5P41_RR00785 13_SUMEX_Annual_Report Year_11_Jun84 5P41 RR00785 13 SUMEX Annual Report Year 11 Jun84
5P41_RR00785-13_SUMEX_Annual_Report-Year_11_Jun84 5P41_RR00785-13_SUMEX_Annual_Report-Year_11_Jun84
User Manual: 5P41_RR00785-13_SUMEX_Annual_Report-Year_11_Jun84
Open the PDF directly: View PDF .
Page Count: 233
Download | |
Open PDF In Browser | View PDF |
SUMEX SUMEX STANFORD UNIVERSITY UNIVERSITY STANFORD MEDICAL EXPERIMENTAL EXPERIMENTAL COMPUTER COMPUTER RESOURCE RESOURCE MEDICAL RR - 00785 00785 ANNUAL REPORT ANNUAL REPORT - YEAR 11 Submitted Submitted to BIOTECHNOLOGY BIOTECHNOLOGY RESOURCES RESOURCES PROGRAM PROGRAM NATIONAL NATIONAL INSTITUTES INSTITUTES OF OF HEALTH HEALTH June June 1,1984 1,1984 STANFORD STANFORD UNIVERSITY UNIVERSITY SCHOOL SCHOOL OF MEDICINE MEDICINE Edward Edward A. Feigenbaum, Feigenbaum, Principal Investigator Investigator Edward Edward H. Shortliffe, Co-Principal Co-Principal Investigator Investigator DEPARTMENT OF DEPARTMENT OF HEALTH HEALTH AND AND HUMAN HUMAN SERVICES SERVICES PUBLIC HEALTH SERVICE PUBLIC HEALTH SERVICE NATIONAL INSTITUTES OF NATIONAL INSTITUTES OF HEALTH HEALTH OF DIVISION DMSION OF RESEARCH RESEARCH RESOURCES RESOURCES BIOTECHNOLOGY RESOURCES BIOTECHNOLOGY RESOURCES PROGRAM PROGRAM ANNUAL ANNUAL PROGRESS PROGRESS REPORT REPORT PART PART I., I., TITLE TITLE PAGE PAGE 1. 1. PHS GRANT GRANT NUMBER: NUMBER: 6P41RR00786-11 SP41RR00785-11 2. TITLE TITLE OF GRANT: GRANT: SUMEX SUMEX Stanford University Medical Stanford University Medical Experimental Computer Resource Experimental Computer Resource 3. 3. NAME NAME OF OF RECIPIENT RECIPIENT INSTITUTION: INSTITUTION: 4. 4. HEALTH HEALTH PROFESSIONAL PROFESSIONAL SCHOOL: SCHOOL: Stanford University Stanford University School of Medicine Medicine School of S. 5. REPORTING REPORTING PERIOD: PERIOD: Sa. 5a. Sb. 5b. FROM: FROM: TO: TO: 08-01-83 08-01-83 07-31-84 07-31-84 6. PRINCIPAL PRINCIPAL INVESTIGATOR: INVESTIGATOR: 6a. 6b. NAME: NAME: TITLE: TITLE: 6c. SIGNATURE: SIG NATURE: 7. DATE DATE SIGNED: SIGNED: 8. TELEPHONE: TELEPHONE: Edward A. A. Feigenbaum, Feigenbaum, Ph.D. Ph.D. Edward Professor of of Computer Computer Science Science Professor ?%.1 AIM Projects Projects II.A.4.1. Project 11.,4.4.1. PATHFINDER PATHFINDER Project II.A.4.2. II.A.4.2. RXDX RXDX Project Project II.B. 1I.B. Books, Papers, and Abstracts Abstracts H.C. J1.C. Resollrce Resource Summary Summary Table Table Appendix Committee Membership Appendix A. A. AIM AIM Management Management Committee Membership Appendix Growth Appendix B. B. Community Community Growth and and Project Project Abstracts Abstracts E. A. Feigenbaum ii 106 106 121 129 130 130 136 136 142 149 149 159 159 166 166 167 175 179 179 183 183 185 190 191 191 197 202 202 202 202 203 203 209 209 Appendix Appendix 5P·n SF’41 RR0078S-11 RR0078S-11 List List of of Figures Figures Figure Figure Figure Figure Figure Figure Figure Figure Figure Figure Figure Figure Figure Figure Figure Figure Figure Figure Figure Figure Figure Figure Figure Figure Figure Figure Figure Figure Figure Figure 1: 2: 3: 3: 4: 4: 5: 5: 6: 6: 7: 8: 9: 10: 10: 11: 11: 12: 12: 13: 13: 14: 14: 15: 15: Current Current SUMEX-AIM SUMEX-AIM Decsystem Decsystem 2060 Computer Computer Configuration Configuration Current Current SUMEX-AIM SUMEX-AIM 2020 Computer Computer Configuration Configuration Current AX Computer Current Shared V VAX Computer Configuration Configuration SUMEX-AIM SUMEX-AIM Ethernet Ethernet Configuration Configuration SUIV1EX-AIM SUMEX-AIM File File Server Server {SAFE} {SAFE} SUt\.tEX-AIM Development A....X} SUMEX-AIM Development Vax Vax {ARDV {ARDVAX} Total by Month Total CPU CPU Time Time Consumed by Month Peak Number Number of Jobs by by Month Month Peak Load Average by Month Average by Month Monthly by Community Monthly CPU CPU Usage Usage by Community Monthly Terminal Connect by Community Monthly Terminal Connect Time Time by Community Cumulative by Project Cumulative CPU CPU Usage Usage Histogram Histogram by Project and Community Community TYMNET TYMNET Terminal Terminal Connect Connect Time Time ARP ANET Terminal ARPANET Terminal Connect Connect Time Time SUMEX-AIM SUMEX-AIM Growth Growth by Community Community ... in Ill 10 10 11 12 13 13 14 15 31 32 32 34 35 37 45 45 209 E. A. Feigenbaum 5P41 RR00785-U RR00785-11 Annual Annual Report Report I. Description I. Narrative Narrative Description This This third third year year of of the 5-year 5-year renewal renewal of of the the SUMEX SUMEX resource resource grant grant has been an active year year not not only only for for the the SUMEX SUMEX staff, staff, but but for for the SUMEX-AIM SUMEX-AIM community community involved involved active in developing Successes developing expert expert systems. systems. Successes in developing developing such systems, systems, many many of of them them stemming stemming from from projects projects in the the SUMEX-AIM SUMEX-AIM community, community, continue continue to to stimulate stimulate strong strong and growing growing interest interest in AI AI research research on many many educational, educational, governmental, governmental, and industrial industrial fronts. fronts. This is an annual annual report report for for the Stanford Stanford University University Medical Medical EXperimental Experimental This computer computer resource resource for for applications applications of of Artificial Artificial Intelligence Intelligence in Medicine Medicine (SU1vfEX-AIM) (SUMEX-AIM). It It covers the period period between between May May 1, 1983 and April April 30, 1984. This This third third year year of of the 5-year 5-year renewal renewal of of the the SUMEX SUMEX resource grant grant has been an active active year year not not only only for for the the SUMEX SUMEX start, staff, but but for for the SUMEX-AIM SUMEX-AIM community community involved involved Successes in developing developing expert expert systems. systems. Successes in developing developing such systems, many many of of them them stemming stemming from from projects projects in the the SUMEX-AIM SUMEX-AIM community, community, continue continue to to stimulate stimulate strong strong and growing growing interest interest in AI AI research research on many many educational, educational, governmental, governmental, and industrial industrial fronts. fronts. In addition, addition, this this past past year year has seen concurrent concurrent development development of of SUMEX-AIM SUMEX-AIM as a distributed distributed scientific scientific resource. resource. Our Our approved approved project project goals focus principally principally on the merging of of state-of-the-art state-of-the-art community community research research in biomedical biomedical AI AI applications applications with with new merging computing computing tools tools and on the challenges they they will will bring bring to to the the SUMEX-AIM SUMEX-AIM community community and resource. The The SUMEX SUMEX start staff continues continues to to exploit exploit these advances in professional professional resource. workstations and communication communication technology, technology, while while at at the the same time time maintaining maintaining our our high high worksta.tions standards for for a computing computing resource. resource. sta.ndards This This third third year year also saw the initiation initiation of of a number number of of SUMEX-AIM SUMEX-AIM pilot pilot projects. projects. pilot projects projects provide provide new activities activities and research research directions directions for for the community community to to These pilot replace replace existing existing projects projects which which have matured matured and moved moved off off the the SUMEX-AIM SUMEX-AIM resource. resource. The The earlier earlier phases of of the SUMEX-AIM SUMEX-AIM resource were characterized characterized by by the building building of a3 national national community community of of biomedical biomedical AI AI collaborators collaborators around around a central central resource located located of St,nnford University. University. Beginning with with 5 projects projects in 1973, the AIM AIM community community grew grew to at Stanford Beginning 11 11 major major projects projects at our our renewal renewal in 1978. This This past past year year saw the the completion completion of of two t.wo long term DENDRAL and PUF'F jVM. There term and successful projects projects on SUMEX-AIM; SUMEX-AIM; DENDRAL PUFF/VM. There currently currently are 13 fully-authorized fully-authorized projects projects plus seven pilot pilot efforts. efforts. t-.lany Many of of the computer computer programs programs under under development development by by these groups groups are maturing maturing into useful intro tools tools increasingly increaingiy useful to the respective respective research or or clinical clinical communities. communities. \Ve We continue to to seek out out new AI AI applications applications in our our community community of of biomedical biomedical and computer comput.er continue scientists The community scientists who who interact interact through through electronic electronic media. media. The SUMEX-AIM SUMEX-AIM community is beginning to evolve as a highly highly distributed distributed resource, resource, with with the the SUMEX SUMEX staff staff and computer computer beginning facilit.y facility serving serving a.<; as the t,he backbone backbone to electronic electronic communication communication and systems systems support. support. The The community is becoming becoming more and more more involved involved in personal personal computers computers and professional professional commnnity workststiors, more heavily heavily dependent dependent on network network communication communication facilities facilities for for workstations, and more in teractions, collaborations, interactions, collaborations, and sharing. sharing. The following following sections sections cover cover the activities activities of of the SUMEX-AIM SUMEX-AIM resource resource this this past The year, including including brief brief summaries summaries of of the the our our objectives, objectives, a characterization characterization of of biomedical biomedical AI AI year, rese:1fch, resesrch, resource organization organization and operating operating procedures, procedures, recent recent core progress progress in system system development development and basic AI AI research, and progress progress in the collaborative collaborative projects. projects. 1 Feigenbaum E. A. Feigenbaum 5P41 5P41 RR00785-11 RR00785-11 Summary Summary of of Research Research Progress Progress I.A. of Progress I.A. Summary Summary of Research Research Progress LA.1. of and I.A.1. Overview Overview of Objectives Objectives and Rationale Rationale SUMEX-AIM SUMEX-AIM ("SUMEX") (“SUMEX”) isis aa national national computer computer resource resource with with aa dual mission: mission: 1) 1) promoting promoting applications applications of of computer computer science science research research in in artificial artificial intelligence intelligence (AI) (AI) to biological biological and and medical medical problems, problems, and and 2) 2) demonstrating demonstrating computer computer resource sharing sharing within within aa national community of health research projects. The central SUMEX-AIM facility The central SUMEX-AIM facility is is nationa. community health projects. located physically in the Stanford University Medical located physically Stanford University Medical School and serves as as aa nucleus for aa community community of medical medical AI AI projects projects at universities universities around around the country. country. SUMEX SUMEX provides provides computing computing facilities facilities tuned tuned to the needs needs of AI AI research and communication communication tools to facilitate facilitate remote remote access, access, interinter- and intra-group intra-group contacts, contacts, and the demonstration demonstration of developing developing computer computer programs programs to biomedical biomedical research collaborators. collaborators. I.A.1.1. is Intelligence LA. 1.1. What What is Artificial Artificial Intelligence The The subfield subfield of computer computer science known known as as Artificial Artificial Intelligence, Intelligence, or or AI, AI, deals with with symbolic reasoning using large amounts of heuristic knowledge. Many of world’s symbolic re-aoning using amounts heuristic knowledge. Many of the world's difficult problems electronic electronic or or mechanical mechanical difficult problems are symbolic, symbolic, such as troubleshooting troubleshooting equipment, medical diagnosis and therapy planning, and configuring elemental equipment, medical diagnosis therapy planning, configuring elemental parts parts into into a whole system. system. For For these kinds kinds of of problems, problems, AI AI offers offers new opportunities opportunities for for developing developing computerbased solutions. computer-based solutions. AI also uses heuristic heuristic In addition addition to t.o using using symbolic symbolic representations representations of of knowledge, knowledge, AI methods thumb, judgmental judgmental rules mles that that met,hods for for processing processing information. information. Heuristics Heuristics are rules rules of of thumb, aid in finding plausible other areas of of computing computing in its its bk solutions. solutions. AI AI is distinguished distinguished from from other finding plauso’ att,ention heuristic (non-algorithmic) (non-algorithmic) att.ent,ion to to both both symbolic symbolic (non-numeric) (non-numeric) information information and heuristic methods methods for for solving solving problems. problems. Placing AI AI in in Computer Science Placing Computer Science The major major focus of of AI AI is understanding understanding intelligence intelligence through through construction construction (or (or The programming) of of machines machines that that behave intelligently. intelligently. That That is a grand grand goal. In In the the shortshortprogramming) term, AI AI research research focuses on non-numerical non-numerical problem problem solving solving in in order order to to build build experience experience term, with problem problem solving solving methods, methods, techniques techniques for for representing representing various various kinds kinds of of knowledge, knowledge, with interfaces with with users. users, and numerous numerous other other issues. issues. interfaces of t,he the distinguishing distinguishing features features of of problems problems for for which which AI AI methods methods have have been One of t,hat the problems problems are not not well-structured. well-structured. That That is, is. one does not not already already developed is t*hat developed know in advance advance (from (from the the problem problem description description alone) alone) what what the the best best method method is for for solving solving know the problem. problem, In In short, short, there there are no algorithms. algorithms. Broadly Broadly speaking, speaking, AI AI substitutes substitutes the exploratory search search for for precise, precise, algorithmic algorithmic solution solution methods. methods. explorat.ory Expert Systems Systems and and Applications Applications Expert The national national SUMEX-AIM SUMEX-AIM resource resource is an outgrowth outgrowth of of a long, long, interdisciplinary interdisciplinary line line The of artificial artificial intelligence int.elligence research research at at Stanford Stanford and and elsewhere elsewhere concerned concerned with with the the of development of concepts and techniques for building "expert systems" [1]. An "expert development of concepts and techniques for building “expert systems” [l]. An “expert system"” is an inteliigent intelligent computer computer program program that that uses knowledge knowledge and and inference inference procedures procedures sy,item to solve solve problems problems that that, are difficult difficult enough enough to to require require significant significant human human expertise expertise for for their their to solution. For For some fields fields of of work, work, the the knowledge knowledge necessary necessary to to perform perform at at such such aa level, level, solution. E. A. A. Feigenbaum Feigenbaum E. 2 Overview Overview of Objectives Objectives and Rationale Rationale 5P41 SP41 RR00785-11 RR00785-11 plus the inference inference procedures procedures used, used, can be thought thought of as as aa model of the expertise expertise of the expert expert practitioners practitioners of that that field. Two features Two important important features that that distinguish distinguish expert expert systems from from conventional conventional programs Expert flexible in the sense Expert systems are flexible sense programs are flexibility flexibility and understandability. understandability. that in the sense that they they can be be changed and extended extended easily, and they they are understandable understandable sense that that they they can explain explain the contents contents of of their their own own knowledge knowledge bases bases and their their own own lines of reasoning features are especially especially important important in medicine, medicine, where where knowledge knowledge is is reasoning [10]. [lo]. These features changing changing rapidly rapidly and where practitioners practitioners have to understand understand the reasons for for a program's program’s decisions because because they they have to accept responsibility responsibility for following following (or (or not not. following) following) those decisions. The The application application areas range from from medicine medicine to electronics, electronics, from from machinery machinery to software. software. The problems problems ra.nge range from from diagnosis and troubleshooting troubleshooting (analysis) (analysis) problems problems to planning pla,nning and configuration configura.tion (synthesis) (synthesis) problems. problems. Knowledge Knowledge bases bases for for expert expert systems are built built iteratively iteratively -- usually usually through through long interactions interactions over over many many months months between between a human human specialist. specialist who who underst.ands understands the details details of of the domain domain and a knowledge knowledge engineer engineer who who understands understands the programming programming details details of of the system. system. The The knowl-edge knowledge of of an expert expert system system consists of of facts and heuristics. heuristics. The The "facts" “facts” constitute constit,ute a body body of of irrformation &formation that that is widely widely shared, publicly publicly available, available, and generally generally agreed upon by experts experts in a field. field. The The "heuristics" “heuristics” are the mostly-private, mostly-private, little-discussed little-discussed judgment (rules rult's rules of good judgment (rules of of plausible plausible reasoning, reasoning, rules rules of of good guessing) that that charact.erize charact,erize expert-level expert,-level decision making making in the field. field. The The performance performance level of of an expert expert system system is primarily primarily a function funct.ion of of the size and quality quality of of the knowledge knowledge base that that it it possesses. is the possesses. One of of the key key ideas in maintaining maintaining flexibility flexibility and understandability understandability clean separation program that separation of of elements elements of of the knowledge knowledge base from from elements elements of of the program that interpret base. interpret the knowledge knowledge base. The building expert The major major issues issues in building expert systems, systems, at the moment, moment, are: •l selecting problem (in importance, selecting an appropriate appropriate problem (in terms terms of of size, difficulty, difficulty, importance, decomposability, decomposability, risk) risk) •l selecting selecting a representation representation and control control structure structure (or (or framework framework system system that that sllpplies both), supplies both), •l set.t.ling problem, set,t.ling on an a.11appropriate appropriate vocabulary vocabulary and conceptualization conceptualization for for the the problem, •l finding finding an available available expert, expert, transferring the expert’ expert's knowledge into into the the program program (knowledge (knowledge engineering), engineering), l• transferring s knowledge refining t,he the knowledge knowledge base with with feedback feedback from from test test cases, cases, l• refining packaging t,he the system system in a form form that that is acceptable acceptable to to end-users, l• packaging l• validating validating the quality quality of of the the program’ program's s advice. advice. of the best best known known expert expert systems systems is MYCIN MYCIN [3], a program program in which which the the One of separation of of knowledge knowledge (of (of medicine) medicine) from from the rest rest of of the the program program was carefully carefully sepa.rstion engineered. (The (The abstracted abstracted ca.se case of of an arbitrary arbitrary knowledge knowledge base and a framework framework engineered. interpreter, plus plus auxiliary a.uxiliary programs, programs, was achieved achieved in the EMYCIN EMYCIN system system [16], to which which interpreter, [lS], to other domains domains can be added to to build build a diagnostic diagnostic system system in those domains.) domains.) knowledge of of other knowledge Cllrrent.ly authorized authorized projects projects in in the the SUMEX SUMEX community community are concerned concerned in in some Currently way with with the the application application of of AI AI to to biomedical biomedical research*. research·. The The tangible tangible objective objective of of this this way approach is the t.he development development of of computer computer programs programs that that will will be more more general general and effective effective approach • Bri€'f abstracts abstracts of of the the various various projects projects can can be be found found in in Appendix Appendix BB on on page page 209 200 and and more more detailed detailed progress progress Brief summaries in in Section Section IIII on on page page IHI. 60. summaries 3 A. Feigenbaum Feigenbaum E. A. RR00785-111 5P41 RR007851 Overview of of Objectives Objectives and and Rationale Rationale Overview consultative tools tools for for the the clinician clinician and and medical medical scientist. scientist. There There already already have have been consultative promising result*s results in in areas such as chemical chemical structure structure elucidation elucidation and and synthesis, synthesis, diagnostic diagnostic promising consultation, molecular molecular biology, biology, and and modeling modeling of of psychological psychological processes. consult,ation, much is yet yet to to be learned learned in the t.he process of of fashioning fashioning a coherent coherent Needless t.o say, much scientific discipline discipline out out of of the the assemblage of of personal personal intuitions, intuitions, mathematical mathematical procedures, procedures, scientific emerging theoretical theoretical structure st.ructure comprising comprising artificial artificial intelligence intelligence research. research. State-of-theState-of-theand emerging art programs programs are far far more more narrowly-specialized narrowly-specialized and and inflexible inflexible than than the the corresponding corresponding art of human human intelligence intelligence they they emulate; emulate; however, however, in in special special domains domains they they may may be of of aspects of comparable or or greater greater power, power, e.g., in in the the solution solution of of formal formal problems problems in in organic organic chemistry. chemistry. comparable I.A.1.2. Impact Impact of of AI AI in in Biomedicine Biomedicine I.A.1.2. There is a certain certain inevitability inevitability to to the the field field of of Artificial Artificial Intelligence Intelligence and and its its There applications, in particular, particular, to to medicine medicine and biosciences. The The cost cost of of computers computers will will applications, continue to to fall fall drastically drastically during during the the coming coming two two decades. As itit does, many many more more of of t,he the continue practitioners of ot the t.he world’ world's professions will will be persuaded persuaded to to turn turn to to economical economical automatic automatic practit,ioners s professions information processing processing for for assistance in managing managing the the increasing increasing complexity complexity of of their their daily daily information They will will find, find, from from most most of of computer computer science, help only only for for those problems problems that that tasks. They have a mathematical mat.hematical or or statistical statistical core, or or are of of a routine routine data-processing data-processing nature. nature. But But ha.ve such problems problems will will be relatively relatively rare, rare, except except in in engineering engineering and and physical physical science. In In management, indeed indeed in most most of of the the world’ world's work, the the daily daily tasks tasks are medicine, biology, biology, management, medicine, s work, requiring symbolic symbolic reasoning reasoning with with detailed detailed professional professional knowledge. knowledge. The The computers computers those requiring that will wiII act act as intelligent intelligent alrsivtants assistants for for these professionals profeSSionals must must be endowed endowed with with that symbolic symbolic reasoning reasoning capabilities capabilities and knowledge. knowledge. The The growth growth in medical medical knowledge knowledge has far far surpassed the the ability ability of of a single practitioner to master practitioner master it it all, all, and the computer's computer’s superior superior information information processing processing capacity capacity there by offers thereby offers a natural natural appeal. Furthermore, Furthermore, the reasoning reasoning processes processes of of medical medical experts experts are poorly poorly understood; understood; attempts attempts to to model model expert expert decision-making decision-making necessarily necessarily require require a degree of of introspection introspection and a structured structured experimentation experimentation that that may, may, in turn, turn, improve improve the the quality quaiity of of the the physician's physician’s own own clinical clinical decisions, making making them them more more reproducible reproducible and defensible. New insights defensible. New insights that that result result may may also allow allow us more more adequately adequately to to teach m~dical medical students students and house staff staff the techniques techniques for for reaching reaching good decisions, rather rst,her than than merely merely to to offpr offer ;).n. collection collect ion of of facts facts which which they they must must independently independently learn learn to to utilize utilize coherently. coherently. The The kno'wledge knowledge that that must must be used is a combination combination of of factual factual knowledge knowledge and heuristic The latter 1atLer is especially especially hard hard to to obtain obtain and represent represent since the heuristic knowledge. knowledge. The experts expertIs providing providing it it are mostly mostly unaware unaware of of the heuristic heuristic knowledge knowledge they they are using. ~ledical hledical and scientific scientific communities communities currently currently face many many widely-recognized widely-recognized problems problems relating relating to the rapid rapid accumulation accumulation of of knowledge, knowledge, for example: example: •l codifying codifying theoretical theoretical and heuristic heuristic knowledge knowledge •l effectively effectively using the wealth wealth of of information information implicitly implicitly available available from from textbooks, textbooks, journal articles and other practitioners journal articles other practitioners •l disseminating disseminating that that knowledge knowledge beyond beyond the intellectual intellectual centers where where it it is is collected collected •l customizing customizing the presentation presentation of that that knowledge knowledge to individual individual practitioners practitioners as as well well as ‘3s customizing customizing the t,he application application of the information information to individual individual cases cases \Ve We believe that t.hat computers computers are an inevitable inevitable technology technology for for helping helping to overcome overcome E. A. Feigenbaum Feigenbaum 4 Overview of of Objectives Objectives and Rationale Rationale Overview 5P41 RR00785-11 RR00785-11 While recogOlzmg recognizing the the value value of 01 mathematical mathematical modeling, modeling, statistical statistical these problems. While problems. classification, decision decision theory theory and other other techniques, techniques, we believe believe that that effective effective use use of of such classification, methods depends on using using them them in conjunction conjunction with with less less formal formal knowledge, knowledge, including including methods contextual contextual and strategic strategic knowledge. knowledge. Artificial inklligence intelligence offers offers advantages advantages for for representing representing and using using information information that that Artificial will allow allow physicians physicians and scientists scientists to to use use computers computers as as intelligent intelligent assistants. assistants. In In this this way way will envision a significant significant extension extension to to the the decision-making decision-making powers powers of of specific specific practitioners practit.ioners we envision without without reducing reducing the importance importance of of those individuals individuals in that that process. Knowledge is power, power, in the profession profession and in the intelligent intelligent agent. agent. As we proceed Knowledge to model model expertise expertise in medicine medicine and its its related related sciences, sciences, we find find that that the power power of of our our to programs derives derives mainly mainly from from the the knowledge knowledge that that we are able to to obtain obtain from from our our programs collaborating practitioners, practitioners, not not from from the sophistication sophist,ication of of the inference inference processes processes we collaborating observe them them using. Crucially, Crucially, the knowledge knowledge that that gives power power is not not merely merely the the knowledge of of the textbook, textbook, the lecture lecture and the the journal, but the knowledge knowledge of of good knowledge journal, but good practice--the experiential knowledge knowledge of of good knowledge practice--tlle experiential good judgment judgment and good good guessing, gue88ing, the knowledge of the practitioner's practitioner’s art art that that is orten often used in lieu lieu of of facts facts and rigor. rigor. of This This heuristic heuristic knowledge knowledge is mostly mostly private, private, even in the the very very public public practice practice of of science. It It is almost almost never never taught taught explicitly, explicitly, is almost almost never never discussed and critiqued critiqued among among peers, and most most orten often is not even in the t.he moment-by-moment moment-by-moment awareness of 01 the the practitioner. practitioner. not Perhaps the the most most expansive expansive view view of of the significance significance of of the work work of of the the SUMEXSUMEXPerhaps AIhl community communit.y is that that a methodology methodology is emerging emerging for for the systematic systematic explication, explication, testing, testing, AIM dissemination, and teaching teaching of of the heuristic heuristic knowledge knowledge of of medical medical practice practice and scientific scientific dissemination, performance. performa,nce. It It may may be less less important important that that computer computer programs programs can be organized organized to to use use this lcnowledge than Lhan that that the knowledge knowledge itself itself can be organized organized for for the use of of the the human human this practitioners practitioners of of today today and tomorrow. tomorrow. Evidence of of the impact impact of of SUMEX-AIM SUMEX-AIM in promoting promoting ideas such as as these, and Evidence developing the pertinent pertinent specific specific techniques, techniques, has been the explosion explosion of of interest interest in medical medical developing artificial intelligence intelligence and the specific specific research efforts efforts of of the SUMEX SUMEX community. community. artificial As SU?rlEX has ha.s entered entered its second decade, we have found found that that the small small community community of of SU!\fEX researchers 970'ss has now researchers that t.hat characterized characterized the AIM AIM field field in the the early early 11970’ now grown grown to a large, accomplished, and respected respected research community. community. The The American American Association Association for for Artificial Artificia.1 accomplished, Intelligence Intelligence (AAAI), (A,4.41), the principal principal scientific scientific membership membership organization organization for for the AI AI field, field, has members, over over 1000 of of whom whom are members members of of the medical medical special interest interest group group 4000 members, known as as t.he the AAAI-M. AAAI-M. This subgroup subgroup was founded founded by by members members of of the SUMEX-AIM SUMEX-AIM known This community who who were active active in AAAI AAAI and is the only only active active subgroup subgroup in the Association. Association. community The organization organization distributes distributes semiannual semiannual newsletters newsletters on medical medical AI AI and provides provides a focus The for co-sponsoring levan t medical co-sponsoring re relevant medical computing computing meetings meetings with with other other societies societies (such as as the American Association Association for for Medical Medical Systems Systems and Informatics Informatics -- J\.-\MSI). k4MSI). Medical AI AI papers American Medical prominently featured featured at at’ both both medical medical computing computing and artificial artificial intelligence intelligence meetings, meetings, are prominently artificial int.elligence intelligence is now now routinely routinely featured featured as as a specific specific subtopic subtopic for for specialized specialized and artificial sessions at a,t medical mcdicsl computing computing and other other medical medical professional professional meetings. meetings. For sessions For example, example, members of of the AIM .41bf community communit,y have represented represented the field field to to physicians physicians at the American American members of Pathology Pathology and American American College College of of Physicians Physicians meetings meetings for for the last last several several College of A mere decade ago, the words words "artificial “artificial intelligence" intei!igence” were never never uttered uttered at such years. A The growing growing interest interest and recognition recognition are largely largely due to to the activities activities of of the the conferences. The SU?.lES-.4IM community. community. SU!'.fEX-AIM Another indication indication of of the growing growing impact impact of of the SUMEX-AIM SUMEX-AIM community community is its its Another effect, on medical medical education. education. effect For For reasons such as as those outlined outlined above, there there is an increasing recognition recognition of of the t,he need for for a revolution revolution in the way way medicine medicine is taught taught and increasing 5 E. A. A. Feigenbaum Feigenbaum Overview Overview of Objectives Objectives and Rationale Rationale 5P41 RR00785-11 RR00785- I I medical Computing medical students students organize organize and access access information. information. Computing technology technology isis routinely routinely cited as as part part of this this revolution, revolution, and artificial artificial intelligence intelligence (and SUMEX-AIM SUMEX-AIM research) research) generally generally figures figures prominently prominently in such discussions. Such diverse organizations organizations as as the Na.tional Na.tional Library Library of Medicine, Medicine, the American American College of Physicians, Physicians, the Association Association of American American Medical Medical Colleges, and the Medical Medical Library Library Association Association have all called for for sweeping changes in medical medical education, education, increased educational educational use use of computing computing technology, technology, enhanced research in medical medical computer computer science, and career development development for people working working at the interface interface between between medicine medicine and computing; computing; reports reports of all four four organizations orga,nizations have ha.ve specifically specifically cited cited the role of of artificial artificial intelligence intelligence techniques techniques in future future medical programs medical practice practice and have used SUMEX-AIM SUMEX-AIM programs as as examples of where where the technology technology is is gradually gradually heading. heading. In summary, summary, the logic which which mandates mandates that that artificial artificial intelligence intelligence playa play a key key role in enhancing enhancing knowledge knowledge management management and access access for biomedicine biomedicine -- a logic in which which we have long believed gradually become evident evident to much much of the biomedical biomedical community. community. We believed -- has gradually are encouraged encouraged by by this this increased recognition, recognition, but but realistic realistic about about the significant significant research cha.llenges cha.llenges that that remain. remain. Our Our goals are accordingly accordingly both both scientific scientific and educational. educational. We We continue continue t.o to pursue pursue the research objectives objectives that that have always always guided guided SUMEX-AIM, SUMEX-AIM, but but must must also undertake undertake educational educational efforts efforts designed to inform inform the biomedical biomedical community community of of our our results results while while cautioning cautioning it it about about the challenges remaining. remaining. A. Feigenbaum Feigenbaum E. A. 6 Details of of Technical Technical Progress Progress Details 5p·H 5P4 1 HH00785-11 R.R00785- 11 I.A.2. Details of Progress I.A.2. Details of Technical Technical Progress I.A.2.1. Facility Management and I.A.2.1. Facility Management and Operation Operation The The following following materia.l materia.1 covers the SUMEX-AIM SUMEX-AIM resource activities activities over over the past past year year in greater greater deta.il. detail. Individual Individual sections sections cover cover progress progress in ; •l Facility Facility Management Management and Operation Operation Timesharing Systems Systems •l Timesharing •l Professional Professional \Vorkstations Workstations •l Networking Networking and Communications Communications These sections sect.ions outline outline accomplishments accomplishments in the context context of of the resource staff staff and resource management. management. Details Details of of the progress and plans plans for for our our external external collaborative collaborative projects projects are presented presented in Section Section II II beginning beginning on page 69. I.A.2.2. Facility Management and I.A.2.2. Facility Management and Operation Operation SUMEX-AIM continues SUMEX-AIM continues to to manage and and operate operate it's it’s computing computing resources in a effect.ive effect,ive and efficient efficient manner manner conducive conducive to to providing providing a reliable reliable and robust robust computing computing environment. environment. While While the the previous previous year year (Year (Year 10) involved involved a major major move from from the KIlO KIlO Tenex Tenex syst.em system to a new DECsystem DECsystem 2060, this this year year saw more emphasis emphasis to to our our gradual gradual move to to distributed distributed processing, processing, while while continuing continuing to to improve improve our our excellent excellent timesharing timesharing environment environment on the 2060. This This development development is covered covered in full full in section section I.A.2.2 I.A.2.2 starting starting on page 15. Ollr Our continued continued movement movement to to professional professional workstations workstations has taken taken on several several forms. forms. We We have continued continued to to acquire acquire Lisp Lisp machines machines for for use use by by the SUMEX SUMEX community community while while at the same time time investigating investigating t.he the use of 01 remote remote virtual virtual graphics graphics and new lower lower cost workstat,ions workstations such as as the Apple Apple Macintosh, Macintosh, Sun workstations, workstations, and others others that that are appearing The The development development of of professional professional workstations workstations is covered covered in apprnring on the market. market. more more detail detail in section section I.A.2,3 I.A.2.3 starting starting on page 21. SUMEX SUMEX cont,inues continues to to expend expend a great great deal of of effort effort in the the support support and development development of Key communications facilities. facilities. Key to to our our ability ability to provide provide the the of our our networking networking and communications maximum maximum computing comput,ing power power available available to to the t,he greatest greatest number number of of users is a mechanism mechanism for for making making it it, irrelevant irrelevant where where that that user is physically physically located. located. By By having having a robust robust networking networking and communications communications environment, environment, we are able to extend extend our our facility facility to to any any user or or group group of of users, thereby thereby making making available available to to them them the power power and convenience convenience of of SUMEX. SUMEX. Further Further information information on the the progress made in networking networking and communications communications can be found found in section section I.A.2.4 I.A.2.4 starting starting on page 23. In In the area of of fa.cility fa.ci!ity management management and operation, operation, several severa.! noteworthy noteworthy events events occurred occurred over over the past past year year which which will will be explained explained in more detail detail here. 7 E. A. Feigenbaum Feigenbaum Details Details of Technical Technical Progress 5P41 RR00785-11 RR00785-1 I SUA/EX/IIPP Facility SUhdEX/HPP Welch Welch Road Road Computing Computing Facility A major major development development this this past year year at SUMEX SUMEX was the move of the Heuristic Heuristic Programming Project to their new location at 701 Programming Project their location 701 Welch Welch Road, adjacent adjacent to the Stanford Stanford Medical Medical Center. Center. Since this this group group is a major major user of the SUMEX-AIM SUMEX-AIM resource and the focus for most most of the core AI ‘41 research, a good deal of effort effort was expended to provide provide a robust robust computing computing environment environment at their their new location. location. This This development development involved involved several stages and levels of technical technical deveL:;pment, devei;;pment, ranging ranging from from construction construction of the machine machine room, room, new cable and wiring wiring installation, installation, procurement procurement and setup of networking networking hardware, hardware, to major major new developments developments in the networking networking software software and a twisted twisted pair pair Ethernet Ethernet communication communicat.ion link link between between this this site and the main main SUMEX SUMEX Computer Computer room. room. All A!! of the hardware hardware and facilities facilities purchases were funded funded from from sources other other than than SUMEX. SUMEX. \Ve We setup the general genera! communications communications capabilities capabilities for for the two two buildings buildings occupied occupied by by the HPP. HPP. This This involved involved wiring wiring up local terminals, terminals, installing installing local Ethernets Ethernets (both (both 3 and 10 10 mt'gabit megabit capability), capability), and acquiring acquiring and installing installing networking networking hardware hardware such as as terminal terminal interface interface processors (TIP) (TIP) and gateways, gateways, as as well we!! as as extending extending the current current SUMEX SUMEX TIP TIP and GATE\\, A Y software to handle both 3 and 10 megabit network traffic. GATEWAY software handle both megabit network traffic. But But the most most important important and most most interesting interesting development development in this this process was the twisted pair' “twist,ed pair’ ethernet ethernet developed developed by by the SUMEX SUMEX engineering engineering staff stall to allow allow high high speed reliable reliable communications communications between between this this Welch Welch Road facility facility and the SUMEX SUMEX machine machine room. room. Further Further information information on this this new ethernet ethernet can be found found in Section Section I.A.2.4 I.A.2.4 on page 23. II HPP HPP researchers are routinely routinely using this this link link to to communicate communicate with with SUMEX SUMEX and the the central central university university network. network. In addition, addition, various various Lisp Lisp machines machines and printers printers located located in the HPP HPP facility facility and connected connected to to a local network network are able to communicate communicate with with the the university university network. network. The The end result result is that that we have successfully successfully been able to to extend extend the the SUMEX SUMEX compnting providing a high computing environment environment to a remote remote site, site, providing high speed link link to to the facilities facilities of of SUMEX processing. We SUMEX While while also allowing allowing for for local local distributed distributed processing. We see see this this experience experience has being most most valuable valuable in the future future as as we move further further into into a distributed distributed environment, environment, while provided by by large while still still needing needing the sharing sharing of of resources and communication communication links links provided t.imesharing timesharing systems systems and local area networks. networks. Dig£tal Equipment Equipment Corporation stops development 0/86-6it product line Digitnl Corporation stops development of S&bit product line Digital Equipment Equipment Corporation, Corporation, a long long time time supplier supplier of of high high speed 36-bit 36-bit Digital timesharing computers computers to to the the Artificial Artificial Intelligence Intelligence community, community, announced announced that that itit was timesharing stopping all all development development of of future future 36bit 36-bit products, products, and instead instead sta.rting starting a program program to to stopping provide a migration migration path path to to its its line line of of VAX V A..,{ minicomputers. minicomputers. provide Many DEC DEC 20 customers customers had had been anticipating anticipating a new yet yet unannounced unannounced machine machine Many from DEC DEC code named named the the'‘Jupiter’ Jupiter'. which had been reported reported to to be a order order of of magnitude magnitude from , which faster than than the t.he current current KLlO KLlO processor processor used in DEC20’ DEC20's DEClO's. However, DEC’ DEC's fa.ster s and DEClO’ s. However, s announcement means this this effort effort has stopped, stopped, and we can expect expect no more more 36bit 36-bit products products announcement from Digits! Digital Equipment Equipment Corporation. Corporation. from The effect effect of of t’ this announcement to to the the AI AI community community is disappoint,ing, disappointing, alt,hough although The his announcement not totally totally unexpected. unexpected. The The DECsystem20 DECsystem20 has been the the predominant predominant timesharing timesharing machine machine not to support support Artificial Artificial Intelligence Intelligence based research, research, but but yet yet researchers researchers have been in in used to of more more processing processing power power and and larger larger address spaces for for quite quite a few years. years. DEC DEC has need of clearly decided decided to to devot,e devote their their resources resources to to VAX VAX development. development. For For those tho~e in in need of of clearly greater 36bit 3u-bit processing processing power power or or address space, you you must must now now look look to to newer newer less less greater E. A. A. Feigenbaum Feigenbaum 88 RR00785-11 5P41 RR0078S-11 Details of of Technical Technical Progress Progress Details experienced experienced companies companies such as as Foonley Foonley and Systems Systems Concepts Concepts for for follow follow on 36-bit 36bit products products which which those firms firms are preparing. preparing. The The impact impact of of this this decision decision on the the SUMEX-AIM SUMEX-AIM community community must must be examined examined in conjunclion conjunclion with with our our development development of of AI AI systems systems on personal personal Lisp Lisp machines. machines. We We have outlined outlined very very clearly clearly our our plans plans to to move move to to distributed distributed Lisp-based Lisp-based workstations workstations for for AI AI t,his is clearly clearly where where we see see the the AI AI computing computing market market heading. heading. These Research, and this ma.chines machines offer offer much much better better cost./performance cost/performance ratios ratios than than timesharing timesharing machines, machines, high high resolution bit-mapped bit-mapped screens, and powerful powerful Lisp Lisp programming programming environments environments for for the resolution development development, and eventual eventual dissemination dissemination of of AI AI based systems. However, However, this this is not not too too say we still still do not not see see a role for for the large timesharing timesharing machine machine in our our environment. environment. We We still still of a large central central mainframe mainframe computer computer as the the anchor anchor for for a large believe in the use of con1munit.y of of users. The The mainframe mainframe also functions functions as as a central central facility facility for for communication communication community collaboration, and provides provides fast fast Lisp Lisp cycles for for program program development development where where the and collaboration, application application is not not in need of of a specialized specialized workstation. workstation. Other SUA/EX SUMEX C’omputing Facilities Other Computing Facilities SUMEX continues continues to to support support other other mainframe mainframe computers, computers, file servers, professional professional SUMEX workstations, workstations, and assorted assorted printers printers and terminals terminals for for use use by by the the SUMEX-AIM SUMEX-AIM community. community. 1. The The SUMEX-AIM SUMEX-AIM File File server, server, based on a VAX VAX 11/750 11/750 computer, computer, continues continues to to of the workstation workstation users within within SUMEX-AIM. SUMEX-AIM. The The use use of of serve the needs of SAFE by by users of of our our 2060 is minimal. minimal. We We plan plan to extend extend the the use use of of SAFE SAFE in SAFE future by by providing pro\-iding more more convenient convenient access access by by 2060 users than than is currently current,ly the future available. available. The VAX VAX 11 11,/780 computer system, system, originally originally purchased purchased with with DARPA DARPA funds funds 2. The /780 computer and previously previous!y located located in Margaret Margaret Jacks Hall Hall on campus, campus, has become a SUMEX-AIM resource this this past past year. year. The The system system was moved moved to a new location location SUMEX-AIM Stanford campus campus which which provides provides a better better environment environment for for a computer computer on the Stanford of this this size. This This VAX VAX is now now shared shared between between the the Computer Computer Science of Department Department and the SUMEX-AIM SUMEX-AIM community. community. 3. SUMEX SUMEX continues continues to support support a wide wide range of of professional professional workstations workstations from from vendors as as Xerox, Xerox, Symbolics, Symbolics, and Hewlett Hewlett Packard Packard for for the development development such vendors testing of of AI AI applications. applications. Additional and testing Additional work work has been started started to explore explore use of of the Apple Apple Macintosh Macintosh and Apple Apple Lisa Lisa within within SUMEX. SUMEX. More More the use information on these developments developments can be found found in section section I.A.2.3. I.A.2.3. information 9 E. A. Feigenbaum Feigenbaum 5P41 RR00785-11 RIl00785-11 Details of of Technical Technical Progress Progress Details 7 G3nmil Central Proceaor Processor DECKLlo-E DEC KLlO-E 2M words warns of of memoly, memory, c¶cho Cache 2M RH20 RH2o L I II RH20 RtKo MassBus " . .sus DisK colltr0ller Controller D&k and Dnvc Onve and DECRP07 DEC UP07 I Disk Controlkr ContrOller Disk and Dnve Drive MCI DECRP07 DEC R PO7 I RH20 I I RH20 IJNIEUS UNIBUS MusSus M&s&m L 4 OIB20 DIE20 DEC AN20 ARPAnet Interlace ~ and OnY8 DECRPOe I MEIS MassBus Ettlemet Interlace ~ ~ ~ H K1.INIKUne ~glnQTTY -- DEC LP·26 LP-26 DEC urm Pnnta UnePnntar t- 6 Une Scannens DEC OH·', 96 Unes total III 6L.ines TYMNET Interface Figure 1: Figure 1: f>. .A. Feigenballm ~ IZ Current SUMEX-AIM SUME?(-AIM Decsystem Decsystem 2050 2060 Computer Computer Configuration Configuration Current 10 7I DEC DEC LPlO LP10 Line Une Printer Printer Tape Drives and Controller I ~ ~ DEC BAlD L Console TTY 20ECTU·78 2 DEC TU-78 J l/O bus 1I0bua Disk ContrOller ~ t fo-' 11/40FROUTEND 111.-0 FRONT END , . RR00785-11 5P41 RR00785-11 Details Details of of Technical Technical Progress Progress Central Central Processor Processor DEC KS·10 DECKS-10 512K 512K words words Memory Memory LA-36 LA-36 Console Console I I I Unibus Unibus Adapter Adapter Unibus Unibus Adapter Adapter I II 1 I UNIBUS UNIBUS I I Massbus Massbus Adapter Adapter I Massbus Adapter Ethemet UNIBUS Interface I Line Scanner DEC OZ·" f-16 Lines MassBus I Disk Disk Controller Controller and and Drive Drive DEC DEC RP·06 RP-06 Figure 2: Figure 2: UNli3US UNIBUS MassBus r-l Tape Tape Controller Controller and and Drive Drive DECTU....s DEC TlJ-45 Current Current SUMEX-AIM SUMEX-AIM 2020 Computer Computer Configuration Configuration 11 11 E. A. Feigenbaum Feigenbaum RR00785- 11 5P41 RR00785-11 Details Details of of Technical Technical Progress Progress Central Central Processor Processor DEC DEC VAX VAX 11nOO 11 f70O 4 Mbytes Mbytes Memory Memory Floating Floating Pt Pt Unit Unit LA·36 LA-36 Console Console MassBus MSSBUS Disk Controller and Drive DEC RP·06 Line Scanner DEC DZ·" 16 Lines Disk Disk Controller Controller and and Drive Drive DEC DEC RP·07 RP-07 Ethernet UNIBUS Interface Tape Controller and Drive DEC TU·n Figure 3: Figure 3: E. .r\. A. P('ig('nbaum Feigenbsum UNIBUS UNIBUS I- Chaosnet Chaosnet UNIBUS UNIBUS Interface Interface , Current AX Comput,er Current Shared Shared V VAX Comput,er Configuration Configuration 12 DeWIs Details of or Technical Technical Progress Progress SP41 RR0078S-11 SF’ 41 RR0078.5-11 Margaret Jacks Jacks Margaret Heuristic Prog Prog Prof PrOI Heunstic Electrical Electrical Engineering Engmeenng Pine Hall Hall Pine CSLI 10 Mb.t ethernet I SCORE 2060 2060 SCORE Ether TIP TIP Ether Dover Pnnter Printer Dover Other CSD CSD Eoulp Equip Other r I I . I Sumex VAX CSLl2060 Oandeltons Dandelions NS NS file file server server G Oncocin OnCOCin Offices Offices piiiJq SUMEX Offices Offices SUMEX Medical Center Medical Center SUMEX Mach SUMEX Mach Rm Rm ~ El Xerox Alto Alto Xerox Dolph.n Dolphm Dandelions Dandelions SUN. Devel SUN· Devel Ether TIP Ether TIP Imprmt-10 Printer Imprint·10 Printer 2060 2060 2020 2020 VAX· VAX. File File Server Server VAX· VAX. Devel Devel Ether Ether TIP TIP PDP·11 POP-1 1 . Devel Oevel Repeater Repeater ~ M Gateway Gateway Ether TIP Dandelions HP 9836's 10 ethernet 10 Mbit Mbit ethernet Whelan Building (Welch Road) Whelan Buflding (Welch Road) Heuristic Heuristic Prog Prog Proj Proi 1.5 1.5 Mbit Mbit phone phone line line 10 10 Mbit Mbit ethernet ethernet /I I '-====l~~ I 11 I Dandelions Dandelfons LM·3600's LM-3600’S Ether Ether TIP TIP Dolphins Dorado Dorado !=laven Printer 3 Mb.t Mbtt ethernet ethernet 3 Figure 4: Figure 4: SUMEX-AIM SUhlES-AIhl Ethernet Ethernet Configuration Configuration 13 13 E. .\. Feigcnhnrim Det3 Dehilsils of of Technical Technical Progress Progress 5P41 5P41 RR00785-11 RR00785-11 Central Central Processor Procesor DEC DEC VAX VAX llnSO I t n!x 22 Mbytes Mbytes Memory Memory MassBus MassBus 1 I I TUSS H UNIBUS UNIBUS Kenne<1y 600/1600 bpI CDC 256 Mbyte Removable· MedIa Disk Drive Fujitsu Eagle 414 Mbyte Disk Drive I1 It System System Industries Industries 9900 8900 Disk Disk Controller Controller f- ~ I I Fujitsu Eagle Eagle Fuji&u 414 Mbyte Mbyte 414 Disk Drove Drive Disk Fujitsu Eagle Eagle Fujitsu 414 Mbyre Mbyte 414 Disk Drive Drive Disk f- f- TIlDe Drove & Emulex Controller Disk Controller and Drive DEC RK07 Ethernet UNIBUS Interface OZ·11 I-- Figure 5:5: Figure E. ;\. Ft'i~l'nbd'lrn Line Scanner BLines SUMEX-AIM File File Server SerVt'r {SAFE} {SAFE} suhim-AI?.1 14 14 Console TTY I Details Details of Technical Technical Progress 5P41 5P41 RROOi85-11 RR00785-11 Central Processor DEC VAX 111750 2 Mbytes Memory "- I Console TTY TU 58 UNIBUS UNIBUS Disk Controller and Drive DEC RK07 Ethernet UNIBUS Interface Disk Controller and Drive DEC RK07 DEC OZ-11 Line Scanner 8 Lines Figure 6: 6: Figure SUMEX-AIM Development Development Vax Vax {ARDVAX} SUMEX-AIM {ARDVAX} 15 A. Feigenbsum Feigenbaum E. A. Progress Progress -- Timesharing Timesharing Systems Systems 5P41 5P41 RR00785-11 RR00785- 11 I.A.2.3. Timesharing Systems I.A.2.3. Timesharing Systems Continued Continued support support and and development development ot of our our timesharing timesharing systems systems this this past past year year has has concentrated concentrated on on several several areas, areas, including including improvement improvement of of user user services services such such as as printing printing spoolers spoolers and and archiving archiving support, support, implementation implementation of of features features from from our our KilO KIlO Tenex Tenex system, system, enhancing interface network interface service, service, correcting correcting encountered encountered system system bugs, and and enhancing network implementing In implementing new features features for for better better user community community support. support. In addition, addition, we have invested system invested further further effort effort in in supporting supporting the VAX/UNIX VAX/UNIX system in in conjunction conjunction with with the SUMEX-AIM SUMEX-AIM file file server server installation. installation. DECsystem 2060/I'OPS-20 System DECsystem 206O/TOPS-20 System Support Support of our our main main timesharing timesharing machine, machine, the DECsystem DECsystem 2060, 2060, has continued continued during during grant grant year year II. 11. •l Hardware development Hardware development I. 1. The The DECsystem DECsystem 2060 system system now in operation operation at SUMEX SUMEX differs differs greatly greatly from from our our previous previous KilO KIlO Tenex Tenex system. system. Whereas Whereas before, the KilO KIlO system system and TENEX TENEX software software required required much much in-house development development and support, support, life is easier with with the 2060. Being Being at Stanford Stanford University, University, where where there there are at least 7 other DECsystem 20's with similar hardware and software other DECsystem 20’s with similar hardware software is a great great advantage. advantage. We We are able to to share our our experiences with with other other sites, and have become an integral integral part part of of the Stanford Stanford DEC DEC community. community. In addition, addition, the DEC2060 DEC2060 hardware hardware has been more more reliable reliable and easier to maintain maintain than than the the KilO KIlO system. system. 2. Additional provide support Additional modems modems were added to to the the 2060 to to provide support for for BELL BELL 212A baud protocols. protocols. This baud 212A 1200 baud This adds an alternative alternative to Vadic Vadic 1200 baud service. use the Bell service. Modems Modems which which use Bell 212A 212A standard standard are more more widely widely available available for for much much less less cost than than Vadic Vadic modems modems.. •a TOPS-20 TOPS-20 Monitor AIonitor Software Software Enhancements Enhancements 1. A A significant significant enhancement enhancement to to our our TOPS-20 TOPS-20 monitor monitor occurred occurred this this year year when we implemented implemented the the software software from from our our Tenex Tenex system system which which allowed allowed when extended support support for for the the ‘'?' feature of of TOPS-20 TOPS-20 when when parsing parsing filenames. filenames. extended ?’ feature This feature feature allows allows a user at at any any time time to to get get a list list of of possible choices This when needing needing to to input input a file file name by by just just typing typing ‘'?'. This returns returns an ?‘. This when actual list list of of file file names, whereas whereas in the the standard standard TOPS-20 TOPS-20 monitor, monitor, just just actual the st.ring string ‘'input filespec' was returned. returned. This This is a very very significant significant and and the input filespec’ useful change to to our our TOPS-20 TOPS-20 system. system. useful We continued continued to to keep up-to-date up-to-date on on the the various various bug bug fixes fixes and and monitor monitor 2. We improvements that that we received received from from DEC DEC and and other other TOPS-20 TOPS-20 sites. improvements These included included several several fixes fixes and and rewrites rewrites to to the the Internet Internet IP/TCP IP /TCP code These which went went under under a major major revision revision this this past past year. year. which \Ve installed installed the the capability capability for for users to to access access their their subdirectories subdirectories as ifif 3. \Ye they were were the the owners owners of of such. such. While While this this may may seem to to be the the logical logical way way they to implement implement subdirectories subdirectories to to begin begin with, with, DEC’ DEC's models of of to s models subdirectories was was a bit bit different. different. Our Our changes have have since been installed installed subdirectories on other other DEC20’ DEC20's at Stanford Stanford and and elsewhere. elsewhere. on s at 4. We 'We inst,slled installed the the capability capability to to vary vary the the allocation allocation of of windfall windfall cycles cycles in in -1. E. ,4. A. Feigenbaum Feigenbaum E. 16 Progress - Timesharing Timesharing Systems Systems Progress 5P41 RR00785-11 RR00785-11 accordance accordance with with user classes. classes. This This will will allow allow us more more flexibility flexibility in assigning jobs enough cycles to assigning jobs to run run comfortably comfortably while while limiting limiting their their usage usage to a strict strict percentage percentage of of the the machine. machine. 5. vVe We installed installed several several new features features in our our TOPS-20 TOPS-20 EXEC EXEC to to facilitate facilitate the use use of of the system system by by the users. Among Among these features features was the ability ability to to edit edit any any previous previous command commai.nd you you had entered, entered, and then then re-execute re-execute that that command. command. This This saves extra extra keystrokes keystrokes and has proven proven to to be very very useful. useful. The The code to to do this this came from from the University University of of Texas Texas at Austin. Austin. 6. We We switched switched our our system system this this year year to to using using encrypted encrypted passwords. passwords. This This means that that passwords passwords are not not stored stored in any any readable readable form form on t.he the computer computer system, system, and if if an illegal illegal user should should gain access access to to the system, system, he/she he/she would would not not be able to to find find out out the passwords passwords of of any any other other users. We We feel this this feature feature is quite quite important important as the frequency frequency of of computer computer break-ins/attempts break-ins/attempts increases. 7. Software Software was added to to our our monitor monitor in order order to to record record the last last reader reader of of a file. Previously, date of of the last last read was recorded, recorded, while while both both Previously, only only the date the writer writer and date date were recorded recorded for for creating creating and writing writing a file. This This gives users the ability ability to to determine determine which which other other persons may may have been read ing files reading files.. •a Printer Printer Support Support The The support support on the 2060 for for various various printers printers in the SUMEX SUMEX community community has been greatly greatly enhanced this this past past year. year. 1. Support Support was added for for the Xerox Xerox Raven Raven printer printer at the the Welch Welch Road Road facility facility to allow allow spooling spooling and direct direct output output to to the printer. printer. In In addition, addition, code was added to to the spooler spooler to print print out out a header header page identifying identifying the user, filename, job. filename, and date date for for each job. 2. Similar Similar spooler spooler support support was added for for the the Xerox Xerox Dover Dover printer printer in Margaret hlargsret Jacks Hall. Hall. 3. SUMEX SUMEX installed installed a Printronix Printronix line printer printer at at Welch Welch Road to allow allow users print out out files remotely remotely from from the 2060. The The Printronix Printronix is connect.ed connected to to print SUMEX SUh?EX via via a twisted twisted pair pair serial serial line. We transformed transformed TENEX TENEX software software to the normal normal TOPS-20 TOPS-20 line printer printer 4. \Ve spooler spooler program program to look look out out for for users who who had accidentally accidentally printed printed an 'unprintable ‘unprintable file', file’, meaning meaning a binary binary file file of of some sort sort which which does not not cont:1in contain legible legible characters. characters. We We do this this both both by by counting counting the number number of of binary binary characters characters in the first first page of of the file, and by by not not printing printsing the file if t.he the count count exceeds exceeds a certain certain threshold. threshold. A A similiar similiar scheme is also used taking taking int.o into account account the vertical vertical motion motion of of the first first page. 5. Additional Additional modifications modifications were made to the LPlO LPlO line printer printer driver driver software in the TOPS-20 TOPS-20 monitor monitor to to improve improve the reliability reliability of of using using this this software line printer, printer, which which came from from our our KilO KIlO system. system. 6. \Ve We greatly greatly enhanced our our support support of of the IMAGEN IMAGEN Imprint-lO Imprint-10 laser printa /TCP Ethernet printer this this past past year. year. A A new IP IP/TCP Ethernet interface interface was installed installed on the prin ter (discussed further printer further in Section Section LA.2.4 I.A.2.4 replacing replacing the existing existing serial serial interface. int,erface. This This new interface interface allows allows for for more more efficient efficient printer prinber 17 Feigenbaum E. A. Feigenbaum Timesharing Systems Progress - Timesharing 5P41 RR00785-11 operation, and greater greater flexibility flexibility in choosing output output modes, number number of operation, pages, and other other features. features. We have also implemented implemented aa copies, header pages, TOPS-20 spooler for the Imagen Imagen as as well. well. TOPS-20 •l User User System System Software Software 1. We have continued continued to assemble and maintain maintain a broad broad range of utilities utilities 1. support software software on the 2060. These include and user support include operational operational aids, statistical packages, DEC-supplied DEC-supplied programs, programs, text text editors, editors, text text search statistical programs, file space space management management programs, programs, graphics graphics support, support, texttextprograms, formatting and justification justification assistance, magnetic magnetic tape conversion conversion aids, formatting many more. We also are importing importing software soft,ware tools and packages and many wherever necessary to avoid avoid reinventing reinventing the wheel and wasting wasting our our own own wherever efforts. Packages have been imported imported from from Texas Instruments, Instruments, Columbia Columbia efforts. University, the University University of Texas at Austin, Austin, Yale University, University, and other other University, Stanford sites. Stanford SUMEX has continued continued to provide provide to to its users the latest latest releases releases of 2. SUMEX various Lisp Lisp dialects dialects that that run run under under TOPS-20. TOPS-20. This This past past year year we agreed various provide disk disk space to store store the 'official' ‘official’ version version of of Interlisp-20 Interlispfrom to provide from XEROX due to the fact fact that that the machine machine at Xerox Xerox used to support support XEROX Interlisp before before was being removed. removed. Interlisp-lO Interlisp is now officially Interlispofficially maintained at SUMEX SUMEX by our our staff staff and XEROX XEROX personnel. personnel. In addition, addition, maintained continue to support support the full full variet.y variety of of LISPUSERS LISPUSERS packages. we continue Portable Standard Standard Lisp Lisp (PSL) (PSL) developed at the University University of Utah Utah has Portable a.lso been installed installed on SUMEX. SUMEX. also We continue continue to use use MM, MM, a very very powerful flexible mail mail system, system, on the 3. We powerful and flexible 2060. Electronic bulletin bulletin boards boards are also extensively-supported extensively-supported at 2050. Electronic at SUMEX. These provide rather informal informal mechanism mechanism for for community community SUMEX. provide a rather Other bulletin bulletin boards, boards, read and contributed contributed to to discussions and debates. Other throughout the INTERNET INTERNET community, are available throughout community, available for for perusal perusal at SUMEX. These bulletin bulletin boards boards cover cover such topics topics as AI AI Discussions, Discussions, Micro Micro SUMEX. Computers, Terminals, Terminals, and \Vorkstations. Workstations. Computers, SIJhIEX participates part,icipates with with other other Stanford Stanford sites in a general general license for for 4. SUMEX access to to the SCRIBE SCRIBE text-formatting text-formatting system system from from UNILOGIC, UNILOGIC, including including access versions too run run under under TENEX, TENEX, TOPS-20, TOPS-20, and UNIX. UNIX. SCRIBE SCRIBE is the the versions t’ preferred preferred tool tool for for text text preparation preparation at at SUMEX. SUMEX. 5. S. Versions Versions of of various various user utilities utilities and system system utilities utilities were updated updated throughout throughout the the year. year. These programs programs included included network network server server processes, processes, statistical stat istical packages, packages. system system daemon daemon programs, programs, and several several programs programs for for processing processing electronic electronic mail. mail. 6. 5. Various Various system system network network tables tables and networking networking softwltre softwa.re were updated updated to to sccommoda,te accommodate the ARPANET ARPANET split split that that occurred occurred this this year. yea.r. The The network network change effectively effectively split split the the ARPANET ARPANET into into two two networks, networks, the the MILNET, MILNET, which which is a secure private private part, part, and the the rest rest of of the the ARPANET, ARPANET, which which operates operates as the the original original ARPANET ARPANET did did before. before . • Docermentc?tion Documentation and and Education Education l \Ve \Vt> have expended expended considerable considerable effort effort to to develop, develop, maintain, maintain, a.nd and facilitate facilitate access access to to our our documentation documentation so to to accurately accurately reflect reflect available available software. software. The The Feigenbaum E. A. Feigenbaum 18 Progress Progress - Timesharing Timesharing Systems Systems 5P41 RR00785-11 RR00785-11 HELP IHELP and Bulletin Bulletin Board Board subsystems subsystems have hazTebeen important important in this this effort. effort. As As subsystems publish a bulletin bulletin or subsystems are updated, updated, we generally generally publish or small small document document describing more and more more changes occur, occur, it it becomes more more describing the changes. As more difficult pointers. Within difficult, for for users to to track track down down all of of the the change pointers. Within manpower manpower limits, limits, we are in a continual continual process of of reviewing reviewing the existing existing documentation documentation system programs now system for for compatibility compabibility with with the programs now on line and to integrate integrate cha.nges This also will will be done with with a view view toward toward changes into into the main main documents. documents. This developing better tools developing better tools for for maintaining maintaining up-ta-date up-to-date documentation documentation, . •l Software Sharing Software Sharing stated previously, we firmly firmly believe in importing importing rather rather than than stated previously, reinventing reinventing software software where where possible. As As noted noted above, a number number of of the the brought up are from packages we have brought from outside outside groups. groups. Many Many a.venues avenues between the system projects, other exist exist for for sharing sharing between system staff, staff, various various user projects, other facilities, and vendors. The facilities, vendors. The advent advent of of fast and a.nd convenient convenient communication communica.tion facilities facilities coupling coupling communities communities of of computer computer facilities facilities ha.s has ma.de cooperation made possible effective effective intergroup intergroup cooperation and decentralized decent,ralized maintenance maintenance of of software software packages. 1. As 2. The The TENEX, TENEX, TOPS-20, TOPS-20, and UNIX UNIX sites on the ARPANET ARPANET have been a good model model for for this this kind kind of of exchange based on a functional functional division division of of labor by-product labor and expertise. expertise. The The other other major major advantage advantage is that that as a by-product of particular software, personal of the constant cons.tant communication communicat.ion about about particular soft.ware, personal relat.ionships between staff relationships between staff members members of of the various various sites develop. develop. These collegial collegial interactions interactions serve to to pass general general information information about about software software tools tools and to encourage encourage the the exchange of of ideas among among the sites. Certain Certain common problems are now common problems now regularly regularly discussed on a mUlti-site multi-sit,e level. 3. vVe We continue continue to to draw draw significant significant amounts amounts of of system system software software from from other other ARP ANET sites, reciprocating ARPANET developments. reciprocating with with our our own own local developments. Interactions backup support, Interactions have included included mutual mutual backup support, experience experience with with various various hardware hardware configurations, configurations, experience experience with with new types types of of computers computers and operating operating systems, systems, designs for for local networks, networks, operating operating system system enhancements, utility or language software, and user project project enhancements, software, utility or collaborations. We collaborations. We have been able to to import import many many new pieces of of software soft.wsre and improvements improvements to to existing existing ones in this this way. way. Examples Examples of of imported program MM, imported software software include include the message message manipulation ma.nipulat,ion program Mhf, SAIL, SAIL, PASCAL, A.X Ethernet PASCAL, SOS, INTERLISP, INTERLISP, the C compiler, compiler, V VAX Ethernet code, the PHOTO program, ARPANET host PHOTO program, ARPANET host tables, tables, various various user utilities, utilities, and many many others. others. 4. Fin:.dly, Finally, we also have assisted groups groups that that have interacted interacted with with SUMEX SUMEX user projects projects in acquiring acquiring access access to to software software available available in our our community. community. \Ve providing tape preparation and copy \Ve are repeatedly repeatedly providing tape preparation copy service to to many many SU~[EX-AIM projects to SUME.X-AIM projects to aid in sharing sharing their their software software with with outside outside requestors. requestors. 19 E. A. A. Feigenbaum Feigenbaum 5P41 5P 41 RR00785RR00785- 11 Progress - Timesharing Timesharing Systems Systems Progress DECsystem ZOL?O/TOPS-20 2020/I'OPS-20 System System DECsystem Monitor Upgrade Upgrade --- Our Our 2020 system system has continued continued to to run run very very reliably reliably this this 1. Monit.or past year. year. We We have have updated updated the the 2020 monitor monitor with with bug bug fixes fixes and and performance performance past improvements regularly. regularly. There There will will likely likely be few further further monitor monitor releases for for improvements the 2020 since itit does not not support support extended extended addressing addressing and and there there are no plans plans the to add add this this feature. feature. to Demo Cont,rols Controls --- We \Ve continue continue to to use the the 2020 system system for for demos of of AI AI systems systems 2. Demo developed at at SUMEX. SUMEX. This This demo demo system system takes takes advantage advantage of of the the “pie-slice” "pie-slice" developed scheduler in the the TOPS-20 TOPS-20 release 4 monitor. monitor. We vVe now now guarantee guarantee dedicated dedicated scheduler large fraction fraction of of the the machine machine but but also allow allow others others to to do useful useful work work users a large when the the demo demo demand demand is low. low. This This system system has nicely nicely met met the the needs of of both both when groups. groups. v:4-YjUNIX Systems We 11/780 system \Ve continued continued to to provide provide systems systems support support for for the the VAX/UNIX VAX/UNIX 11/780 system (named ‘'AIMVAX') shared by by the the SUMEX-AIM SUMEX-AIM community community and Stanford Computer (named AIMVAX’) shared Stanford Computer Department. Various Various efforts efforts included included supporting supporting the the UNIX UNIX monitor, monitor, installing installing Science Department. network software, software, and in bringing bringing up various various user subsystems. subsystems, new network Further Further development development has continued continued in support support of of the the SUMEX-AIM SUMEX-AIM File File Server Server (SAFE) based on a VAX V A..'C 11/750. vVe successfully successfully converted converted SAFE SAFE to to Berkeley Berkeley Unix Unix 4.2 (SAFE) llj7.50. We server. server, and with with the help of of the the Computer Computer Science Department, Department, converted converted the the Ethernet Ethernet Pup software software to to run run under under UNIX UNIX 4.2 Pup E. A. Feigenbaum Feigenbaum 20 Progress Progress - Professional Professional Workstations Workstations SP41 FjP41 RR0078S-11 RR00785-1 I I.A.2.4. Professional Workstations I.A.2.4. Professional Workstations Our Our ongoing ongoing movement movement to professional professional workstations workstations is taking taking on several forms. forms. We We continue continue to carry carry out out our our acquisition acquisition plans plans for for acquiring acquiring Lisp Lisp machines machines for for use use by by the SUMEX SUhlEX community, community, as well well as investigating investigating the the use use of of remote remote virtual virtual graphics graphics and new This lower This work work is prototypical prototypical of of what what other other lower cost workstations workstations in our our environment. environment. groups groups will will face and we hope will will serve to to find find effective effective solutions solutions to to common common problems. problems. Lisp-based Scientific Workstations Lisp-based Scientific Workstations SUMEX SUMEX carefully carefully developed developed and implemented implemented our our equipment equipment acquisition acquisition plan plan for for year year eleven by by buying buying seven Xerox Xerox 1108 Lisp Lisp machines machines for for use by by SUMEX-AIM SUMEX-AIM projects. projects. Two purchased with provide floating Two of of these machines ma.chines were purchased with special upgrade upgrade packages to provide floating point. capability, capability, expanded expanded microcode, microcode, and expanded expanded memory. memory. Our Our experiences with with these machims machines will will be reported reported in next next years years report. report. The The XEnOX XEROX Dolphin Dolphin on loan to to Rutgers Rut*gers University University was returned returned to to SUMEX SUMEX this this year. year. This This Dolphin Dolphin had effectively effectively served the Rutgers-AIM Rutgers-AIM community community in setting setting up their their Ethernet provided initial Now Ethernet network network and provided initial exposure to to the Lisp Lisp machine machine technology. technology. Now that that that that experiment experiment is successfully successfully completed, completed, the Dolphins Dolphins will will be used for for AI AI system system development development at Stanford. Stanford. SU~fEX SUMEX installed installed two two SYMBOLICS SYMBOLICS 3600 Lisp Lisp machines, machines, purchased purchased with with DARPA DARPA funding, funding, for for use use within within the Heuristic Heuristic Programming Programming Project Project (HPP) (HPP) at at their their new location location at Welch Welch Road. Road. \Ve We are currently currently awaiting awaiting a new release of of the Symbolics Symbolics Operating Operating System System software before we can provide Ethernet access to our file server from software before provide Ethernet access to our file server from these machines. machines. The The 3600'5 3600’s are used regularly regularly by by members members of of the HPP. HPP. We We st.ill st,ill are using 4 preproduction preproduction models of of the Dolphin Dolphin workstations. workstations. One preproduct.ion preproductzion model model has been exchanged for for a production production system, system, and we are on schedule with with XEROX XEROX to exchange the remaining remaining 4 machines machines for for production production models at no extra extra cost. cost,. This This process is hampered hampered by by the rate rate at which which XEROX XEROX themselves themselves can get prod llction machines. product,ion machines. \Ve studied studied the benefits benefits of of buying buying the extended extended memory memory and microstore microstore upgrades upgrades to the Xerox as Xerox 1108 Dandelion Dandelion announced announced at AAAI-83 MI-83 as being being "under “under development." development.” \Ve We concluded concluded that that some users would would benefit benefit greatly greatly from from these enhancements enhancements and others others not not a.t a.t all. The The most most marked marked improvements improvements came from from system system which which were extremely extremely memory memory limit.ed. NEOM)"CIN. SUMEX limited, such as as NEOMYCIN. SUMEX will will be acquiring acquiring two two 1108's 1108’s with with the upgrades upgrades for for full full t.ime time use use and testing. testing. A close relationship relationship between between SUMEX SUMEX and the newly-formed newly-formed Center Center for for the Study Study of of Langnage Ln.ngusge and Information Information (CSLI) (CSLI) at Stanford Stanford was established. established. This This has already already benefit.('d benefitsed SU~1EX SLJhlEX (and (and the ONCOCIN ONCOCIN project project in particular) particular) in the loan, by by CSLI CSLI to to SUMEX, SUMEX, of of two two Xerox Xerox 1108's 1108’s which which have been in constant constant use use by by researchers researchers since January January 12th. 12th. The The SUI,1EX SUhlEX staff staff assisted CSLI CSLI in bringing bringing their their DecSystem20 DecSystem20 and network network environment environment online. line. CSLI CSLI has informally informally expressed an interest interest in working working on the problem problem of of distributed distributed AI AI computation computation with with SUMEX SUMEX researchers. researchers. CSLI CSLI will will have 110 1108's 1108’s on the Ethernet Ethernet wit.hin within the ye:1r. year. This This resource suggests some exciting exciting solutions solutions to former former compute-bound compute-bound problems. problems. The The ONCOCIN ONCOCIN group group has already already implemented implemented a preliminary preliminary network-based network-based Interactor to run Int,eractor which which permits permits elements elements of of ONCOCIN ONCOCIN run concurrently concurrently on different different machines. of this this writing, writing, the Reasoner and its its Debugger Debugger have been made to to run run ma.chines. As of transparently transparently in this this mode, and to make good use of of both both processors. 21 21 E. A. A. Feigenbaum Feigenbaum 5P41 RR00785-11 RR00785- 11 Progress - Professional Professional Workstations Workstations Virtual Virtual Graphics Graphic3 SU!'vlEX SUMEX continued continued the development development of aa Virtual Virtual Graphics Graphics system system written written in Interlisp-IO and running Interlisprunning on our 2060. 2060. Any Any user running running the V system system on aa workstation workstation can then then use use the t,he package on the 2060 to drive drive the graphics graphics display display on the workstation. workstation. A A current application is to take nuclear magnetic resonance data on the 2060 and display current a.pplication is nuclear magnetic data display t,he the atoms atoms and a.nd their their bonds on a SUN SUN workstation workstation by using splines. This This development development is is in its infancy, infancy, but but, is is opened ended and has great great potential potential with with the price price of workstations workstations capable capa.ble of decent graphics graphics reaching reaching the two two to three three thousand thousand dollar dollar range. It It allows allows those users who who cannot cannot afford afford expensive lisp machines machines to have full full graphics graphics power power available available to them them by doing doing the actual actual graphics graphics applications applications on a large time time shared system, system, and then then doing doing t.he the graphics graphics itself itself remotely remotely on a less less expensive workstation. workstation. This This development development can help users take ta.ke advantage advantage of the computing computing power power of the DECsystem DECsystem 20, while while providing providing many many of the t,he high high speed speed graphics graphics advantages advantages of the Lisp Machines. Machines. Apl)/e Development .4pple Workstation Workstation Development SUMEX-AIM SUMEX-AIM has initiated initiated a development development project project to pursue pursue the effectiveness effectiveness and possible use of low cost personal workstations within our environment. use personal workstations within our environment. After After examing examing a number number of new personal personal computers computers and workstations workstations on the market, market, such as as the Hewlett Hewlett Packard Packard 150. 150, IBM IBM PC, Sun workstations, workstations, and others, others, we chose chose the Apple Apple Macintosh Macintosh and Apple Apple Lisa on which which to begin our our work work in this this area. These machines machines were chosen technically t*echnically due to their their built built in graphics, graphics, networking, networking, mouse, windows, windows, and menu support. support. vVe We also considered considered the very very beneficial beneficial relationship relationship formed formed between between Apple Apple and Stanford Stanford University which provides provides us direct direct access access to Macintosh Macintosh hardware hardware and software software University which documentation plan to documentation which which is a necessity necessity for for the type type of of work work we plan to do. Our Our Macintosh Macintosh development development encompasses several areas ; 1. INFO-MAC INFO-AIAC Discussion Di.'3cu8sion List List An An electronic electronic discussion discussion list list was originated originated at SUMEX, SUMEX, and is currently currently maintained maintained here, to to foster foster sharing sharing and communication communication among among research groups groups and pursuing the serious use a.nd universities universities that that are interested interested is pursuing use of of the Macintosh Macintosh within within their their respective respective environments. environments. This This list list has been highly highly successful in collaborating collaborating on Macintosh Macintosh development development and the sharing sharing of of ideas. The The discussion over 50 sites, and well well over over 1000 participants. participants. discussion list list currently currently contains contains over C Development Development Environment Environment 2. C A vital vital link link in our our development development of of Macintosh Macintosh software software is creating creating a C based A development system system on our our VAX VAX computers computers for for the the coding coding and downloading downloading of of development software. Utilizing Utilizing existing existing MC68000 MC68000 C cross compilers compilers on our our VAX, VA.,X, we are software. developing the the necessary linkages linkages in order order to to make make the the appropriate appropriate system system developing to rout,ines rOll tines in the the Macint.osh Macintosh ROM’ ROM's for sophisticated sophisticated graphics graphics and system system calls to s for rela,t,ed functions. functions. related 3. Macintosh Afacintosh print print servers servers In order order to to effectively effectively use the the Macintosh Macintosh as a stand-alone stand-alone workstation, workstation, we will will In provide the ability to print out Macintosh developed files on our IMAGEN provide the ability to print out Macintosh developed files on our IMAGEN printers. laser printers. Applebu8 to to Ethernet Ethernet Interface Inter face 4. Applebus This development development involves involves the the hardware hardware and and software software necessary to to be able to to This A. Feigenbaum Feigenbaum E. A. 22 Progress - Professional Professional Workstations Workstations 5P41 5P41 RR00785-11 RROO785-11 access access various various file and print print servers on our our Ethernet Ethernet from from aa Macintosh. Macintosh. Macintosh blscintosh will will be be connected to the Apple Apple network network called Applebus. Applebus. hardware hardware provides provides an interrace interface between between Applebus Applebus and 10MB 1OMB Ethernet. Ethernet. software software necessary for this this project project involves involves formulating formulating Macintosh Macintosh file and block block level I/O I/O requests into into properly properly formatted formatted Internet Internet packets. packets. The Our Our The level 5. Virtual Virtual Graphics Graphics on on a Lisa Lisa The The Virtual Virtual Graphics Graphics System, as as previously previously reported, reported, is in great great need need of a low cost workstation workstation on which which it it can run. run. We We have started started a project project to port port the Virtual Virtual Graphics Graphics system syst,em to a Apple Apple Lisa in hopes of providing providing to our our users high speed will report report further further on this this project project speed graphics graphics at remote remote locations. locations. We will in next next year's year’s report. report. Anticipating Ant,icipating the popularity popularity of our our Macintosh Macintosh developments, developments, we are fully fully prepared prepared to make our our efforts efforts available available to other other research sites, sibes, Universities, Universities, and non-profit non-profit institlltions institutions on a.a royalty-free royalty-free basis in hopes of fostering fostering continued continued development development and communal communal sharing. sharing. In addit.ion addition to Lisp-based Lisp-based scientific scientific workstations, workstations, we believe the use use of of low low cost workst.at,ions, workstnt,ions, which which offer offer suitable suitable local processing processing power, power, high high resolution resolution screens with with easy to t.o use use user interfaces, interfaces, and networking networking and communications communications abilities, abilities, are vital vital to the future future of of our our resource. Our Our Macintosh Macintosh and Lisa Lisa development development efforts efforts will will allow allow us to use use and experiment experiment with w&h these workstations workstations in our our environment. environment. Hewlett Hewlett Packard Packard Development Development SUl\fEX SUMEX assists the ONCOCIN ONCOCIN Project Project is developing developing a computing computing environment environment for for developing part developing AI AI applications applications based on HP HP 9836 workstations. workstations. These workstations workstations were part of peripherals for of aa. gift gift from from Hewlett Hewlett Packard Packard to the Oncocin Oncocin Project. Project. Additional Additional support support peripherals for t.he the 9835's 9836’s included included large capacity capacity disk disk drives, drives, color color monitors, monitors, graphic graphic tablets, tablets, and a laser printer. \Vork proceeding to print.er. Work is proceeding to network network these machines machines onto onto the SUMEX SUMEX Ethernet Ethernet as as soon as as suitablke suit.ablke networking networking hardware hardware is available available from from HP. HP. These machines machines will will be used for for projects within new and a.nd existing existing projects within Oncocin. Oncocin. 23 A. Feigenbaum Feigenbaum E. A. RR00785-11 5P41 RR00785-11 Networking and Communications Net,working Communications I.A.2.5. Networking and I.A.2.5. Networking and Communications Communications A aspect of A highly-important highly-important of SUMEX-AIM SUMEX-AIM is effective effective communication communication with with remote remote users and between between the growing growing number number of of machines machines available available within within the SUMEX SUMEX resource. In addition addition t.o t.o the the economic economic arguments a.rguments for for terminal terminal access, access, networking networking offers offers other other advantages including improved improved inter-user inter-user communications communications and a.dvantages for for shared shared computing, computing, including more more effective effective software software sharing. sharing. Users accessing a remote remote computer computer will will use a hardline hardline connection connection to to the computer computer as as aa. standard standard of of comparison. comparison. Local Local networks networks stand stand up well well in this this comparison comparison but but remote remote network network facilities facilities do not. not. Data Data loss loss is not not a problem problem in most most network network communications; communications; in fact, fact, wit·h with the t,he more more extensive extensive error error checking checking schemes, data data integrity integrity is higher higher than than for for a long distance distance phone link. link. On On the the other other hand, hand, remote remote networking networking relies upon upon shared shared use use of of communication communication lines for for widespread widespread geographical geographical coverage at at substantially substantially reduced reduced cost. However, However, unless enough enough total total line capacity capacity is provided provided to meet meet peak loads, substantial substantial queueing jams result queueing and traffic traffic jams result in the the loss loss of of terminal terminal responsiveness. We We continually continually monitor monit,or the load statistics statistics for for our our direct, direct, dialup, dialup, and TYMNET TYMNET lines to to avert avert logjam logjam situations. situations. n-AfNET TYMNET TYMNET provides provides broad broad geographic geographic covera.ge covera,ge for for terminal terminal access access to to SUMEX SUMEX from from throughout throughout the country country and increasingly increasingly from from foreign foreign countries. countries. With With the installation installation of of our our new DEC2050 DEC2060 computer computer system system in January January of of 1983, we installed installed new TYMNET TYMNET equipment. After After the initia.l initia,l debugging debugging of of the new equipment equipment (called (called TYMCOM) TYMCOM) the equipment. equipment However, some months months after after the installation installation it it was equipment has been quite quite reliable. reliable. However, discovered that the XON/XOFF protocol between the Tymcom and the 2050 had discovered t,hat XON/XOFF protocol between Tymcom 2060 had not not been properly The The number number of of user properly specified specified in the Tymcom Tymcom and was corrected. correct.ed. complaints complaints about about connection connection problems problems have been greatly greatly reduced. reduced. This This is thought thought to to be the result result of of improved improved "backbone" “backbone” lines within within Tymnet Tymnet and the installation installation of of triple-duty triple-duty modems modems which which simplify simplify things things for for the users users.. •-tRPA.NET \Ve of We retain retain our our advantageous advantageous connection connection to to the Department Department of Defense's Defense’s ARPANET. now Agency This ARPANET. now ma.naged ma.naged by by the Defense Communications Communications Agency (DCA). (DCA). This connection connection has facilitated facilitated close collaboration collaboration with with the Rutgers-AIM Rutgers-AIM facility facility and many many other ot.her computer computer science groups groups that that are also on the net. \Ve We have maintained maintained good working working relationships relationships with with other other sites on the ARPANET ARPANET for for system system backup backup and software software interchange. day-bo-day working working interactions interactions with with remot.e remote facilities facilities would would not not be interchange. Such day-t.o-day possible without wit,hout the integrated integrated file transfer, transfer, communication, communication, and terminal-handling terminal-handling capabilities The capa.hilities nnique lrnique to the ARPANET. ARPANET. The ARPANET ARPANET is also key key to to maintaining maintaining ongoing tt' lIectual contact.s going in intellectual c0ntact.s between between SUMEX SUMEX projects projects such as as the Stanford Stanford Heuristic Heuristic Programming Progra.m.ming Project Project authorized authorized to to use use the net net and other other active active AI AI research groups groups in the ARPANET ARPANET community. community. This This past past year, year, SUMEX-AIM SUMEX-AIM participated participated in the split split of of the ARPANET ARPANET into into two two networks; networks: the MILNET, MILNET, which which is a highly highly secure strictly strictly DOD-related DOD-related part part of of the network, network, and the ARPANET, which ARPANET, which is the remainder remainder of of the ARPANET ARPANET sites. This This latter latter net net functions NET can only functions as as we knew knew t,he the ARPANET ARPANET before. before. The The MIL MILNET only be accessed accessed via via mail mail gat.eways. ga.t.eways. No TELNET TELI”:ET or or FTP FTP to to MILNET MILNET sites is allowed. allowed. In In addition, addition, access access to to the ARPANET AC's ARP,4NET T TAC’ s (Terminal (Terminal Access Controllers) Controllers) was restricted restricted this this past past year year to to only only t.hose t,hose users who who were granted granted TAC TAC access access cards, which which meant meant their their username username was rt'gistered Network Information registered with with the Network Information Center, Center, and they they were given given a password password with with E. A. Feigenballm Feigenbslim 24 5P4l 5P41 RR00785-l1 RRO0785-11 Networking Networking and Communications Communications which which they they could dial dial into into the ARPANET. ARPANET. SUMEX SUMEX arranged arranged for lor guest cards for for those users users who needed such access. access. \Ve We continue continue to be be called upon to interact interact with with outside outside organizations organizations which which are (or (or wish to be) connected connected to our our IMP. IMP. The The line to Advanced Advanced Information Information and Decision Decision Syst.ems Systems occasionally occasionally causes causes trouble trouble requiring requiring diagnosis. The The intended intended connection connection to Perceptronics Perceptronics Inc. has evidently evidently been canceled. EI'HERNEI' ETHERNET A sllbstantial substantial portion portion of our our system system effort effort this this past past year year went went into into continued continued development development of local Ethernet Ethernet facilities facilities which which link link the SUMEX SUMEX resource hardware hardware with with other other parts parts of the campus, campus, namely namely to 701 701 Welch Welch Road, which which is is the new location location of the Heuristic Heuristic Programming Progra.mming Project, Project, and to the Computer Computer Science Science Department Department building building on campus. \Ve We have also invested invested a great great amount amount of of effort effort this this year year to begin our our transfer transfer to a 10 megabit megabit Ethernet, Ethernet, while while continuing continuing support support of of our our current current 3 MB MB ethernet. ethernet. Specific Specific areas of of Ethernet Ethernet development development include: include: -- We We continued continued support support of of the Sequin reliable reliable packet packet protocol protocol and Leaf Leaf byte-level byte-level file transfer transfer protocol protocol to enable our our Xerox Xerox D machines machines to access access files on our our DEC20 DEC20 systems. The The Leaf Leaf server server had to be modified modified on the DEC20 DECPO t,his this past past year year when when we switched switched to using using encrypted encrypted passwords. The The LEAF LEAF server server implementation implementation for for the 4.2 BSD release of Unix Unix was also debugged and installed installed at SUMEX. SUMEX. This This allows allows us to to access access files stored stored on our our VAX VAX file servers servers from from either either our our 10MB 1OMB or or 3MB 3MB networks. networks. 1. Leaf Leaf server server The protocol is built built into The Leaf Leaf protocol into the lowest lowest levels of of the Dolphin Dolphin I/O I/O system, system, and allows allows any any file on a remote remote file file server server to to be accessed accessed as as easily easily as as a disk disk file in both paged or both or random random access access mode. The The latest latest updates updates to to the Sequin transport t,ransport level have made marked marked improvements improvements in efficiency. efficiency. The The 2020 now now performs Leaf performs Leaf file file transfers transfers with with a speed approaching approaching that that of of XEROX's XEROX’s dedicated dedicated file server. server. 2. TOPS-20 Ethernet Server We continued continued to to maintain maintain and improve improve the TOPS-20 Ethernet Seroer -- We Et.hernet under the Ethernet service under to t.he the TELNET TELNET and FTP FTP mentioned Leaf server, and mentioned Leaf server, TOPS-20 updates TOPS-20 operating operating system. system. This This included included updates programs, as well previously programs, well as as mail mail software, software, the previously network table table maintanence maintanence programs. programs. network Ethernet Ciate.way Gateway ---- Our Our Ethernet Ethernet gateway gateway software software has continued continued to to run run 3. Eth,ernet reliably and effectively. effectively. The The previous previous problems problems with with lost lost packets packets and delayed delayed reliably terminal response has been fixed, fixed, the the cause of of which which was a bad memory memory board board t.erminsl software bug bug in the the TOPS-20 TOPS-20 operating operating system. system. Serious Serious problems problems that that and a software affected our our net net connectivity connectivity to to other other parts parts of of campus campus were also discovered discovered affected repaired this this past past year, year, thereby thereby providing providing us with with over over 99% 99% net net and repaired connect.ivity to to the rest rest of of campus. campus. The The changes involved involved board board repair repair and connect.ivit,y modifications to to the topology topology of of the the campus campus Ethernet. Ethernet. modifications The gateway gateway itself itself was generalized generalized The networks where where previously previously itit had had networks currently have two two gateways, gateways, each currently networks. two two of of which which are 3MB 3MB and and networks, to handle handle 3 or or more more directly directly connected connected to only dealt dealt with with 2 such networks. networks. We We only handling the the traffic traffic between between three three local local handling 10MB network. network. one a 1OMB 4. Ethernet Ethernet TIP TIP (EtherTIP) (EtherTlp) --- The The EtherTIP EtherTIP provides provides multiple multiple terminal terminal access access to to the Ethernet. Ethernet. A A PUP PUP ethernet ethernet operating operating syst,em system was written written for for MC68000MC68000t.he and a MC6800O-based MC68000-based EtherTIP EtherTIP was built built base.d based on on this. this. based processors, and 25 E. A. A. Feigenbaum Feigenbaum E. Networking Networking and Communications Communications 5P41 RR00785-11 RR00785-11 The The EtherTIP EtherTIP software software has undergone undergone further further enhancements enhancements in the Portions of this work was done in conjunction Portions this work conjunction with with the Stanford Stanford Science Department. Department. Among Among those enhancements enhancements are the following: following: past year. year. Computer Computer a. It It now now accepts accepts incoming incoming connections connections to ‘to line printer printer ports, ports, and for for remote remote system system diagnosis. diagnosis. b. It It can simulate simulate the the "old" “old” Stanford Stanford EtherTIP EtherTIP for for users who who have not not yet yet made the transition transition to the new environment. environment. c. The The user interface interface is more more flexible flexible to suit suit the needs of an increasingly increasingly diverse diverse user community. community. The The EtherTIP EtherTIP software software has developed developed into into a very very stable stable system, system, and one enjoying enjoying good use use within within the SUMEX SUMEX community. community. 5. 10 AlB/SEC - SUMEX SUMEX made a major major move this this past past hfB/SEC Ethernet Ethernet Development Development -- year year to begin our our transfer transfer to to a 10 megabit/sec megabit/set network. network. While While the current current 3 megabit/sec megabit/set network network continues continues to to serve us well, well, many many new workstations workstations and printers printers are coming coming on the market market with with only only 10 MB/SEC MB/SEC interfaces, interfaces, and in addition, a.ddition, since 3 MB/SEC MB/SEC networks networks were only only used a very very few selected settings. settings. it it is becoming becoming increasingly increasingly difficult difficult to find find replacement replacement parts parts when when failures do occur. failures occur. Therefore, Therefore, this this past past year year saw several efforts efforts involved involved in installing installing and supporting supporting the SUMEX SUMEX 10MB/SEC lOMB/SEC Ethernet; Ethernet ; a. Reworking Reworking the the entire entire Ethernet Ethernet system system software software to to handle handle both both 3 and and 10 megabit megabit link link level level standards, standards, i.e., addressing addressing and encapsulation encapsulation are transparent transparent to to the the user user levels. We We similarly similarly made the the network network link link level protocols transparent protocols transparent to to the the the the user level software. software. In In this this way way one can communicated protocols on a 10MB/SEC communicated using using PUP PUP protocols lOMB/SEC ethernet ethernet and the the user software software does not not have to to change. protocols for b. Adding Adding address resolution resolution protocols for PUP PUP and and IP IP so that that the the 3~tB/SEC byte byte addresses can be translated 3MB/SEC translated to to 10 MB/SEC MB/SEC hardware hardware addresses for for the the link link level. This This enables one to to communicate communicate using using PUP PUP or IP IP between between 3 and and 10 megabit megabit hosts. or Integrating XNS XNS and and IP IP into into the the PUP PUP routing routing mechanism. mechanism. c. Integrating Solving some rather rather subtle subtle software/hardware software/hardware integration integration problems problems in d. Solving order to to simulate simulate “ethernet” "ethernet" on on the the HPP/Welch HPP/\Velch Road Road “twisted "twisted pair” pair" order ethernet. ethernet. e. Bringing Bringing up the the 3 MB/SEC MB/SEC EtherTIP EtherTIP on the the 10 MB/SEC MB/SEC network network was a e. proof tha,t that the the above worked. worked. ItIt was done without without any any changes to to the the proof TIP software software itself itself by by simply simply relinking relinking itit with with the the 10 MB/SEC MB/SEC system system TIP software. This This required required only only one additional additional piece of of logic. logic. When When a 10 software. MB/SEC host host wants wants to to communicate communicate using using PUP PUP which which is a 3 MB/SEC MB/SEC MB/SEC protocol. then then itit must must find find its its PUP PUP address address from from some host host on the the 10 protocol, MB/SEC network. network. The The gateway gateway maintains maintains a translation translation table, table, and and listens listens MB/SEC requests, thus thus translating translating the the 10 MB/SEC MB/SEC hardware hardware address into into for such requests, for "soft PUP PUP address,” address," and and replying replying to to the the requesting requesting host. host. a “soft HPP-SUAIEX Communication Communication link link -- The The Heuristic Heuristic Programing Programing Project Project 6. HPP-SUrZ1E.Y A. Feigenbaum Feigenbaum E. A. 26 Networking Communications Networking and Communications 5F41 RR00785-11 RR00785- 11 5P41 (HPP) 701 Welch Welch Road, adjacent adjacent to to the (HPP) relocated relocated from from its campus location location to to 701 Stanford Stanford campus. Since this this group group is a primary primary user of of the SUMEX SUMEX computer computer facility facility and the principal principal focus for for core AI AI research, a communication communication link link location and SUMEX SUMEX machine room room was imperative. imperative. Several between the new location communication communication schemes schemes for for establishing establishing a reliable reliable and relatively relatively fast link link were m icrowave, laser, infrared, infrared, direct direct ethernet ethernet (by (by trenching trenching considered, namely; namely ; microwave, and placing placing a direct direct ethernet ethernet cable), ATT's ATT’s Tl Tl service and others. others. All All of the above schemes schemes would would have necessitated necessitat,ed large budgetary budgetary outlays outlays and some some would would have imposed lengthy lengthy time time delays (getting (getting permits permits and the like) like) The idea of using bare copper telephone pair pair due to jurisdictional jurisdictional boundaries. The already in place looked very very attractive attractive especially if if reasonable reasonable speed speed and already reliability reliability could be achieved. The The wire wire distance between the above mentioned mentioned locations locations is approximately approximately 2000 2000 ft. ft. A A design goal was established to to try try to to communication link link with with Ethernet Ethernet type type speed speed ( 3MB/SEC) 3MB/SEC ) between develop a communication these these two t.wo locations. locations. Utilizing Ut.ilizing high high driving driving capacity capacity drivers drivers (differential) (differential) and ultra ultra high speed, speed, high high sensitivity sensitivity receivers a transceiver transceiver was designed designed and tested for for maximum maximum transmission transmission speed speed with with maximum maximum reliability. reliability. The The final final configuration configuration resulted resulted in aa, half half duplex duplex transmission transmission over over a bare copper twisted twisted pair pair in each direction direction utilizing Manchester Manchester coding at at a reliable reliable transmission transmission speed speed of of 1.25MBs/sec. 1.25MBs/sec. utilizing each direction direction for for an aggregate a.ggregate speed speed of of 2.5MBs/sec. 2.5MBs/sec. This This communication communication link link each has been been in operation operation for for about about six months months now without without any appreciable appreciable down time time or or noticeable error error rate rate or or data data delays. Many Many HPP HPP researchers are utilizing utilizing t,his t,his link link to communicate communicate with with SUMEX SUMEX and the University University Ethernet Ethernet network. and printers network. In addition, addition, various various Lisp Lisp machines machines printers located in the HPP HPP facility facilit*y a.nd and connected to to a local network network there are also able to communicate communicate with the University University network. network. with Ii'VTERNET SOF'1.1VARE One major major issue issue we face face at SUMEX-AIM SUMEX-AIM in support support of our our network network environment environment. is the lack of standardization standardization in network network protocols protocols among various various vendors. Currently, Currently, many many vendors /TCP) protocols. vendors are adding adding support support to their their products products for the Internet Internet (IP (IP/TCP) protocols. SUMEX /TCP protocols SUMEX continues cont.inues to support support the the IP IP/TCP protocols on the DEC2060, and we are current,ly alpha-testing alpha.-testing a release release of of Interlisp-D Inter-lisp-D which which also supports supports IP IP/TCP protocols. In currently /TCP protocols. addition, /TCP software nddit,ion, we sucessfully adapted the IP IPjTCP software to our our VAX VAX systems running running UNIX UNIX 4.2BSD. L~BSD. This This Vax Vax TCP TCP adaptation adaptation involved involved provisions provisions for subnet subnet routing, routing, 3 MB/SEC MB/SEC byte swap problems, encapsulation encapsulation problems problems and 10 10 MB/SEC MB/SEC debugging with with our our byte gateways. gntewa.y.5. 27 27 E. A. Feigenbaum Progress - Progress in Core Research 5P41 RR00785-11 I.A.2.6. Progress in I.A.2.6. Progress in Core Core Research Research Over Over the past year year we have continued continued to support support several core research activities activities aimed at developing developing information information resources, basic AI AI research, and tools tools of general interest interest to the SUMEX-AIM SUMEX SUMEX is providing providing only only partial partial support support for for these SUMEX-AIM community. community. projects, project,s, with with complementary complementary funding funding coming coming from from ARPA, ARPA, ONR, ONR, NLM NLM and NSF contracts contracts and grants grants to the Stanford Stanford Heuristic Heuristic Programming Programming Project. Project. Core Research Research Core Research at SUMEX-AIM SUMEX-AIM focuses focuses on understanding understanding the roles of knowledge knowledge in symbolic symbolic problem problem solving solving systems, its representation represent.ation in software software and hardware, hardware, its use use for for inference. inference, and its acquisition. acquisition. We are continuing continuing to develop new tools tools for system system builders builders and to improve improve old ones. ones. The The research crosses crosses aa number number of application application domains, domains, as as reflected reflected in the subprojects subprojects discussed earler, earler, but but the main main issues issues that that we are addressing addressing in this this rese3rch research are a.re those fundamental funda.mental to all aspects of AI. AI. We We believe this this core research is broadening broadening and deepening deepening the groundwork groundwork for for the design and construction construction of of even more more capable and effective effective biomedical biomedical systems. systems. As mentioned mentioned above, although although our our style style of research is largely largely empirical, empirical, the questions questions we are addressing addressing are fundamental. fundamental. The The three three major ma.jor research issues issues in AI AI have, since its beginning, beginning, been knowledge knowledge representation, representation, control control of of inference inference (search), and learning. learning. \Vithin Within these topics, topics, we wiII will be asking asking the following following kinds kinds of of questions. As our our work work progresses, we hope to leave behind behind several several prototype prototype systems systems that that can be developed by others others in the t.he medical medical community. community. 1. 1. Knowledge Knowledge Representation Representation -- How How can we represent represent causal models and structural What benefits of st.ructura,l information? information? What are the relative relative benefits of logic-based, logic-based, rulerulehased, based, and frame-hased frame-based systems? systems ? How How can we represent represent temporal temporal relations relations and events events so that t,hat rea.'Soning reasoning over over time time is efficient? efficient? 2. Knowledge Knowledge Acquisition Acquisition -- How How can an expert expert system system acquire acquire new knowledge knowledge wit,hout consuming without consuming substantial substantial time time from from experts? Can Can we improve improve the knowledge paradigm enough knowledge engineering engineering paradigm enough to to make make a difference? difference? Can Can automatic automatic learning programs be designed that will work across many disciplines? lea,rning programs that will work many disciplines? Will Will cooperative cooperat.ive man-machine man-machine systems systems be able to open the communication communication channel channel het,ween expert and expert expert system? bet,wern expert Knowledge Utilization Utilization -- By By what what inference inference methods methods can a variety variety of of sources of of 3. Knowledge knowledge of of diverse diverse types types be made to to contribute contribute jointly jointly and efficiently efficiently knowledge towa.rd solut,ions? sollltions? What \Vhat is the the nature nature of of strategy strategy and control control information? information? toward Plans for for the Coming Coming Year Plunv Several syst,ems systems have been developed developed in recent recent years years to to serve as vehicles vehicles for for Severa, knowledge engineering engineering and research research on knowledge knowledge representation representation and its its use. Knowledge Knowledge knowledge acquisition (including (including machine machine learning) learning) and and advanced advanced architectures architectures for for AI AI will will be the the two two acquisition areas of of most most new activity activity in the coming coming year. year. Research on these topics topics obviously obviously must must draw on on-going on-going work work in representation representation and and control. control. draw In part,iculsr, particular, we will will focus on In • Induct,ive Inductive learning learning of of MYCIN-like MYCIN-Iike rules rules from from case case data data in the the domain domain of of diagnosing disorders disorders where where the the chief chief complaint complaint is jaundice; jaundice; diagnosing l • Learning Learning from from experience experience in domains domains where where the the means for for interpreting interpreting new new l A. Feigenbaum Feigenbaum E. A. 28 Progress Progress - Progress Progress in Core Core Research SP'l1 5P4 1 RR0078S-11 RROO785-11 data data are largely largely contained contained in the the emerging emerging (and (and thus thus incomplete incomplete and not not wholly wholly correct,) correct) theory; theory; •l Learning by watching Learning by watching NEOMYCIN; NEOMYCIN; a medical medical expert expert diagnose cases presented cases presented by by •l Investigating Invest,igating complex complex signal signal understanding understanding systems systems for for ways ways to exploit exploit and represent represent concurrency concurrency with with a view view toward toward hardware hardware and software software architectures architectures that that may may be capable of of several several orders orders of of magnitude magnitude improvement improvement in performance. performance. Further and the Heuristic Further information information on the core research at at SUMEX-AIM SUMEX-AIM Heurist,ic Programming Project can be found in the Projects section starting on page 89. Programming Project found Projects section starting 29 E. A. A. Feigenbaum Feigenbaum 5Phl RR00785-11 RR00785- 11 5P~1 Progress Progress - Resource Operations Operations Statistics Statistics I.A.2. Operations Statistics I.A.2.7. 7. Resource Resource Operations Statistics The The following following data data give an overview overview of of various various aspects of of SUMEX-AIM SUMEX-AIM resource usage. usage. There There are 5 subsections subsections containing containing data data respectively respectively for: for: 1. 1. Overall Overall resource loading loading data data (page 31). 2. Relative Rela,tive system system loading loading by by community community (page 33). 3. Individual Individual project project and community community usage (page 36). 4. Network Network usa.ge usage data data (page 44). 5. System System reliability reliability data data (page 44). For For the most most part, part, the the data data used for for these plots plots covers the the entire entire span of of the SUMEX-AIM SUMEX-AIM project. project. This This includes includes data data from from both both the TENEX TENEX KIlO KIlO system system and the curren current t DECsystem DECsystem 2060. At At the point point where where the SUMEX-AIM SUMEX-AIM community community switched switched over over to the 2060 (February, (February, 1983), you you will will notice notice severe changes in most most of of the graphs. graphs. This This is due to many rnany reasons rea.sons which which I will will mentioned mentioned briefly briefly here; here ; 1. 1. Even Even t.ho1]gh though the Tenex Tenex operating operating system system used on the KilO KIlO was a forerunner forerunner of of t.he the current current Tops20 Tops20 operating operating system, system, the Tops20 Tops20 system system is still still different different from from Tenex is many many ways. ways. Tops20 Tops20 uses uses a radically radically different different job scheduling Tellex job scheduling mechanism, mechanism, different different methods methods for for computing computing monitor monitor statistics, statistics, different different I/O I/O routines, routines, etc. In general. general, it it can not not be assumed that that statistics statistics measured measured on the Tenex Trnex syst·em syst,em correlate correlate one to one with with similar similar statistics statistics under under Tops20. Tops20. 2. 3. The The KLlO KLlO processor processor on the 2060 is a faster faster processor processor than than the KilO KIlO processor processor used previously. job running previously. Hence, a job running on the KLlO KLlO will will use use less less CPU CPU time time than than the sa.me job running sa,me job running on the KilO. KIlO. This This aspect is further further complicated complicated by by the fact fact that that the SUMEX SUMEX KilO KIlO system system was a dual dual processor processor system. system. 3. The The SUMEX-AIM SUMEX-AIM Community Community was changing changing during during the time time of of the transfer transfer to the just been t,he 2060. SOGO.The The usage of of the GENET GENET community community on SUMEX SUMEX had just pha:sed phased out. out. This This part part of of the community community accounted accounted for for much much of of the the CPU CPU time time used by by the AIM AIM community. community. Since the the purchase purchase of of the 2060 was partially partially funded funded by by the Heuristic Heuristic Programming Programming Project Project (HPP), (HPP), an additional additional number number of of HPP HPP Core Core Research Projects Projects started started using the 2060, increasing increasing the Stanford Stanford communities usage usage of of the machine. machine. And And finally, finally, the move to to the 2050 2060 occurred occurred communities during during a pivotal pivotal time time in the community community when when more more and more more projects projects were either eit,her moving moving to their their own own local timesharing timesharing machines, machines, or or onto onto specialized specialized Lisp Lisp workstations. workstatJions. It It also was the time time for for the closure of of many many long time time SUMEXSUMEXAIM projects, projects, like like Dendral Dendrsl and Puff/VM. Puff/VM. AIM Any ,4ny conclusions conclusions reached by comparing comparing the the data data before before and after after February, February, 1983 should be done with with caution. caution. The The data data is included included in this this years years annual annual report report mostly mostly for for should casual comparison. comparison. Starting Starting next next year, year, only only data data from from the 2060 will will be recorded recorded on the annual annual report. report. Readers will will be referred referred to to previous previous annual annual reports reports (such as as this this one) for for data da.ta from from the KIlO KIlO Tenex Tenex system. system. Feigenbaum E. A. Feigenba.um 30 Progress Progress - Resource Operations Operations Statistics Statistics 5P41 5P4 1 RR00185-11 RR00785fI 1 , Ot'erail Overall Res~urce Reshree Loading Loading Data Data The over the life life The following following plots plots display display several different different aspect.s aspects of system system loading loading over of t.he the current current t,he project. project. This This data data includes includes usage of the the Tenex Tenex KilO KIlO system system and the DECsystem DECsystem ~050. 2060. These plots of jobs jobs plots include include total total CPU CPU time time delivered delivered per per month, month, the peak number number of logged in, and the peak load average. The variable is The monthly monthly "peak" “peak” value value of of a given given variable the average of Thus, these of the daily daily peak values values for for that that variable variable during during the month. month. Thus, "peak" not reflect reflect “peak” values are representative representative of of average monthly monthly loading loading maxima maxima a.nd a.nd do not the days, which which are much much higher. higher. t,he largest largest excursions excursions seen seen on individual individual days, 800 800r *‘Or Total Total CPU CPU Usage Usage Hours/Month Hours/Month 600 600 - 400 400 - 200 200 - 1983 $74 1975 1976 1977 1978 1979 1980 1981 1982 1984 1985 4 o~--~----~----~----~--~~----~----~--~~--~----~----~ 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1974 Figure 7: Figure 7: Total Total CPU CPU Time Time Consumed Consumed by by Month Month 31 31 E. A. Feigenbaum Feigenbaum 5P41 5P41 RR00785-11 RROOTSS-11 Progress -- Resource Resource Operations Operations Statistics Statistics Progress 60 50 ” Peak Daily Daily Jobs Jobs Peak. 4040 3030 ‘. 2020 1010 O~--~----~----~--~----~----~--~-----L----~--~~--~I I%74 1974 1975 1975 1976 1976 1977 1977 1978 1978 Figure 8: Figure 8: *r 8 7 7 1979 1979 1980 1980 1981 1981 1982 1982 1983 1983 1984 1984 1985 1985 Peak Number of Jobs by by Month Month Peak Number of Peak Peak. Daily Daily Load Load Average Average 66 55 44 33 22 11 tI OL-____ L -_ _ _ _L -_ _ _ _L -_ _ 1974 p974 1975 1975 1976 1976 1977 1977 ~~ Figure Figure 9:9: E. E. A. A. Fcigenb3.um Feigenbaum __ 1978 1978 ~~ __ 1979 1979 ~~ __ 1980 1980 ~~ ____ 1981 1981 Peak Peak Load Load Average Average by by Month Month 32 32 ~ __ 1982 1982 ~~ __ 1983 1983 ~~ __ 1984 1984 I 1985 1985 ~ Progress - Resource Operations Operations Statistics Statistics Progress RR00785-11 5P41 RR00785-11 Relative System Loading by Community Relative System Loading Community The SUMEX SUMEX resource is divided, divided, for for administrative administrative purposes, into into three three major major The projects based at the Stanford Stanford Medical Medical School (Stanford (Stanford Projects), communities: Projects), communit#ies: user projects project’s based outside outside of of Stanford Stanford (National (Nutiomzl Alhf common system system user projects AlA! Projects), Projects), and common development As defined development efforts efforts (System (System Staff). stajfi. defined in the resource management management plan plan approved approved by the BRP BRP at at the the start start of of the project, project, the available available system system CPU CPU capacity capacity and divided between between these communities communities as follows: follows: file space resources are divided Stanford Stanford AIM AIhI Staff Staff 40S; 40% 40S; 40% 20S; 20% The The "available" “available” resources to be divided divided up in this this way way are those remaining remaining after after various various monitor monitor and community-wide community-wide functions functions are accounted accounted for. for. These include include such things as as job scheduling, overhead, overhead, network network service, file space for for subsystems, subsystems, things job scheduling, documentation, documenlation. etc. The monthly monthly usage usage of of CPU CPU resources and terminal terminal connect connect time time for for each of of these The three communities communities relative relative to to their their respective respective aliquots aliquots is shown shown in the plots plots in Figure Figure three 10 and Figure Figure 11. As mentioned mentioned on page 30, these plots plots include include both both KilO KIlO and 2060 usage data. data. 33 E. A. Feigenbaum Feigenbaum 5P41 5P41 RRO0785-11 RR00785-11 Progress -- Resource Resource Operations Operations Statistics Statistics Progress 40 40- National Projects Projects National %CPUTime % CPU Time 30 30. 20 2010 10I%74 I .A, 1975 1976 1977 1978 1979 P974 1975 1976 1977 1978 1979 40 JOT % CPU CPU Used % Used 1980 1980 1981 1982 1983 1984 1985 4 1981 1982 1983 1984 1985 Stanford Projects Projects Stanford 30 30- 20 2010 IOI I o~----~----~----~----~----~----~----~----~----~----~----~ 1974 I%74 40 40 1975 1975 1977 1977 1976 1976 1978 1978 1979 1979 % 36 CPU CPU Used Used 1980 1980 1981 1981 1982 1982 1983 1983 19S4 1984 1985 1985 System System Staff Staff 30 20 20 10· 70 - I 1 _I b--L+ o.~----~--~~ p974 1975 1975 1974 __ ____ ____ ____ ____ _____ 1976 197G ~ 1977 1977 Figure 10: Figure 10: EE.. .-\. ‘-1. Feigenbaum Feigcnb:Lum ~ ~ ~ ~ 1978 1978 1979 1979 1980 19SO 1981 1981 ~ ____ 1932 1932 tvIonthly Monthly CPU CPU Usage Usage by Community Community 34 34 ~ ____ 1£83 1553 .~ __ 1984 19c& ~ 1935 Progress Progress -- Resource Resource Operations Operations Statistics Statistics 5P41 5P4 1 RR00785-11 RR00785- 11 6000 5000 National National Projects Projects Connect Connect Time Time Hours/Month 4000 4000 3000 3000 2000 2000 1000 7000 - ---f 1975 ,‘ 974 I 1976 1977 1978 1979 1980 1981 1983 1985 4 1982 O~--~----~----~----~----~----~----~--~~--~----~----~ 1975 1976 1977 1978 1979 1980 1981 1982 1983 .1984 1984 1985 1974 8000 Fizz 7000 6000 6000 5000 4000 3000 2000 1000 Connect Connect Time Time Hours/Month Hours/Month Stanford Stanford Projects Projects 1975 lo974 I 1976 1977 1978 1979 1980 I O~----~----~----~----~----~----~----~----~----~----~~--~ 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1985 7983 1984 1991 7982 1974 6000 6000 Connect Time Time Connect Hours/Month 5000 5000 - Hours/Month System Staff Staff System 4000 4000 3000 3000 2000 2000 1000 1000 O~----~----~--~~--~~--~----~----~----~----~----~----~I 1974 IS74 1975 1975 1976 1976 1977 1977 Figure 11: 11: Figure 1978 1978 1979 1979 1980 7980 1981 1981 1982 1982 1983 1983 1984 1984 Monthly Terminal Terminal Connect Connect Time Time by by Community Community Monthly 35 35 E. A. A. Feigenbaum Feigenbaum E. 1985 1985 Progress Progress -- Resource Resource Operations Operations Statistics Statistics 5P41 5P41 RR00785-11 RR00785-11 Individual Individual Project Project and and Community Community U8age Usage The The following following histogram histogram and and table table show show cumulative cumulative resource resource usage usage by by collaborative collaborative project and community during the past grant year. The histogram displays the project project project and community during the past grant year. The histogram displays the distribution of the total CPU time consumed between May 1, 1983 and April 3D, distribution of the total CPU time consumed between May 1, 1983 April 30, 1984, 1984, on the t,he SUMEX-AIM SUMEX-AIhl DECsystem2050 DECsyst.em2060 system. system. In entries include include aa text text summary summary of the funding funding sources In the the table table following, following, entries (outside (outside of of SUMEX-supplied SUMEX-supplied computing computing resources) resources) for for currently currently active active projects, projects, total total CPU CPU consumption consumption by by project project (Hours), (Hours), total total terminal terminal connect connect time time by project project (Hours), (Hours), and average file space space in use use by project project (Pages, 11 page page = 512 512 computer comput,er words). words). These data data were accumulated accumulated for each each project project for the months months between May, May, 1983 and May, May, 1984. 1984. Several of the projects projects admitted admitted to the National National AIM AIM community community use use the RutgersRutgersAIM AIM resource as as their their home base. base. We do not not explicitly explicitly list list these projects projects in this this annual annual report report covering covering the St.anford Stanford SUMEX-AIM SUMEX-AIM resource. We do record record information information about about the Rutgers Rutgers resource itself, itself, however, however, and note its separate resource status status with with the flag "[Rutgers-AIM]" . ” [Rut’gers-AIM] ‘I. E. A. A. Feigenbsurn Feigenbaum E. 35 36 Progress Progress -- Resource Resource Operations OperatGons Statistics Statistics 5P41 5P41 RR00785-11 RROOi85- 11 National National AIM AIM (24.4% (24.4% Total) Total) AIM AIM Administration Administration AIM AIM Pilots Pilots AIM AIM Users Users ACT ACT I I Caduceus Caduceus SEeS SECS 1 I Models Models of of Human Human Cog Cog Solver Solver Puff-VM Puff-VM Rutgers Rutgers 1 P Stanford Stanford (59.6% (59.6% Total) Total) AGE AGE AI Handbook Al Handbook DENDRAL DENDRAL EXPEX EXPEX Guidon Guidon Hpp Research Hpp Research HPP Assoc HPP Assoc Med Med Info Info Sci Sci MOLGEN MOLGEN Oncocin Oncocin Protein Structure Protein Structure RX Stanford Pilots Pilots Stanford Stanford Assoc Assoc Stanford ~ ..... ~- P Staff (16.0% (16.0% Total) Total) Staff Staff Staff Staff Assoc Assoc Staff 0o Figure 12: 12: Figure p P I I 55 10 10 15 20 25 25 15 20 Percent of of Total Total CPU CPU Used Used Percent Cumulative CPU CPU Usage Histogram Histogram by by Project Project and and Community Community Cumulative 37 37 E. A. A. Feigenbaum Feigenbaum E. 5P41 5P41 RR00785-11 RR00785-11 Progress -- Resource Resource Operations Operations Statistics Statistics Progress Resource Use Use by by Individual Individual Project Project -- 5/83 5/83 through through 4/84 4/84 Resource National A1A[ Community Community i\‘ ational .4IM Connect Connect (Hours) (Hours) 0.37 0.37 33.88 33.88 2866 2866 58.15 58.15 895.52 6852 6852 1.38 1.38 209.34 750 750 ACT Project Project 1) ACT "Acquisition of of “Acquisition Cognitive Procedures” Procedures" Cognitive John R. Anderson, Anderson, Ph.D. Ph.D. John Carnegie-Mellon Univ. Univ. Carnegie-Mellon NSF IST-80IST-80-15357 NSF 15357 2/81-2/84 2/81-Z/84 File File Space (**) (**) (Pages) (Pages) CPU CPU (Hours) (Hours) $186,000 $186,000 CADUCEUS 2) CADUCEUS "Clinical Decision Decision Systems Systems “Clinical Resea.rch Resource" Resesrch Resource” Jack D. Myers, Myers, M.D. M.D. Jack Harry E. Pople, Pople, Jr., Jr., Ph.D. Ph.D. Harry University of of Pittsburgh Pit,tsburgh Universit,y NIH RR-01 r·m-Ol101-07 NIH 101-07 $1,507,717 7/80-5/8S $1.607,717 7/80-6/&5 $369,484 7/83-6/84 $369,484 7,i83-f/84 NLM Lh1037 Lr-.W371O-04 NLM IO-04 $817,884 7/80-5/8S 7/80-S/85 $817,884 $196,710 7/83-5/84 7/83-6/84 $196,710 NLt\l New New Invest NLhI Invest L~103889-02 Lh¶03889-02 Gordon Gordon E. Banks. Banks. M.D. M.D. $107,675 4/82-3/8.5 4/‘82-3jS.S $107,675 $35,975 4/83-3/84 4/83-3,/84 $35,975 $35,975 4/84-3/85 4/84-3185 $35.975 3) CLIPR CLIPR Project Project "Hierarchical “Hierarchical Models Models of Human Hllman Cognition" Cognition” \Valter \Valter Kint.sch. Iiint.sch. Ph.D. Ph.D. Peter Pet.er G. Polson, Poison, Ph.D. Ph.D. University Llniversity of Colorado Colorado NI!\'1H NIXtH MH-15872-14-15 PIIH-1.5872-14-16 (Kintsch) (Kintsch) 7/81-5/8'1 7/,81-6,/&l $281,085 77;83-6/8-1 / 8~~-o/8·t $59,878 $60,878 NSF NSF (Kintsch) (E;int.sch) $200,000 8/8a-7/85 8,‘83-7/‘86 $200,000 mr-.l (Polson) IBhi (Poison) Da.vid David Kieras Kieras lTniversity Universit.y of of Arizona Arizona $364,000 1/82-12/84 l/82-12/8.4 $364.000 $145,000 1/84-12/84 l/84-12/84 $145,000 E. E. A. A. Feigenbaum Feigenbaum 38 38 Progress - Resource Operations Operations Statistics Stat.istics Progress SP41 5P41 RR0078S-11 RROO785-11 0.65 61.20 303 264.61 9877.34 10500 S.76 356.23 5.76 “Problem Solving Solving "Problem Expertise” Expertise" Paul E. Johnson, Ph.D. Ph.D. P::J.1l1 v.,'illiam William B. Thompson, Thompson, Ph.D. Ph.D. Control Control Data Dat.a Corp. Corp. (Johnson) (Johnson) $90,000 1983-85 $90,000 Microelect. Microelect. and a.nd Info. Info. Ctr. Ctr. Univ. Univ. of of rvlN MN (Plus (Plus Two Two Colleagues) Colleagues) 1984-1987 $800,000 $800,000 H18'i-1987 492 4) PUFF-VM PUFF-VM Project Project “Biomedical Knowledge Knowledge "Biomedical Engineering Engineering in Clinical Clinica. Medicine" Medicine” John J. Osborn, Osborn, M.D. M.D. John Med. Research Inst., Inst., Med. Francisco San Francisco Edward Edward H. ShortIiffe, Shortliffe, M.D.,Ph.D. M.D.,Ph.D. Stanford University University Stanford Johnson & k Johnson Johnson Johnson 1 year $SO,OOO $50,000 (*) (*) year 5) SECS Project Project "SiI1llllation “Simulation & Sr Evaluation Evaluation of Chemical Chemical Synt.hesis" Synthesis” of W. W. Todd Todd Wipke, Wipke, Ph.D. Ph.D. U. 1J. California, California., Santa Santa Cruz Cruz NIHEHS ES02845-02 NIHEHS ESO~845-02 4/82-3/85 4/82-3/8,5 $257.801 4j84-3i85 $89,140 4/84-3/85 $89,140 Evsns & & SlItherland Slltherland Corp. Corp. Evans Equipment Equipment gift gift Value $9S,000 Value $95,000 Stauffer Sta,uffer Chemical Chemical Co. $5,000 $6,000 5) 6) SOLVER SOLVER Project Project 39 E. A. A. Feigenbaum Feigenbaum Operations Statistics Statistics Progress - Resource Operations 5P41 5P41 RR00785-11 RR00785-11 *** [Rutgers-AIM] [Rutgers-AIM] *** *** 7) *** Rutgers Research Resource Rutgers “Computers in Biomedicine" Biomedicine” "Computers Saul Amarel, Amarel, D.Sc. Ca.simir Kulikowski, Kulikowski, Ph.D. Ph.D. Casimir Shalom vVeiss, Weiss, Ph.D Ph.D Shalom Rutgers U U.... New Brunswick Brunswick Rutgers NIH RR-00543-12 RR-00643-12 (Amarel, Kulikowski) Kulikowski) NIH (Amarel, 12/82-11/83 $405,304 12/82-11/83 $405,304 (Kulikowski, Weiss) Weiss) NIH RR-02230-01 RR-02230-01 NIH (Kulikowski, 0.52 0.52 38.59 38.59 1117 1117 65.85 55.85 2227.48 2227.48 2461 2451 .93 .93 118.75 118.75 686 585 57.36 57.36 3836.19 3836.19 9649 9649 ----------------- ------------------- ----m-w 455.56 455.56 17654.52 17554.52 35676 35576 12/83-11/87 $3,198,075 12/83-11/87 $3,198,075 12/83-l l/84 $989,275 $989,276 12/83-11/84 AIM Pilot Pilot Projects Projects 8) AIM 9) AIM AIM Administration Administration 10) AIM AIM Users 10) Community Totals Totals Community E. ‘A. 2. Feigenbaum Feigenbaum 40 40 Progress - Resource Operations Operations Statistics Statistics SP41 5P41 RR0078S-11 RR00785-11 CPU CPU (Hours) (Hours) Connect Connect (Hours) (Hours) File File Space Space (Pages) (Paws) 1) AGE AGE Project. Project (Core) (Core) "Attempt “Attempt to Generalize" Generalize” Edward Ph.D. Edward A. Feigenbaum, Feigenbaum, Ph.D. Dept. Dept. Computer Computer Science Science ARPA (U*) ARPA MDA903-8Q-C-0107 MDA903-80-C-0107 (***) (partial (partial support) support) 11.80 11.80 845.30 845.30 4076 4076 2) AI AI Handbook Ha.ndbook Project Project (Core) (Core) Edward A. Feigenbaum, Ph.D. Edward Ph.D. Dept. Dept. Computer Computer Science Science ARPA (**) hID,4903-80-C-0107 (**) ARPA4 MDA903-8Q-C-0107 (partial (partial support) support) 1l.03 11.03 980.94 980.94 4425 4425 3.72 3.72 183.81 183.81 2980 2980 4) EXP EX Proj ect EXPEX Project "Expert “Expert Explanation" Explanation” Edward Edward H. Shortliffe, Shortliffe, M.D.,Ph.D. hl.D.,Ph.D. Dept. Dept. Medicine Medicine ONR NR 049-479 ONR NR 049-479 $-156,622 1/81-12/8:3 l/81-12,,‘8:3 $456,622 ONR NR049-479 ONR NR049-479 Michael hlichael Gt'nesereth Genesereth 1/84-12/85 $312.070 l/84-12/‘86 NSF IST83-12H8 IST83-12148 Bruce G. Buchanan Buchanan $330,000 3/84-2/87 3/‘84-3,‘87 $330.000 (*) (*) $99,410 3/84-2/8.5 3 /8+2;‘8Fi $99,410 (*) (“) 53.75 53.75 239l.40 2391.40 4920 4920 5) GUIDON-NEOMYCIN GUIDON-NEO~lYCIN Project Project "Exploration “Exploration of Tutoring Tutoring S: 6 Problem-solving Problem-solving Strategies" Strategies” Bruce G. Buchanan, Buchanan, Ph.D. Ph.D. William William J. Clancey, Clancey, Ph.D. Ph.D. Dept. Dept. Computer Computer Science Science ONR/ NOOOI4-79-C-0302 ONR/ARIARI N00014-79-C-0302 $583,892 3/79-3/85 3,‘79-3/85 $683,892 45.44 45.44 4418.68 4418.68 5967 5967 Stanford Stan ford Community Community 3) DENDRAL DENDRAL Project Project "Resource “Resource Related Related Research: Research: Computers Computers in Chemistry" Chemistry” Carl Carl Djerassi, Djerassi, Ph.D. Ph.D. Dennis H. Smith, Smith, Ph.D. Ph.D. Dept. Dept. Chemistry Chemistry NIH RR-00612-13 NIH RR-00612-13 $170,710 5/82-4/83 5/82-d/83 $170,710 41 41 E. A. Feigenbaum Progress Progress - Resource Operations Operations Statistics Statistics 5P41 5P41 RR00785-11 RR00785- 11 6) Project 6) fv[OLGEN kIOLGEN Project "Applications “Applications of of Artificial Artificial Intelligence Intelligence 106.92 106.92 7734.34 7734.34 10448 10448 239.97 14404.62 14389 4.79 635.43 635.43 1296 1296 t.o t,o Molecular Molecular Biology" Biology” Edward Edward A. Feigenbaum, Feigenbaum, Ph.D. Ph.D. Peter Peter Friedland, Friedland, Ph.D. Ph.D. Charles Charles Yanofsky, Yanofsky, Ph.D. Ph.D. Depts. Depts. Computer Computer Science/ Science/ Biology Biology NSF MCS-8310236 NSF MCS-8310236 (Feigenbaum, (Feigenbaum, Yanofsky) Ya.nofsky) 11/83-10/84 $139,215 (*) 11/83-lo/84 $139,315 (*) 7) ONCOCIN ONCOCIN Project Project "Knowledge “Knowledge Engineering Engineering for for Med. Med. Consult.ation Consultation” II EdWard Edwa.rd H. Shortliffe, Shortliffe, M.D.,Ph.D. M.D.,Ph.D. Dept. ~[edicine Dept. hfedicine NLM LM-03395 (Shortliffe/ONCOCIN) NLhI LM-03395 (Shortliffe/ONCOCIN) Edward A. Feigenbaum, Edward A. Feigenbaum, Ph.D. Ph.D. 7/79-5/84 7/79-6184 $497,420 $497,420 7/83-5/84 $95,424 7,/83-6/84 $95,424 NLM LM-00048 NLhf LbI-00048 $196,425 7/79-5/84 7/79-6/84 $196,425 7/83-5/84 $39,502 7,‘83-6184 $39,502 ONR NR 049-479 ONR NR $456,522 (*) 1/81-12/83 l/81-12/83 $456,622 (*) NIH RR-01513 NIH RR-01613 $524,455 7/83-6/85 7!83-6186 $624,455 $220,371 7/83-6/84 7/83-6/84 $220,371 NLM L~1-04136 NLM Lhl-04136 $211,851 8/83-7/85 8/83-7/86 $211,851 $50.517 8/83-7/84 8/83-7/84 $60,517 H.J. H.J. Kaiser Kaiser Family Family Fdn. Fdn. $150,000 7/83-6/86 7!‘83-6/86 $150,000 $50,000 7/83-6/84 7;83-6,/84 $50,000 ONR NOOO14-81-K-0004 ONR N00014-81-K-0004 Michael Michael R R. Genesereth Genesereth (Shortliffe) (Shortiiffe) 1/84-12/86 $512.070 l/84-12/86 $512,070 (*) (*) NSF IST83-12148 NSF IST83-12148 Bruce Bruce G. Buchanan Buchanan (Shortliffe) (Shortliffe) 3/84-2/87 $330.000 (*) 3 /84-2/‘87 $330.000 (*) $99,410 3/84-2/85 3/83-2/85 $99,410 (*) (*) 8) PROTEIN PROTEIN Project Project "Heuristic “Heuristic Compo Comp. Applied Applied to to Prot. Prot. Crystallog. Crystallog.” II Edward Edward A. A. Feigenbaum. Feigenbaum, Ph.D. Ph.D. Dept. Dept. Computer Computer Science NSF MCS-81-17330 NSF MCS-81-17330 $28,976 1/82-1/83 l/82-1/83 $28,976 E. A. Feigenbaum Feigenbaum 42 42 Progress - Resource Operations Operations Statistics Statistics 5P411 RR00785RR00785-11 5P4 11 79.44 RADIX Project Project 9) RADIX "Deriving Medical Medical Knowledge Knowledge from from TimeTime“Deriving Oriented Clinical Clinical Databases” Databases" Oriented Robert L. Blum. Blum. M M.D. Robert .D. Gio C.hl. C.I,I. VViederhold, \Viederhold, Ph.D. Ph.D. Gio Depts. Computer Computer Science/ Science/ Depts. Electrical Engrg. Engrg. Electrical IST-8317858 (Blum) (Blum) NSF IST-8317858 (*) 3/84-3186 3/84-3/86 $89597 (*) NLM (Wiederhold) (Wiederhold) NLM $291,192 S/84-llj86 S/84-11/86 $291,192 3140.27 3140.27 8777 10) Stanford Stanford Pilot Pilot Projects Projects 61.55 61.55 4115.02 4115.02 6097 6097 11) HPP Core Core AI AI Research Research 11) HPP 383.07 383.07 29073.96 29073.96 42202 42202 12) HPP Associates 12) HPP 57.37 57.37 1600.31 1600.31 2997 2997 13) Stanford Stanford Associates 27.01 27.01 1016.59 1016.59 1681 1681 5.62 5.62 1315.64 1315.64 587 587 --------- ---------- _------ 1091.46 1091.46 71856.29 71856.29 110842 110842 CPU CPU (Hours) (Hours) Connect Connect (Hours) (Hours) Space File Space (Pages) 288.21 288.21 16.65 16.65 17591.82 17591.82 1983.43 1983.43 23292 23292 7847 7847 --------- ------------------- _------ 304.87 304.87 19575.25 19575.25 31139 31139 CPU CPU (Hours) (Hours) Connect Connect (Hours) (Hours) File Space Space File (Pages) (Pages) 530.54 530.54 67375.43 67375.43 167863 167863 ----===== ----===== ----===== 2382.43 2382.43 176461.50 176461.50 345520 345520 14) Medical Information Information Sciences Sciences 1-i) Medical Community Community Totals Totals SUMEY Stf'!! St a f f SUMEX 1) Staff Staff 2) 2) System Associates Associates Community Community Totals Totals System System Operations Operations 1) 1) Operations Operations Resource Resource Totals Totals (*) (*) Award Award includes indirect indirect costs. costs. (**) awarded to (**) Supported Supported by aa larger larger ARPA ARPA contract contract MDA-903-80-C-0107 MDA-903-80-C-0107 awarded to the Stanford Stanford Computer Computer Science Science Department: Department: 43 43 E. A. Feigenbaum Progress Progress - Resource Operations Operations Statistics Statistics 5P41 5P41 RR00785-11 RR00785- 11 System Reliability System Reliability System better than System reliability reliabilit,y for for the DECsystem DECsystem 2060 has been much much better than with with our our previous KilO periods of particular hardware previous KIlO system. system. "Ve We have had very very few periods of particular hardware or or software software problems. The below covers the period in which problems. The data data below the entire entire period which the SUMEX-AIM SUMEX-AIM community community has used the 2050. 2060. The The actual actual downtime downtime was rounded rounded to to the nearest nearest hour. hour. 7 18 1 1 Feb Mar Apr Table per Month Table 1 : System System Downtime Downtime Hours Hours per Month - February February 83 through through Apr Apr 83 11 11 1 2 6 o0 11 11 15 15 26 26 13 13 16 16 28 28 May hlay Jun Jul Jul Aug Au Sep Sep Oct Ott Nov Nov Dec Dee Jan Feb Mar Mar Apr -Or Table per Month Ta.ble 2 : System System Downtime Downtime Hours Hours per Month - May May 83 through through Apr Apr 84 Reporting period Reporting period Total UpTime Total Up Time Pl\I Phi Downtime Downtitne Act.ual Act,ual Downtime Downtime Tot,al Tor.al Downtime Downtime Mtbf Mtbf Uptime Upt,ime Percentage Percentage 462 days, 23 hours, hours, 41 minutes, minutes, and 42 seconds 454 days, 5 hours, hours, 16 minutes, minutes, and 57 seconds 1 days, 14 hours, hours, 2 minutes, minutes, and 55 seconds 7 days, 4 hours, hours, 21 21 minutes, minutes, and 50 seconds 8 days, 18 hours, hours, 24 minutes, minutes, and 45 seconds 2 days, 16 hours, hours, 30 minutes, minutes, and 16 seconds 98.45 Network Usage Statistics Network Statistics The plots in Figure The plots Figure 13 and Figure Figure 14 show the monthly monthly network network terminal terminal connect connect tim(' broader term time for for the TYMNET TYMNET and the INTERNET INTERNET usage. The The INTERNET INTERNET is a broader term for for what previously referred usage. Since many what was previously referred to as as Arpanet Arpanet usage. many vendors vendors now now support support the INTERNET protocols (IP JTCP) in addition JTCP INTERNET protocols (IP/TCP) addition to to the Arpanet, Arpanet, which which converted converted to to IP IP/TCP in Janllary 198~3, it between Arpanet Janua.ry of of 1083, it is no longer longer possible to to distinguish distinguish between Arpanet usage and Internet Internet usage usage on our our 2060 system. system. E. A. A. Feigenbaum Feigenbaum 44 44 Progress Progress- -Resource Resource Operations OperationsStatistics Statistics SP'll RR00785-11 5P4 1 RR0078511 1400 TYMNETConnect ConnectTime Time TYMNET Hours/Month Hours/Month ‘400 1200 7200 10001000 800 800 600 600 400 400 200 zoo- jr 4 OL---~----~----~----~----~----~----~--~~--~----~----~ I?374 I 1975 1974 1975 1976 1976 1977 1977 1978 1978 1979 1979 1980 1980 1981 1981 1982 1982 7983 1983 1984 1984 7985 1985 Figure 13: 13: Figure TYMNET TYMNET Terminal Terminal Connect Connect Time Time 1200 ARPAnet Connect Connect Time Time r ARPAnet Hours/Month Hours/Month 1000 1000 ‘200 1 800 800 600 600 400 400 t 200 200 t , O~----~----~----~----~----~----~----~----~----~----~~--~ 1975 1976 1977 7978 1979 1980 1981 ! 982 1983 1984 1985 It374 I 1974 1975 1976 1977 Figure Figure 14: 14: 1978 1979 1980 1981 1982 1983 1984 ARP ANET Termina.l ARPANET Terminal Connect Connect Time Time 45 45 E. A. A. Feigenbaum Feigenbaum E. 1985 Progress Progress -- SUMEX SUMEX Staff Staff Publications Publications 5P41 5P41 RR00785-11 RR00785- 11 I.A.2.8. SUMEX Staff I.A.2.8. SUMEX Staff Publications Publications The The following following are are publications publications for for the the SUMEX SUMEX staff staff and and include include papers papers describing describing t,he the SUMEX-AIM SURIEX-AIM resource and and on-going on-going research as as well as as documentation documentation of of system system and and program EX-AIM community program developments. developments. Many Many of of the publications publications documenting documenting SUM SUMEX-AIM community research are are from from the individual individual collaborative collaborative projects projects and and are are detailed detailed in in their their respective respective reports (see Section II on page 69). Publications for the AGE and AI Handbook reports (see Section II on page 69). Publications for AGE and AI Handbook core core research resesrch projects projects are are given given there. 1. 1. Carhart, Carhsrt, R.E., R.E., Johnson, S.M., S.M., Smith, Smith, D.H., D.H., Buchanan, Buchanan, B.G., B.G., Dromey, Dromey, R.G., R.G., and Lederberg, A Lederberg, J., Networking Networking and and aa Collaborative Collaborative Research Research Community: Community: A Programs. IN Cllse Case Study Study Using Using the DENDRAL DENDRAL Programs. IN P. Lykos Lykos (Ed.), (Ed.), CO~1PUTER AND COMPUTER NETWORKING NETWORKING AND CHEMISTRY, CHEMISTRY, ACS ACS Symposium Symposium Series, Series, No. 19, 1975. 19, 1975. 2. 2. Levinthal, Levinthal, E.C., E.C., Carhart, Carhart, R.E., R.E., Johnson, S.M., S.M., and Lederberg, Lederberg, J.: Computer,'! Computers Talk Talk to Computers. Computers. Industrial Industrial Research, November, November, 1975. 1975. When 3. Wilcox, - A Machine-Independent Programming Wilcox, C.R., C.R., MAINSAIL MAINSAIL Machine-Independent Programming System. Syatem. Proc. Proc. DECUS DECUS Symposium Symposium 2(4), Z(4), Spring, Spring, 1976. 1976. 4. Wilcox, Project: for Software C.R.: The AL4INSAIL MAINSAIL Project: Developing Wilcox, C.R.: Developing Tools Tools for Software Portability. Portability. Proc. Proc. SCAMC, SCAMC, October, October, 1977, pp. 76-83. 5. Lederberg, and Lederberg, J.L.: J.L.: Digital Digital Communications Communication8 and the Conduct Conduct of of Science: Science: The New New Literacy. Literacy. Proc. Proc. IEEE IEEE 66(11), November, November, 1978. 1978. 6. Wilcox, MAINSAIL - Language Language Wilcox, C.R., C.R., Jirak, Jirak, G.A., G.A., and Dageforde, Dageforde, M.L.: M.L.: MAINSAIL Afanual. Stanford Manual. Stanford University University Computer Computer Science Report Report STAN-CS-80-791, STAN-CS-80-791, 1980. 1380. C.R., Jirak, G.A., and Dageforde, M.L.: AIAINSAIL C.R., Dageforde, M.L.: Jirak, G.A., MAINSAIL - Implementation Implementation Overview. Overview. Stanford Stanford University University Computer Computer Science Report Report STAN-CS-80-792, 1980. STAN-CS-80-793, 7. \Vilcox, \Vilcox, addition, a substantial substantial continuing continuing effort effort has gone into into developing, developing, upgrading, upgrading, In addition, extending documentation documentation about about the the SUMEX-AIM SUMEX-AIM resource. resource. These efforts efforts include include user and extending help files, and introductory introductory notes, an ARPANET ARPANET Resource Handbook Handbook entry, entry, and and guides, help policy guidelines. guidelines. policy E. A. A. Feigenbaum Feigenbaum E. 46 46 Progress Progress - Future Future Plans Plans 5P41 RR00785-11 RR00785-11 I.A.2.9. Future Plans I.A.2.9. Future Plans Our plans for plans for Our plans for the next next grant grant year year are based on the Council-approved Council-approved plans for our our 5-year plans for 5-year renewal renewal that that began in August, August, 1980. In In addition addition to the specific specific plans for the the next next grant period to serve grant year, year, we include include a summary summary of of the overall overall objectives objectives for for this this 5-year 5-year period as background. NearNear- and long-term plans for as a background. long-term objectives objectives and plans for individual individual collaborative collaborative projects are discussed in Section beginning on page 69. projects Section II II beginning Ot.erall Overall Goals Goals The The goals of of the SUMEX-AIM SUMEX-AIM resource are long-term long-term in supporting supporting basic research in artificial broad range of biomedical applying these techniques techniques to to a broad of biomedical artificial intelligence, intelligence, applying problems, experimenting with technologies promote scientific with communication communication technologies to to promote scientific experimenting problems, interchange, better tools interchange, and developing developing better tools and facilities facilities to to carryon carry on this this research. research. Just Just as as the tone of proposal derives of our our renewal renewal proposal derives from from the the continuing continuing long-term long-term research research objectives objectives of philosophy of the SU~fEX-AIM SUMEX-AIM community, community, our our approach approach derives derives from from the methods methods and philosophy already We will will continue continue to to develop useful useful knowledge-based knowledge-based already established established for for the resource. We software biomedical research based on innovative, softwa,re t,ools t,ools for for biomedical innovative, yet yet accessible computing computing t.echnologies. t,echnologies. For For us it it is important important to to make systems systems that that work work and are exportable. exportable. Hence, our our approach is to integrate available state-of-the-art hardware technology approach to integrate available state-of-the-art hardware technology as a basis for the the underlying software research and development development necessary to support support the AI AI work. work. underlying software SUMEX-AIM SUMEX-AIM will will retain retain its broad broad community community orientation orientation in choosing choosing and implementing implementing its its resources. \Ve will will draw draw upon upon the expertise expertise of of on-going on-going research efforts efforts where where possible and This orientation orientation has a.nd build build on these where where extensions extensions or or innovations innovations are necessary. This proved to be an effective proved effective way way to build build the current current facility facility and community. community. We We have built built ties to a broad broad computer computer science community; community; have brought brought the the rE'sults results of of their t.heir work work to to the AIM AIM users; users: and have exported exported results results of of our our own own work. work. This This broader particularly active broader community community is particularly active in developing developing technological technological tools tools in the form form of of new machine machine architectures, architectures, language support, support, and interactive interactive modalities. modalit.ies. Toward Distributed Resource Resource Toward a Afore More Distributed The The initial initial model model for for SUMEX SUhLIEX as as a centralized centralized resource was based on the high high cost of powerful computing of powerful computing facilities, fa.cilit,ies, which which were not not readily readily duplicated. duplicated. This This role is evolving, evolving, though, though, with with the t.he introduction introduction of of more more compact compact and inexpensive inexpensive computing computing technology. technology. Our by community Olur fnture future goals are guided guided by community needs for for more more computing computing capacity capacity and improved build more improved tools tools to build more effective effective expert expert systems, and to to test test operational operational versions versions of of AI programs in real-world In order order to to meet meet these needs, we must must take take advantage advantage AI programs real-world settings. settings. In of of a ralll,;e range of of newly-developing newly-developing machine machine architectures architectures and systems. As a result, result, SUIvIEXSUhsIEXAIM AIM will will become a more more distributed distributed community community resource with with heterogeneous heterogeneous computing computing facilit,ies fncilit.ies tE'thE'rE'd tethered to t,o each other other through through communications communications media. media. Many Many of of these machines machines will physically near the projects projects or will be located located physically or biomedical biomedical scientists scientists using using them. them. The C{Jntinuing Role of Continuing Role of SUAfEX-Central SUMEX-Central Even Even with with more more distributed distribut.ed computing computing resources, the central central resource will will continue continue to play play an important important role as as a communications communications crossroad, crossroad, as as a research research group group devoted devoted to integrating integrating the new software software and hardware hardware technologies technologies to to meet meet the needs of of medical medical AI ,41 projects, and as applicat.ions, applications, as a spawning spawning ground ground for for new application application projects, as a base base for for local AI AI projects. A projectas. A key key challenge challenge will will be to to maintain maintain the scientific scientific community community ties that that grew grew naturally previous colocation naturally out. out of of the t,he previous colocation within within a central central facility. facility. 47 E. A. Feigenbaum Feigenbaum 5P41 RR00785-11 RR00785- 11 Progress - Future Future Plans Summary Summary of of Five-!lear Five-year Objectives Objectives The long-term long-term objectives objectives of the SUMEX-AIM SUMEX-AIM resource nucleus during during the follow-on follow-on 5 year period (of which we are in the third year) are summarized year period which third year) summarized below. These are broken resource operations, operations, training training and and education, education, and core broken into into three three categories: categories: resource research. research. Resource Resource Operations Operations l. will continue continue to encourage 1. Maintain Ma.inta.in the vitality vitality of the AIM AIM community community -- We will and explore explore new applications applications of AI AI to biomedical biomedical research and improve improve mechanisms mechanisms for interinter- and intra-group intra-group collaborations collaborations and communications. communications. While While AI AI is is our our defining defining theme, theme, we may may entertain entertain exceptional exceptional applications applications justified justified by some other other unique unique feature feature of SUMEX-AIM SUMEX-AIM essential for for important important biomedical biomedical research. To To minimize minimize administrative administrative barriers barriers to the communitycommunityoriented oriented goals of SUMEX-AIM SUMEX-AIM and to direct direct our our resources toward toward purely purely scientific scientific goals, we plan plan to retain retain the current current user funding funding arrangements arrangements for for projects User projects projects working working on SUMEX SUMEX facilities. facilities. projects will will fund fund their their own own manpower manpower and local needs; will will actively actively contribute contribute their their special expertise expertise to the SUMEX-AIM SUMEX-AIM community; community; and will will receive an allocation allocation of of computing computing resources under under the control control of of the the AIM AIM management mana.gement committees. committees. There There will will be no "fee “fee for for service" service” charges for for community community members. members. We also will will continue continue to exploit exploit community community expertise expertise and sharing sharing in software software development, development, and to facilitate facilitate more effective effective information-sharing information-sharing among among projects. projects. 2. Provide Provide effective effective computational computational support support for for AIM AIM community community goals -- We We will will continue intelligence continue to expand expand support support for for artificial artificial intelligence research and new applications applicat.ions work, work, to develop develop new computational computational tools tools to support support more more mature mature proj ects, and to facilitate projects, facilitate testing testing and research dissemination dissemination of of nearly nearly operational programs. We operational programs. We will will continue continue to to operate operate and develop develop the existing existing central cent,ral facility facility as as the nucleus nucleus of of the resource. We We will will acquire acquire additional additional 11ipment to meet eq equipment meet developing developing community community needs for for more more capacit,y, capacity, larger larger program address New computing program a.ddress spaces, spaces, and improved improved interactive interactive facilities. facilities. New computing hardware becoming available hardware technologies t,echnologies becoming available now now and in the next next few years years will will play a key play key role in these developments, developments, and we expect expect to take take the lead in this this community biomedical AI community for for adapting ada.pting these new tools tools to biomedical AI needs. Provide effect.ive effective and geographically geographically accessible communication communication facilities facilities to to the the 3. Provide SUMEX-AIM commiinit,y community for for effective effective remote remote collaborations, collaborations, communications communications SUMEX-AIM among dist,ribut,ed distributed computing computing nodes, and experimental experimental testing testing of of AI AI programs programs among We will will retain retain the the current current ARPANET ARPANET and TYMNET TYMNET connections connections for for at at -- We the near-term near-term and and will will actively actively explore explore other other advantageous advantageous connections COIJ.nections to to least the communications networks networks and to to dedicated dedicated links. links. new communications Training and Education Education Training and l. Assist Assist new and and established established projects projects in in the the effective effective use of of the the SlJMEX-AIM SUMEX-AIM 1. resource Collaborative projects continue to be responsible for the the resource -- Collaborative projects continue to responsible for development and and dissemination dissemination of of their their own own AI AI programs, programs, but but the the resource resource development staff will will provide provide general general support support and and will will work work to to make make resource resource goals and and AI AI staff systems known known and and available available to to appropriate appropriate biomedical biomedical scientists. scientists. We We will will systems continue to to exploit exploit particular particular areas of of expertise expertise within within the the community community for for continue deve loping pilot pilot efforts efforts in new application application areas. developing E. A. A. Feigenbaum Feigenbaum 48 Progress Progress - Future Future Plans Plans 5P41 RR00785-11 RR00785-11 2. funds to to qualifying qualifying new and pilot pilot 9. Continue Continue to to allocate allocate "collaborative “collaborative linkage" linkage” funds projects projects to to provide provide for for communications communications and terminal terminal support support pending pending formal formal approval approval and funding funding of 01 their their projects projects -- These funds funds are allocated allocated in cooperation prospective user cooperation with with the AIM AIM Executive Executive Committee Committee reviews reviews of of prospective projects. projects. 3. Continue Cont.inue to support support workshop workshop activities activities including including collaboration collaboration with with the the Rutgers Rutgers Computers Computers in Biomedicine Biomedicine resource on the AIM AIM Community Community \Vorkshop Workshop and with with individual individual projects projects for for more specialized specialized workshops workshops covering covering specific specific application application areas or or program program dissemination dissemination Core Research Core Research 1. 1. Continue Continue to to explore explore basic Artificial Artificial Intelligence Intelligence research issues issues for for knowledge knowledge acquisition. reasoning representation, and utilization; utilization; acquisition. representation, reasoning in the presence of of uncertainty; uncertainty; strategy strategy planning; planning; and explanations explanations of of reasoning reasoning pathways pathways with with particular emphasis on biomedical particular biomedical applications applications -- SUMEX SUMEX core research funding to similar funding is complementary complementary similar funding funding from from other other agencies and contributes contributes to to the the long-standing long-standing interdisciplinary interdisciplinary effort effort at at Stanford Stanford in basic AI AI research and expert We expect expect this this work work to to provide provide the expert system system design. We foundation foundation for for increasingly increasingly effective effective consultative consultative programs programs in medicine medicine and for for more practical adaptations of this work within emerging more practical adaptations 01 this work within emerging microelectronic microelectronic technologies. technologies. 2. Support Support community community efforts efforts to to organize organize and generalize generalize AI AI tools tools that that have been developed developed in the context context of of individual individual application application projects projects -- This This will will include include work work t,o to organize organize the present present state-of-the-art state-of-the-art in AI AI techniques techniques through through the development development of of practical practical software software packages for for the acquisition. acquisition, representation. representation, and utilization utiiization of of knowledge knowledge in AI AI programs. programs. The The objective objective is to evolve a body body of of software software tools tools that that can be used to to more more easily build build future future knowledge-based knowledge-based systems systems and explore explore other other bionmedical bionmedical AI AI applications. applications. 49 E. A. Feigenbaum Feigenbaum 5P41 5P41 RR00785-11 RR00785-11 Progress Progress -- Future Future Plans Plans Specific Specific Plans Plan8 for for Year Year 12 12 Specific Specific plans plans for for the the next next grant grant year year (12) (12) are are summarized summarized in in the the paragraphs paragraphs below. below. The The directions directions and and background background for for much much of of this this work work were were given given in in earlier earlier progress progress report report sections sections and and are are not repeated in in detail detail here. Professional Workstations fiofessional WOrk8tatiOn8 We see see our our major major development development efforts efforts in in year year 12 12 to be be in in the area area of of professional professional workstat.ions, workstat,ions, and and specifically, specifically, to fine fine tune the integration integration of of these these workstations workstations into into our our networking networking environment. environment. This This involves involves software software integration, integration, support support of network network protocols, protocols, general access access to network network printing printing facilities, facilities, telnet telnet access access to Lisp ma.chines, ma.chines, and overall overall worksta.tion workststion ma.intenance maintenance and support. support. We will will also continue continue to explore the use use of low cost workstations workstations within wit’hin our our environment, environment, both both as as distributed distributed processors for for text text editing editing and electronic electronic mail, mail, and as as powerful powerful graphic graphic terminals terminals for for use use with with sophisticated sophisticated programs programs running running on our our mainframes. mainframes. We also see see the use use of virtual virtual graphics graphics interfaces interfaces running running on remote remote workstations to be of continued importance to our progress in the future. workstations continued importance our future. Continued Continued Operation Operation of of Existing Existing Hardware Hardware The The current current, SUMEX-AIM SUMEX-AIM facilities facilities represent represent a large existing existing investment. investment. 'vVe We plan plan to continue continue development deve!opment of of our our main main timesharing timesharing machine, machine, the t,he DEC2050/TOPS-20 DEC2060/TOPS-20 syst.ern, syst.em, and the SUMEX-AIM SUMEX-AIM file server server (SAFE), (SAFE), and make changes as as necessary to improve improve the performance performance of of these machines. machines. We We do not not propose any any substantial substantial changes to the ot.her ot,her hardware hardware systems (2020, shared VAX, VAX, and Lisp Machines). Machines). We We expect expect them them to provide effective t.o continne cont,inue to provide effective community community support support and serve as as a nucleus for for our our distribut,ed di,strihut.ed resource. Communication Networks Communication Network8 Networks have been centrally Net,works centrally important important to the research goals of of SUMEX-AIM SUhlEX-AIM and will increasingly distributed computing. Communication will become more more so in the context context of of increasingly distributed computing. Communication will be crucial crucial to to mainta.in maintain community community scientific scientific contacts, contacts, t,o to facilitate facilitate shared shared system system and will software maintenance maintenance based on regional regional expertise, expertise, to to allow allow necessary information information flow flow and soft,ware access at at all all levels, and to to meet meet the technical technical requirements requirements of of shared shared equipment. equipment. access vVe have had reasonable reasonable success success at at meeting meeting the geographical geographical needs of of the We community during during the the early early phases of of SUMEX-AIM SUMEX-AIM through through our our ARPANET ARPANET and and cornmunit,y TY~INET connections. connections. These have allowed allowed users from from many many locations locations within ,vithin the the United United T\-hINET States and and abroad abroad to to gain gain terminal terminal access access to to the the AIhl AIM resources and and through through ARPANET ARPANET States links t,o to communicat,e communicate much much more more voluminous voluminous file file information. information. Since many many of of our our users do links not have ARPANET ARPANET access access privileges privileges for for technical technical or or administratQve administrative reasons, a key key not problem impeding impt'ding remote remote use has been the the limited limited communications communications facilities facilities (speed, file file problem transfer. and and terminal terminal handling) handling) offered offered currently currently by by commercial commercial networks. networks. Commercial Commercial transfer, improvements a.re are slow slow in coming coming but but may may be expected expected to to solve the the file file transfer transfer problem problem improvements the next. few years. years. A A number number of of vendors vendors (AT&T, (AT&T, IBM, IBM, XEROX, XEROX, etc.) etc.) have yet yet to to in the announce commercially-available commercially-available facilities, facilities, but but TELENET TELENET is actively actively working working in in this this announce direction. We \-Ve plan plan to to continue continue experimenting experimenting with with improved improved facilities facilities as offered offered by by direction. commercial or government sources in the next grant term. We have budgeted for commercial or government sources in the next grant term. We have budgeted for cont.inued TYMNET TY~vINET service service and and an additional additional amount amount annually annually for for experimental experimental network network cont,inued connections. connections. Higil-:':peed intera.ctive interactive terminal terminal support support will will continue continue to to be a problem problem since one High-speed cannol. expect expect to to serve serve 1200 to to 9600 9500 baud baud terminals terminals effectively effectively over over shared shared long-distance long-distance ca.nnoi. trllnk lines lines with with gross gross capacities ca.pacities of of only only 9600 9500 to to 19200 baud. baud. We We feel feel this this is is aa problem problem trllnk E. A. A. Feigenbaum Feigenbaum E. 50 50 Progress - Future Future Plans SP41 5P4 1 RR0078S-11 RR00785- 11 that that is is best solved by distributed distributed machines able to effectively effectively support support terminal terminal interactions interactions locally locally and coupled coupled to other other AIM AIM machines and facilities facilities through through network network or telephone t,elephone links. links. As new machine machine resources are introduced introduced into into the community, community, we will will allocate advice to assure a,llocate budgeted budgeted funds with with Executive Executive Committee Committee assure effective effective communications communicat,ions links. links. Resource Resource Software So jtware We will will continue continue to maintain maintain the existing existing system, language, and utility utility support support, software software on our our systems at the most most current current release release levels, including including up-to-date up-to-date documentation. We also will will be be extending extending the facilities facilities available available to users where documentation. appropriate, rely appropriate, drawing drawing upon upon other other community community developments developments where possible. We rely heavily heavily on the needs needs of the user community community to direct direct system system software softwa.re development development efforts. efforts. \Vithin Within the AIM AIM community community we expect to serve as as a center center for software-sharing software-sharing between various This will will include include contributing contributing locallylocallyvarious distributed distributed computing computing nodes. This developed programs, those derived programs, distributing distributing derived from from elsewhere in the communit,y, community, maintaining maintaining up-t.o-date up-to-date information information on subsystems subsystems available, available, and assisting assisting in software software maintenance. ma.int h the management management committees. committees. This redistribut,ed project members by the individual principal investigators. redist,ribut.ed among among project members as directed directed by individual principal investigators. System project allocations Syst’em enforcement enforcement of of project a.llocations is done on a weekly weekly basis. As the weekly weekly file dump is done, if the aggregate aggregate space in use by by a project project exceeds exceeds its its allocation, allocation, files are dump if the archived from from associated user directories directories which which are over over allocation allocation until until the the project project is archived within its its authorized aut,horized lirnits. limits. within vVe are using using the TOPS-20 TOPS-20 class scheduler scheduler to to attempt attempt to to enforce enforce the 40:40:20 We bala,nce in terms terms of of CPU CPU utilization utilization and and to to avoid avoid system system and and user’ user' inefficiencies inefficiencies under under bal:-mce overload conditions. conditions. In practice, practice, the the 40:40 split split between between Stanford Stanford and non-Stanford non-Stanford over1on.d projects is fa.irly fairly well well realized realized (see (see Figure Figure 10 on page 34 and the tables tables of of recent recent project project projects llsage on page 3Gj. :36). usage Our job-scheduling job-scheduling controls controls bias the the allocation allocation of of CPU CPU time time based on per per cent cent time time Our consumed relative relative to to the the time time allocated allocated according according to to the the 40:40:20 40:40:20 community community split. split. consumed Howeyer, the the controls controls are “soft” "soft" in in that that they they do not not waste waste computer computer cycles ifif users below below However, their allocated percentages are not on the system to consume those cycles. In the early their allocated percentages not on the system to In the early years, the the operating operating dispa.rity disparity in in CPU CPU use reflected reflected a substantial substantial difference difference in demand demand years, betwren the the Stanford Stanford community community and and the the developing developing national national projects, projects, rather rather than than between inequity of of access. access. For For example, example, the the Stanford Stanford utilization utilization is spread spread over over a large large part part of of the the inequity 24-hour cycle, cycle, while while national-AIM national-AIM users tend tend to to be more more sensitive sensitive to to local local prime-time prime-time 24-hour A. Feigenbaum Feigenbaum E. A. 64 Resource Management Management and Allocation Allocation 5P41 5P41 RR00785-11 constraints. (The 3-hour 3-hour time time zone zone phase phase shift shift across across the continent continent is is of substantial substantial help constraints. (The in load-balancing). During load-balancing). During peak times under under the overload overload control control system system reported reported previously, previously, the Stanford St,anford community community experienced mutual mutual contentions contentions and delays while while the Alr..l AIM group groap had relatively relat.ively open access access to the system. This This disparity disparity in usage usage has disappeared disappeared in recent recent years with with the growth growth of the national nat)ional user community, community, and we enabled overload overload controls controls for for the national national community community as as well. For For the present. present, we propose to continue continue our our policy policy of "soft" “soft” allocation allocation enforcement enforcement for for the fair fair split split of resource capacity. capacity. Our Our system system also categorizes categorizes users use,rs in terms terms of access access privileges. privileges. These comprise comprise fully-authorized users, pilot projects, associates, guests, and network fully-authorized users, pilot projects, associates, network visitors visitors in descending order order of system syst.em capabilities. capabilities. \Ve We want want to encourage bona fide medical medical and health health research people to experiment experiment with with the various various programs programs available available with with a minimum minimum of red t.ape, tape, while users to bypass the advisory while Dot not allowing allowing unauthenticated unauthenticated advisory group group screening procedures procedures by by coming coming on as as guests. So far, far, we have had relatively relatively little little abuse compared compared to that that experienced experienced by by other other network network sites, perhaps perhaps because of the personal personal attention attent,ion directed direct’ed by senior senior staff staff to logon records, records, and to other other security security measures. However, However, the experience of most most other other computer computer managers managers behooves us to be cautious cautious about about being as as wide wide open as as might might be preferred preferred for for informal informal service to pilot pilot efforts efforts and demonstrations. demonstrations, \Ve !Ye will will continue continue developing developing this this mechanism mechanism in conjunction conjunction with with management management committee committee policy decisions. policy We We also have encouraged encouraged mature mature projects projects to apply apply for for their their own own machine machine resources in order order to preserve preserve the SUMEX-AIM SUMEX-AIM resource for for research and development development efforts efforts and to justify their t,o support support projects projects unable unable to justify their own machines. The The Rutgers Rutgers resource has its own part of project own 2050 2060 machine, machine, part of which which is allocated allocated for for AIM AIM use, use, and the CADUCEUS CADUCEUS project has instalit'd planned development program installed a.a. VAX V.4X 11/780 11/780 machine machine to t,o support support its its planned development and program test,ing work. “Simulation of of Cognitive Cognitive Processes" Processes” Project Project work. Profs. Profs. Lesgold Lesgold and Greeno's Greeno’s "SimUlation has tire ly to their h3.s moved moved en ent.irely their own own local VAX. VAX. 65 A. Feigenbaum Feigenbaum E. A. 5P41 RR00785-11 RR00785- 11 Dissemination Dissemination Efforts Efforts I.E. Efforts I.E. Dissemination Dissemination Efforts Throughout Throughout its existence, SUMEX-AIM SUMEX-AIM has expended substantial substantial effort effort toward toward disseminating disserninating information information about about its activities activities as as aa resource and about about the work work of individual We continue continue to make many many presentations presentations at individual collaborative collaborative projects. projects. professional professional meetings, meetings, to provide provide services to demonstrate demonstrate developed AI AI programs programs to interested visitors, and to work work in organizing organizing interested groups groups and individuals, individuals, to welcome visitors, workshops workshops within within the SUMEX-AIM SUMEX-AIM community community to introduce introduce our our research to collaborating collaborating professional We also directed professional communities. communities. directed considerable considerable effort effort in the past past toward toward working working with with the Research Resources Information Information Center Center to produce produce the "Seeds “Seeds of Artificial Art,ificial Intelligence" Int.elligence” monograph monograph and other other publications publications and press articles articles to address aa broader broader community community of technical technical and lay people. Software Software Distribution Distribution SUr-.IEX SUMEX continues continues to support support various various projects projects in the distribution distribution of of versions versions of their their sonware soft,wa.re to requesting requesting individuals individuals or or groups. groups. Following Following is a summary summary of of software software dissemination dissemination this this past past year: year: EMYCIN EMYCIN Both “source” versions versions of the EMYCIN EMYCIN Both the "executable" “executable” and "source" distribution package were restructured for distribution restructured for clarity clarity and ease ease of installation. Thirty copies of the EMYCIN package have been inst,allation. Thirty EMYCIN generated generated for for distribution distribution of which which about about 6 were sources only. only. An An Interlisp-VAX version Interlisp-VAX version of of EMYCIN EMYCIN is now available, available, thanks thanks to Ray Ray Bates of of USC-lSI, USC-ISI, who who did did the conversion. conversion. This This runs runs under under UNIX UNIX and VMS. VMS. AGE AGE Twenty-two Nearly Twenty-two copies of of the AGE AGE system system have been distributed. distributed. Nearly half half of of these have been copies requested requested in ANSI ANSI format format indicating indicating they they were evidently evidently going going to to non-Tops20 non-Tops20 sites (probably (probably Vaxes). As with with the EMYCIN EMYCIN system, system, Ray Ray Bates Bates at at USC-lSI USC-IS1 has converted converted AGE AGE to to run run under AX. A under Interlisp-V Interlisp-VAX. A version version is also available available for for the Xerox Xerox 1108 series Lisp Lisp workstations. workstations. GENET GENET In conjunction phaseout of conjunction with with the phaseout of the GENET GENET community community on SUMEX, programs and databases SUMEX, a software software package comprised comprised of of programs developed by by researchers researchers at at Stanford Stanford and elsewhere was assembled for for developed distribution to to interested interested GENET GENET users. Versions Versions of of the software software were distribution provided for for use on both both DEC-10 DEC-lO and DEC-20 DEC-20 systems systems operating operating under under provided TOPS-lO, TENEX. TENEX, and TOPS-20. TOPS-20. Installation procedures procedures were were TOPS-lo, Installation documented, and a substantial substantial amount amount of of telephone telephone consultation consultation was documented, provided. The The package has been well-received well-received and appears appears to to be in provided. active use at at many many of of the 21 academic academic sites to to which which itit was sent. Only Only active copy of of the the complete complete Genet Genet system system was set out out in the the past past year. year. one copy However, several several sets of of Genet Genet related related data data files have been distributed. distributed. However, This includes includes several several copies of of the the NIH NIH and EMBL EMBL Sequence Sequence Libraries. Libraries. This A limited limited amount amount of of operations operations support support has been given given to to Brutlag’ Brutlag's A s interaction with with Sam Sam Karlin Karlin of of the the Math Math department department and a variety variety of of interaction other groups. other groups. MRS MRS Twenty-two copies of of MRS MRS have have been distributed distributed through through Sumex. Sumex. Twenty-two Several others others have also been distributed distributed directly directly by by the the HPP. HPP. Most Most Several have been sent sent out out to to VAX/Unix V A.XjUnix sites or or Symbolics Symbolics Lisp Lisp machine machine sites. have A. Feigenbaum Feigenbaum E. A. 66 Dissemination Dissemination Efforts Efforts 5P41 RR00785-11 RR00785-11 5P41 SACON SACON Two copies copies of of SACON SACON have have been been prepared prepared and and distributed. distributed. Two GLISP GLISP Two copies copies of of GLISP GLISP were were distributed. distributed. Two 67 67 E. A. A. Feigenbaum Feigenbaum E. Comments Comments on on the the Biot,echnology Biot,echnology Resources Resources Program Program 5P41 5P41 RR00785-11 RR00785- 11 I.F. on Resources Program I.F. Comments Comments on the the Biotechnology Biotechnology Resources Program Resource Resource Organization Organization We continue continue to believe that that the Biotechnology Biotechnology Resources Program Program isis one one of of the most effective vehicles for developing and disseminating technological tools for biomedical effective for developing and disseminating technological for biomedical research. resea.rch. The The goals goals and and methods methods of of the program program are are well-designed well-designed to encourage encourage building building of groups of the necessary multi-disciplinary multi-disciplinary groups and and merging merging of of appropriate appropriate technological technological and and medical medical disciplines. disciplines. In In our our experience experience with with the SUMEX-AIM SUMEX-AIM resource, several several elements elements of of this this approach approach seem seem to emerge as as key to the development development and management management of an effective effective resource: 1. 1. Effect,ive Effective Management Management Framework Framework -- There There needs needs to be be an explicit explicit agreement agreement between the t.he BRP BRP and the resource principal principal investigator investigator which which establishes aa clear c1ea.r mandate mandate for the resource and its allocation, allocation, provides provides worthwhile worthwhile incentives incentives for the host institution institution and investigator investigator to invest invest the necessary substantial substantial professional professional career time time to develop and manage the resource, and ensures equitable equitable distribution distribution of of resource services to its target target community. community. 2. 2. Close \Vorking Working Relationship Relationship with with the NIH NIH -- A A resource is a major major and often often long-term long-term investment investment of of money money and human human energy. A A close close and mutuallymutuallysupportive supportive working working relationship relationship between between resource management, management, its advisory advisory committees, is essential to assure healthy NIH administration administration healthy committees, and the NIH development development of of the resource and its relationship relationship to its user community. community. We We at SUMEX-AHvl have benefited SUMEX-AIhl benefited immensely immensely from from such a relationship relationship with with Dr. Dr. William William R. Baker, Baker, Jr., Jr., in the evolution evolution of of the SUMEX-AIM SUMEX-AIM community. community. We We look beneficial relationship look forward forward to a continuing continuing mutually mutually beneficial relationship with with Dr. Dr. Baker's Baker’s successor at the NIH. the NIH. 3. Freedom by its Freedom to Explore Explore Resource Potential Potential -- A A resource, by its nature, nature, operates operates at the “cutting edge" edge” in developing t)he "cutting developing its its characteristic characteristic technology technology and learning learning to effectively effectively disseminate disseminate itit to to the the biomedical biomedical community community at at large. The The BRP BRP t,o should not not impose artificial artificial constraints constraints on the the resource for for commercializing commercializing its its should efforts (fees for for service) service) or or developing developing its its potential potential (funding (funding duration duration limits limits or or efforts annual budget budget ceilings). ceilings). artificial policy policy impositions impositions can serve to to annual Such artificial lIndermine the very very goals central central to to the the BRP’ BRP's for existence. existence. llndermine s reason for Satisfactory policies policies in this this regard regard have been worked worked out out and and should should be Satisfact.ory retained. retained. Electronic Communications Communication,Oj Electronic SUMEX-AIM has pioneered pioneered in developing developing more more effective effective methods methods for for facilitating facilitating SUMES-AIM scientific communication. commllllication. Whereas \Vhereas face-teface face-to-face contacts contacts continue continue to to play playa key role, role, in in scient.ific a key the longer-term longer-term we feel that that computer-based computer-based communications communications will will become increasingly increasingly the important to to the the NIH NIH and and the the biomedical biomedical community. community. We We would would like like to to see see the the BRP BRP take take important more adive active role role in promoting promoting these tools tools within within the the NIH NIH and and its its grantee grantee community. community. aa more E. A. A. Feigenbaum Feigenbaum E. 68 68 sp·a RROO785-11 RR00785-11 5P-11 Description Description of of Scientific Scientific Subprojects Subprojects ll. Description Description of of Scientific Scientific Subprojects Subprojects II. il.A. Scientific Scientific Subprojects Subprojects ILA. The following following subsections subsections report report on the the AIM AIM community community of of projects projects and “pilot” "pilot" The efforts including including local local and national national users of of the the SUMEX-AIM SUMEX-AIM facility facility at at Stanford. Stanford. efforts However, those projects projects admitted admitted to to the the National National AIM. AIM community community which which use the the RutgersRutgersHowever, AIM resource resource as their their home base are not not explicitly explicitly reported reported here. AIhf addition to to these det;ailed detailed progress progress reports, reports, abstracts abstracts for for In a.ddition individual users are submitted submitted on a separate separate Scientific Scientific Subproject Subproject individual have included included here briefer briefer summary summary abstracts abstracts of of the the fully-authorized fully-authorized B on page 209. B project and its its each project Form. However, However, we Form. projects in in Appendix Appendix projects The collaborative collaborative project project reports reports and comments comments are the the result result of of a solicitation solicitation for for The contributions sent sent to to each of of the the project project Principal Principal Investigators Investigators requesting requesting the the following following contriblltions information: information: I. SUMhIARY SUMMARY OF OF RESEARCH RESEARCH PROGRAM PROGRAM A. Project Project rationale rationale A. Medica.l relevance relevance and collaboration collaboration B. Medical Highlights of of research research progress progress C. Highlights --Accomplishments this this past past year year --Accomplishments --Research in progress progress --Research D. List List of of relevant relevant publications publications E. Funding Funding support support II. INTERACTIONS \\11TH INTERACTIONS WITH THE THE SUMEX-AIM SUMEX-AIM RESOURCE RESOURCE A. Medical Medical collaborations collaborations and program program dissemination dissemination via via SUMEX SUMEX B. Sharing Sharing and interactions interactions with with other other SUMEX-AIM SUMEX-AIM projects projects (via (via comp1lting computing facilities, facilities, workshops, workshops, personal personal contacts, contacts, etc.) etc.) C. Critique Critique of of resource management management (community (community facilitation, facilitation, computer computer services, communications communications services, capacity, capacity, etc.) etc.) III. III. RESEARCH RESEARCH PLANS PLANS A. Project Project goals and plans --Near-term --Near-term --Long-range --Long-range B. .Justification .Justifica.tion and requirements requirements for for continued continued SUMEX SUMEX use use C. Needs Nerds and plans for for other other computing computing resources beyond beyond SUMEX-AIM SUMEX-AIM D. Recommendations Recommendations for future future community community and resource development development \'Ve We believe that that the reports reports of the individual individual projects projects speak for for themselves themselves as as case, the reports reports are recorded recorded as as submitted submitted and are rationales In any case, rationales for for participation. participation. The only only exceptions exceptions are the respective respective the responsibility responsibilit#y of the indicated indicated project project leaders. The lists of relevant relevant publications publications which which have been uniformly uniformly formatted formatted for parallel parallel reporting reporting on the Scientific Scientific Subproject Subproject Form. Form. 69 59 Feigenbaum E. A. Feigenbaum Stanford Projects Projects St.anford RR00785-11 5P41 RR00785-11 II.A.!. Stanford Stanford Projects Projects II.A.l. The following following group group of of projects projects is formally formally approved approved for for access access to to the the Stanford Stanford The aliquot of of the the S SUMEX-AIM resource. Their Their access access is based on on review review by by the the Stanford Stanford alicluot UXIEX-AIM resource. Advisory Group Group and and approval approval by by Professor Professor Feigenbaum Feigenbaum as Principal Principal Investigator. Investigator. ,4dvisory In addition addition to to the the progress progress reports reports presented presented here, abstracts abstracts for for each project project and and In its individual individual users are submitted submitted on a separate separate Scientific Scientific Subproject Subproject Form. Form. its E. A. ,4. Feigenbaum Feigenbaum 70 EXPEX EXPEX - Expert Expert Explanation Explanation Project Project 5P41 5P41 RR00785-11 IT.A.1.1. EXPEX - Expert Explanation Project II.A.l.l. EXPEX Expert Explanation Project EXPEX Project EXPEX - Expert Expert Explanation Explanation Project Edward M.D., Edward H. H. Shortliffe, Shortliffe, M.D., Ph.D. Ph.D. Departments of Science Departments of Medicine Medicine and and Computer Computer Science Stanford Stanford University University I. I. SUM:MARY OF PROGRAM SUMMARY OF RESEARCH RESEARCH PROGRAM A. A. Project Project Rationale Rationale EXPEX EXPEX is is not not a single project project but but a combination combination of of efforts efforts that that are directed directed at basic ba.sic issues issues in the development development of of representational representational schemes to facilitate facilitate knowledge knowledge acquisition The The work work includes includes not not only only the study study of of fundamental fundamental acquisition and explanation. explanation. representational represent,ations.! formalisms formalisms but but also the encoding encoding of various various types types of of knowledge, knowledge, such as as causal information information and user models. In In addition, addition, to complement complement these research directions, directions, the project project has served as as the focus for for preparing preparing three three books on medical medical computing computing research. vVe We believe that that the productivity productivity of basic computer computer science research tends tends to be heightened heightened by experiments experiments that that deal with with significant significant real world world problem problem domains. domains. Challenges biology have introduced Challenges drawn drawn from from chemistry, chemistry, medicine, medicine, and molecular molecular biology introduced additional but have simultaneously addition4 complexity complexity to to expert expert systems systems work work at at Stanford, Stanford, but simultaneously forced forced system developers to respond to pragmatic constraints and user demands system developers to respond to pragmatic constraints demands that that have had a significant significant impact impact on the basic AI AI techniques techniques selected or or developed. developed. Thus, Thus, we believe that that creative by working creative investigation investigat,ion into into symbolic symbolic reasoning reasoning techniques techniques is facilitated facilitated by working in real world Much world settings sett,ings where where the application application forces us to to avoid avoid oversimplification. oversimplification. Much of of our our research effort effort therefore therefore deals with with medical medical domains domains (viz., (viz., endocrinology endocrinology and renal renal pathophysiology). pathophysiology). Relevance and B. lvfediral Medical Relevance and Collaboration Collaboration Our Our interest interest, in explanation explanation derives derives from from the insights insights we gained gained in developing developing explanatory ca.pabi!ities capabilities for for the the MYCIN MYCIN system. system. In the the case case of of MYCIN MYCIN and and its its explana.tory In descendents, we have been able to to generate generate intelligible intelligible explanations explanations by by taking taking advant,age advantage descendents, of its its rule-based rule-based representation representation scheme. Rules can be translated translated into into English English for for display display of to a user, their interactions explicitly demonstrated. By adding and their interactions can also be explicitly demonstrated. to By adding mechanisms for for understanding understanding questions questions expressed in simple simple English, English, we were were able to to mechanisms create an interactive interactive system system that that allowed allowed physicians physicians to to convince convince themselves themselves tha.t that they they create with t,he the basis for for the the program’ program's recommendations. The limit,ations limitations of of the the agreed with s recommendations. The explanations generated generated in this this way way have become increasingly increasingly obvious, obvious, however, however, and have expIa.nations to improved improved characterization characterization of of the the kinds kinds of of explanation explanation capabilities capabilities that that must must be led t.o developed ifif clinical clinical consultation consultation systems systems are to to be accepted accepted by by physicians. physicians. The The potential potential developed of workstation workstation graphics graphics as a means of of avoiding avoiding natural natural language language issues issues in in the the use of explanation process is also an area area of of great great promise promise with with which which we are currently currently exp!a.nat.ion experimen ting. experimenting. \Vith these motivations motivations in mind, mind, we are involved involved in a series of of research research projects projects With that address medical medical knowledge knowledge representation representation and and explanation. explanation. The The individual individual projects projects that include the following: include following: 71 A. Feigenbaum Feigenbaum E. A. EXPEX EXPEX - Expert Expert Explanation Explanation Project Project 5P41 5P4 1 RR00785-11 RROO785-11 1. 1. Mr. hlr. Greg NESTOR program program uses Greg Cooper's Cooper’s NESTOR uses a detailed detailed knowledge knowledge base regarding pathophysiologic relationships The program is regarding pathophysiologic relationships in hypercalcemia. hypercalcemia. The program designed to physician's proposed explanation to critique critique a physician’ s hypothesis hypothesis regarding regarding a proposed expla.nation for patient manifestations for a set of of patient manifestations when when an elevated elevated serum serum calcium calcium has been observed. particular interest Of particular interest is the techniques techniques Cooper Cooper has developed developed for for observed. Of using knowledge knowledge of of causality causality to to avoid avoid the assumption assumption of of conditional conditional independence independence commonly commonly used in Bayesian Bayesian diagnosis diagnosis systems. 2. Mr. pathophysiology, Mr. John Kunz Kunz has represented represented the the knowledge knowledge of of renal renal pathophysiology, including including the quantitative quantitative relationships relationships that that characterize characterize the way way in which which the body manages water body water and electrolytes, electrolytes, to develop develop a consultation consultation and analysis analysis system system (AI/MM) (AI/MM) that that melds mathematics mathematics and AI AI techniques. techniques. 3. Building Building on his earlier earlier experience experience with with developing developing an explanation explanation capability capability for for NEOM YCIN (in project members NEOMYCIN (in collaboration collaboration with with the GUIDON GUIDON project members as as outlined outlined elsewhere in this this report), report), Dr. Dr. Glenn Glenn Rennels has begun to to work work on a new system syst*em that that uses uses knowledge knowledge of of medicine medicine to help help formulate formulate and resolve complex complex decision better decision analyses. Convinced Convinced that that decision decision analytic analytic techniques techniques would would be better accepted physician were to accepted in medicine medicine if if the physician to interact interact with with a knowledge-based knowledge-based interface int)erface (rather (rather than t,han with with the the decision decision trees themselves), themselves), Dr. Dr. Rennels has made use of "influence diagrams" as a central method central method for for guiding guiding the the of “influence diagrams” as interaction. The explanation issues become especially evident when interaction. The explanation issues especially evident when an analysis analysis is complete complete and his system system needs to to generate generate a defense for for the recommendation recommendation it it has made. 4. Mr. Mr. Curt Curt Langlotz La.nglotz has continued continued to to work work on a hypothesis hypothesis assessment module module for program uses This program uses a critiquing critiquing model model which which for the ONCOCIN ONCOCIN system. system. This inherently inherently involves involves advanced advanced explanation explanation techniques. techniques. The The work work uses uses the the Xerox Xerox 1108 professional professional workstation workstation (Dandelion) (Dandelion) and is further further described described in the ONCOCIN portion of ONCOCIN Project Project portion of this this annual annual report. report. S. project using the Xerox .5. During During 1983, Ms. Shako Shoko Tsuji Tsuji completed completed a project Xerox workstation workstation to experiment experiment with with graphical graphical techniques techniques for for examining, examining, manipulating, manipulating, expanding, base. Also expanding, and editing editing a large medical medical knowledge knowledge base. Also working working in the context of ONCOCIN, her code was designed for use by knowledge context of ONCOCIN, her for use by knowledge engineers. The building an interface The work work has inspired inspired subsequent subsequent work work in building interface for for the nonprogrammer clinician protocols in the programmer clinician who who wishes to to write write and test t,est new protocols ONCOCIN The project is described ONCOCIN environment. environment. The project described in greater greater detail detail in the ONCOCIN portion of ONCOCIN portion of the annual annual report. report. To prepared two booles on To complement complement, these basic research activities, activities, we have hat-e prepared two boo!cs Artifidal beginning work Artificial Intelligence Intelligence in Medicine Medicine and are beginning work on a third third (see (see Section Section C for for details). details). C. Highlights of Research Progress Progress G. Highlights o j Research C.l NESTOR CT.1The NI3’ TOR System Sydem NESTOR is intended patient data NESTOR intended to allow allow a user to t’o input input patient data plus a hypothesis, hypothesis, and then then have the system system critique critique that that hypothesis hypothesis in light light of of the data. data. The The system, system, an evolving evolving thesis project that t.hesis project that is largely largely the work work of of Mr. Mr. Greg Greg Cooper, Cooper, relies on basic associational associational information part from but supplemented information drawn drawn in part from the INTERNIST-J INTERNIST-I knowledge knowledge base but supplemented with with causal and temporal t.emporal associations. associations. The behind this physicians want The motivation motivation behind this research research is the conviction conviction that that physicians want active active E. A. Feigenbaum Feigenbaum 72 EXPEX EXPEX - Expert Expert Explanation Explanation Project Project 5P41 5P41 RR00785-11 R.R00785-11 control cont,rol of the diagnostic diagnostic process and that that they they also want want and need aa system system that that explains, explains, in aa user-tailored user-tailored way, its evaluation evaluation of the physician's physician’s hypothesis. hypothesis. There There may may be be times times when the user wants wants to give complete complete control control to NESTOR NESTOR and just just be be in aa mode of answering answering questions, but but we feel feel that that this this should be be an option option and not not aa requirement. requirement. It It is is observations observations such as as these that that have also accounted accounted for for the hypothesis hypothesis assessment assessment work work underway underway in the ONCOCIN ONCOCIN research, briefly briefly mentioned mentioned above and further further described described in the section section of this this report report dealing dealing with with that that project. project. The The initial initial NESTOR NESTOR system system is is now largely largely complete complete and is undergoing undergoing evaluation evaluation at this time. Of particular interest is the adequacy of the techniques Of particular interest techniques developed for this time. allowing allowing NESTOR NESTOR to avoid avoid the traditional traditional assumption assumption of conditional conditional independence used in Bayesian systems. syst,ems. Also, because because NESTOR's NESTOR’s probabilistic probabilistic model is is more formal formal than than the ad hoc scheme used used in, say, INTERNIST, INTERNIST, the assumptions assumptions made by our our system system are more explicit. explicit. \Ve have also developed search techniques techniques that that allow allow NESTOR NESTOR to to explore explore efficiently efficiently a very very large search space in order order to find find the most most probable probable (multiple (multiple disease) disease) hypothesis. hypothesis. This This technique technique is is general and can be applied applied to many many nonmedical nonmedical problems problems where where the goal is to find find the most most probable probable hypothesis hypothesis among many many possibilities. possibilities. C.2 C.2 Integrating Integrating ~Mathematical Mathematical Models Models with with AI AI Afethods Methods This This research project, project, known known as as AI/MM, AI/MM, is the dissertation dissertation research of of Mr. Mr. John Kunz. The The system system int,egrates integrates AI AI and simple simple mathematics mathematics to analyze a physiological physiological Kunz. model. model. In a selected medical medical domain domain (renal (renal physiology), physiology), we have built built a computer comput,er program It program based on these techniques. techniques. It analyzes physiological physiological behavior, behavior, diagnoses abnormality, program fits abnormality, and explains explains the rationale rationale for for its its analyses. The The program fits data data to the model. model, identifies identifies whether whether the data data are abnormal, abnormal, and identifies identifies the possible causes causes and effects The physiological model effects of of any any abnormalities. abnormalities. The physiological model is based on knowledge knowledge about about anatomy, physiological system, anatomy, the hehavior heha,vior of of the physiological system, and the mechanism mechanism of of action action of of the system. problems discussed system. It's It’s validity va,lidity has been tested tested by by having having it it analyze many many of of the problems in Valtin's Renal Function. Function. Va.ltin‘s text text IZellal The project have been to: The specific specific aims of of this this project 1. physiological model. 1. Dev('lop Develop a vocabulary vocabulary for for a physiological model. The The vocabulary vocabulary represents represents the "basic physiology" of “ba,sic physiology” of a biological biological system system and appears to be adequate to the concepts concepts included introductory professional-level professional-level physiology physiology express Lhe included in an introductory text. text. Develop a reasoning reasoning system system which which can solve problems problems expressed expres;~ed in the 2. Develop vocabulary. vocabul:try. Demonstrate the basic necessity, appropriateness appropriateness and limitations limitations of of the 3. Demonstrate vocabulary and reasoning reasoning procedure. procedure. vocabulary C.3 Knowledge-Based Explanations Explanations in in a Decision Decision Analysis Analysis Environment Environment Cr.3 Kno,ufledge-Based This new project, project, thesis research by by Dr. Dr. Glenn Glenn Rennels, is motivat,ed motivated by by the This observation that that AI AI techniques techniques could could greatly greatly facilitate facilitate a user’ user's effort to to specify specify the the det.ails details observation s effort of a complex complex clinical clinical decision decision task task and and to to seek assistance with with that that task. task. Although Although of decision theoretic theoretic notions notions have been shown shown to to be relevant relevant to to such medical medical problems, problems, they they decision largely been unused by by clinicians, clinicians, even when when computer-based computer-based solutions solutions have been have largely offered. \Ve \Ve believe believe that that an intelligent intelligent system system should should be able to to guide guide the the definition definition of of the the offered. decision t.ask task and explain explain the the results results of or the the analysis analysis without without requiring requiring that that a user be decision familiar wit,h with the the underlying underlying decision decision analytic analytic techniques techniques being being used to to solve the the problem. problem. familiar 73 A. Feigenbaum Feigenbaum E. A. EXPEX EXPEX - Expert Expert Explanation Explanation Project Project 5P41 RR00785-11 RR00785-11 The The basic notion notion is to to use directed directed graphs, graphs, termed termed "influence “influence diagrams" diagrams” as as a language for physician at for communication communication with with a physician at a graphical graphical display display terminal. terminal. Nodes in these graphs by the user who graphs are defined defined by who is seeking advice, and their their structure structure and meaning meaning is largely ia.rgely intuitive. intuitive. The The task task of of converting converting influence influence diagrams diagrams to to decision decision trees is a knowledge-based problem that potentially well-suited knowledge-based problem that is potentially well-suited for for a solution solution that that uses uses AI AI methods. Similarly, the the results results of of a decision decision analysis, analysis, including including the sensitivity sensitivity analysis, analysis, methods. Similarly, will physician user in terms will need to to be explained explained to to the physician terms of of influence influence diagrams diagrams and knowledge The being The necessary knowledge knowledge structures struct)ures are currently currently being knowledge of of the domain. domain. designed, and an early prototype system The The research uses uses a 9835 9836 early prototype system is operational. operational. workstation by Hewlettworkstation donated donated to the Medical Medical Information Information Sciences Sciences Training Training Program Program by HewlettPackard Packard Company Company and soon to to be networked networked to to the SUMEX SUMEX 2060. C.4 Books Books on Medical Afedical Arti Artificial Intelligence and Afedical Computing jicial Intelligence and Medical Computing \Ve both of published 1Ve have completed completed two two books, both of which which are in press and due to be published in mid-1984: mid- 1984: W.J. E.H. Readings in Medical Artificial Artificial W.J. and Shortliffe, Shortiiffe, E.H. Readings in Medical Intelligence: The First First Decade. Reading, Reading, MA: MA: Addison-Wesley, Addison-Wesley, 1984. Intelligence: •0 Clancey. Clancey, Rule-Based Ezpert Expert Systems: Rule-Based Systems: the .MYCIN Experiments of Heuristic Program.ming Programming Project. Project. h,fi%IN Experiments o j the Stanford Stan ford Heuristic Reading. Reading, MA: MA: Addison-\Vesley, Addison-Wesley. 1984. •l Buchanan, Buchanan, \V.J. W.J. and Shortliffe, Shortiiffe, E.H. E.H. In addition, just begun work beginning to addition, we have just work on a textbook textbook for for students students beginning t.o study study medical This medical computing computing and artificial artificial intelligence. intelligence. This multi-authored multi-authored volume volume should should be completed by the end of publication date is contemplated. completed in draft draft form form by of 1984. A A 1985 publication contemplated. •l Shortliffe, An Introduction Introduction to Short.iiffe, E.H., E.H., \Viederhold, Wiederhold, G.C.M., G.C.M., and Fagan, Fagan, L.M. L.M. An Afedical Computer preparation). Rfedical Computer Science. Science. Reading, Reading, MA: MA: Addison-Wesley Addison-Wesley (in (in preparation). D. Publications Publications Since January 1989 D. Since January 1983 1. Shortliffe, Shortliffe, E.H. E.H. Medical In Medical consultation consultation systems: systems: designing designing for for doctors. doctors. In Dcsigning for Human-Computer Communication (M.S. Sime and M.J. Designing for Human-Computer Communication (M.S. M.J. Coombs. Coombs. eds.), eds.). Chapter Chapter 8, pp. 209-238, London: London: Academic Academic Press, 1983. 2. Shortliffe, Shortliffe, E. H. Medical Medical Cybernetics: Cybernetics: The The Challenges Challenges of of Clinical Clinica. Computing. Computing. In Tcchnology International Stability, and Growth, S. Basheer Ahmed In Tech,nology International Stability, and Growth, S. Ahmed and Alice Alice P. Ahmed, Ahmed, editors; editors; Chapter Chapter 12, pp. 148-165; Associated Associated Faculty Faculty Press, Inc., New York, Inc., Port, \Vashington, Washington, New York, 1984. 3. (*) (*) Shortliffe, Shortliffe, E.H. E.H. and Fagan, Fagan, L.M. L.M. Expert Expert systems systems research: modeling modeling the medical An Integrated Integrated Approach Approach to Monitoring Afonitoring In An medical decision decision making making process. In (J.S. Gravenstein. Newbower, A.K. N.T. Smith, Gravenstein, R.S. Newbower, .4.K. Ream, and N.T. Smith, eds.), pp. 183-200, Woburn, Woburn, MA: hIA: Butterworth's, Butterworth’s, 1983. 1983. '1. 4. Duda, Duda, R.O. R.O. and Short-Iiffe, Shortliffe, 220:251-268 220:261-368 (1983). E.H. E.H. Expert Expert systems systems research. Science, Science, 5. (*) (*) Langlotz, Langiotz. C.P. C.P. and Shortliffe, Shortliffe, E.H. E.H. Adapting Adapting a consultation consultation system system to to critique plans. International Journal Journal of Studies, critique user plans. International o j lI.-fan-Afachine Man-Muchine Studies, 19>179-495 19:479-496 (1983) (1983) 6. Shortliffe, Hypothesis Shortliffe, E.H. E.H. Hypothesis generation generation in medical medical consultation consultation systems: systems: art·ificial In lI.-fEDINFO 83 (J.H. artificial intelligence intelligence approaches. approaches. MEDINFO (J.H. van van Bemmel, Bemmel, M. North Holland, M. Ball, Ball, and a.nd O. 0. Wigertz, Wigertz, eds.), eds.). pp. 480-483, North Holland, Amsterdam, Amsterdam, 1983. E. A. Feigenbaum Feigenbaum 74 EXPEX EXPEX -- Expert Expert Explanation Explanation Project Project 5P41 5P4 1 RR00785-11 RR00785- 11 7. Graphical access access to to the the knowledge knowledge base base of of aa 7. (*) (*) Tsuji, Tsuji, S. S. and and Shortliffe, Shortliffe, E.H. E.H. Graphical medical consultation system. Proceedings of AA . A'[SI Congress 83, Congrese 83, pp. pp. Proceedings of A4MSI medical consultation system. 551-555, San Francisco, Ca . , May 1983. 551-555, Ssn Francisco, Ca., May 1983. 8. The In The science science of of biomedical computing. computing. In A1eeting hleetilzg the the 8. Shortliffe, Shortliffe, E.H. E.H. and Medical Challenge: Challenge: Informatics In formatics Medical Education Education (J.C. (J.C. Pages, Pages, A.H. Levy, Levy, F. F. Gremy, Gremy, and and J. J. Anderson, eds.), eds.), pp. 1-10, l-10, Amsterdam: Amsterdam: North-Holland, North-Holland, 1983. 1983. To To be be reprinted reprinted ih ih Medical Medical Informatics, Informatics, 1984. 1984. 9. 9. (*) (*) Kllnz. Kunz, J.C., J.C., Shortliffe, Shortliffe, E.H., E.H., Buchanan, Buchanan, B.G., B.G., and and Feigenbaum, Feigenbaum, E.A. E.A. Comparison Comparison of techniques of computer-assisted decision decision making making in medicine. In 111Pure and Applied .4ppfied Biostructure Biostructure (Claudio (Claudio Niccolini, Niccolini, Ed.), Singapore: World World Press. Press, 1983. 1983. 10. 10. (*) (*) Kunz, J.C., Shortliffe, Shortliffe, E.H., Buchanan, B.G., B.G., Feigenbaum, E.A. ComputerComputerassisted of Philosophy Philosophy and A-fedicine, Medicine, assisted decision making making in medicine. Journal Journal of Summer 1984 1984 (in press). press). 11. ll.‘(*)(*) HasJing, Hssling, D. \V., W., Clancey, W. W. J., a.a diagnostic consultation system. diagnostic consultation Studies, Studies, Spring Spring 1984 1984 (in press). press). and Rennels, Rennels, G. Strategic Strategic explanations explanations for International Journal of International Journal of Afan-Afachine Man-Machine 12. 12. Shortliffe, Shortliffe, E.H. Reasoning methods in medical consultation consultation systems: artificial artificial intelligence intelligence approaches (tutorial). (t,utorial). Computer Computer Programs Programs in in Biomedicine, Biomedicine, January January 1984 1984 (in press). press). E. Funding Funding Support Support Grant Grant Title: Title: "The “The Development Development of Representation Representation Methods Methods to Facilitate Facilitate Knowledge Knowledge Acquisition Acquisition and Exposition Exposition in Expert Expert Systems" Systems” Principal Principa.1 Investigator: Investigator: Edward Edward H. Shortliffe Shortliffe Agency: Naval Research; ID Number: NR NR 049-479 A4gency: Office Office of of Naval ID Number: 049-479 Term: January January 1981 to December 1983 Term: 1981 to Total award: award: $~456,622 $455.622 Total Grant Title: Title: “Research "Research on Introspective Introspective Systems” Systems" Grant Principal Investiga.tor: Investigator: M Michael Principal ichael R. Genesereth Agency: Office Office of of Naval Naval Research; ID ID Number: Number: NR NR 049-479 Agency: Term: Jnnunry January 1984 1984 to to December 1986 Term: Total award: award: $312,070 $:312.070 Total Title: “Information "Information Structure Structure and Use in Knowledge-Based Knowledge-Based Expert Expert Grant. Title: Systems" Systems” Principal Investigat.or: Investigator: Bruce Bruce G. Buchanan Buchanan Principa.1 Agency: Na.tionsl National Science Science Foundation; Foundation; ID ID Number: Number: 83-12148 83-12148 Agency: Term: h,larch t-.larch 1984 to to February February 1987 Term: Tota,l award: $300,000 $300.000 (includes (includes indirect indirect costs) Tot al award: ll. II. INTERACTIONS WITH WITH THE THE SUMIZX-AIM SUMEX-AIM RESOURCE RESOURCE INTERACTIONS A.. Medical Afedical Collaborations Collaborations and and Program Program Dissemination Dissemination via via SUMEX SUAfEX A. None of of t,hese these new new programs programs is is yet yet ready ready for for disseminat,ion. dissemination. None 75 75 They are are mostly mostly They E. A. A. Feigenbaum Feigenbaum E. 5P41 RR00785-11 RR00785- 11 EXPEX EXPEX - Expert Expert Explanation Explanation Project Project fundamental research experiments experiments with with limited limited clinical clinical usefulness usefulness other other than than as funda.mental research demonstration demonstrat,ion projects. projects. Our Our past past experience experience has shown, shown, however, however, that that SUMEX SUMEX provides provides a superb superb vehicle vehicle for for demonstrating demonstrating systems, systems, even at a dist.ance. dist,ance. The The new book book writing writing eCfort effort will will in particular particular be facilitated facilitated by by SUMEX, SUMEX, much much as the AI A1 Handbook Handbook was in the the past. past. A A multi-authored multi-authored text text of of this this type, type, particularly particularly one for for which which the the authors authors are spread spread at at numerous numerous different different universities universities around around the country, country, would be a nightmare nightmare to to compile compile if if it it were not not for for the the SUMEX SUMEX resource. resource. Many Many of of the the would of manuscript manuscript contributors to to the the book book have been assigned SUMEX SUMEX accounts accounts for for purposes purposes of contributors preparation. preparation. Online Online manuscript manuscript work work through through the the shared shared facility, facility, coupled coupled with with messaging messaging capabi\it.ies, capabilit,ies, will will greatly greatly enhance the the efficiency efficiency and and accuracy accuracy oC of the developing developing chapters chapters the editing editing process. and the B. Interaction with UM Project8 Projects B. Sharing Sharing and and Interaction with Other Other SUMEX-.. SUMEX-AIM Although our our EXPEX EXPEX work work is young, young, we are already already benefiting benefiting Cram from interactions interactions Although with ot.her other researchers researchers who who use use the SUMEX-AIM SUMEX-AIM resource. The NESTOR work in with resource. The NESTOR work particular knowledge particular has depended depended on access access to to the the INTERNIST-l INTERNIST-l knowledge base and on Crequent frequent exchange of Similarly, of messa.ges messa.ges with with the researchers researchers at at the the University University of of Pittsburgh. Pittsburgh. Similarly, our our collaboration collaboration with with the the GUIDON GUIDON research research team team for for the the implementation implementation of of an explanation explanation capability capability would would not not have been possible possible without without the the facilitated facilitated communication communication and and shared shared file access access available available via via SUMEX. SUMEX. file C. Critique Resource Management Atfanagement Critique of of Resource SUf\IEX Not SUhIES continues continues to to provide provide a superb superb environment environment for for research research of of this this kind. kind. Not only only is the the 201)0 2060 a well well managed managed resource resource under under Ed Ed Pattermann's Pattermann’s leadership, leadership, but but the hypothesis hypothesis assessment and graphical graphical query query systems systems are dependent dependent upon upon access access to to high high performance performance professional professional workstations, workstations, and and we are delighted delighted with with the resources that that SUMEX SUMEX has provided provided us in this this regard. regard. III. RESEARCH PLANS ID. RESEARCH PLANS A. Project Project Goals Plan8 A. Goal.9 and and Plans \Ve We anticipate ant,icipate completion completion of of many many of of these basic research research efforts efforts during during the the coming coming year. NESTOR work year. Cooper's Cooper’s NESTOR work is largely largely complete, complete, and a thesis thesis document document is anticipated anticipated in June June 1984. Similarly, Similarly, Kunz Kunz has completed completed his work work on AI/MM, AI/MM, and and his dissertation dissertation is approaching approaching completion. completion. Both Both Cooper Cooper and and Kunz Kunz have completed completed their their oral oral examinations examinations this ",,"ork. work. on this The The project project oC of Tsuji Tsuji is complete complete and and she has now now lett left StanCord. Stanford. However, However, the the code she developed developed is being being modified modified Cor for ongoing ongoing use in the the ONCOCIN ONCOCIN environment. environment. The The project project oC 01 Langlotz Langlotz continues continues to to be an active active research research effort effort within within the the ONCOCIN His plan ONCOC’IN project. project. plan for for the the coming coming year year is briefly briefly outlined outlined in the ONCOCIN ONCOCIN portion of of this this annual annual report. report. portion The just getting The work work of of Rennels, Rennels, which which is just getting underway, underway, will will be better better formulated formulated by by next next year year at at this this time. time. We We expect expect the the project project to to last last at at least least two two more more years, years, however. however. The The textbook textbook preparation preparation is scheduled scheduled for for completion completion in approximately approximately one year, year, with publication publication anticipated anticipated during during H185. 1985. with E. A. Feigenbaum Feigenbaum 76 76 EXPEX EXPEX - Expert Expert Explanation Explanation Project Project 5P41 5P4 1 RR00785-11 RROO785-11 B. Requirement8 Requirement.'} for for Continued Continued SUAfEX SUMEX Use All All the work work we are doing doing is largely largely dependent dependent on the SUMEX SUMEX re.,ource. resource. The The new work by the Medical work of of Rennels Rcnnels is using Hewlett-Packard Hewlett-Packard 9835 9836 workstations workstations owned owned by Medical Information program, but but Dr. Information Sciences training training program, Dr. Rennels continues continues to be dependent dependent upon upon SUMEX for projects, only SUhlEX for communication communication and collaboration. collaboration. Of Of the other other projects, only the t,he hypothesis hypothesis assessment and graphical projects are sufficiently justify their graphical query query projects sufficiently mature mature to justify bheir transfer transfer to to one of personal workstations, of the SUMEX SUMEX personal workstations, so the new 2060 continues continues to to be a key key element element in our plan. our research plan. In addition. benefits of addition, we have long appreciated appreciated the the benefits of GUEST GUEST and network network access access to the progrs.ms programs we are developing. SUMEX greatly enhances our ability to obtain feedback developing. SUMEX greatly our ability obtain feedback from physicians and computer programs from interested interested physicians computer scientists scientists around around the country. country. As our our programs continue cant inue to t,o mature, mat,ure, it it will will become increasingly increasingly important important that that we be able to make them them available by distant ava.ilable for for demonstration demonstration and for for access access by distant collaborators collaborators via via the SUrvfEX SUhlEX network. ne twnrk. c. Requirements for for Additional Additional Computing Resources C. Requirement8 Computing Resources The provide a suitable The mainframe ma.inframe machine machine should should continue continue to to provide suitable environment environment for for most plans to NESTOR, or We have no plans to transfer transfer NESTOR, or most of of our our work work in the the months months ahead. We AI/~'IM AI/MM to other other hardware hardware soon. D. Recommwdation,'J for for Future Future Community Resource Development Development D. Recommendations Community and and Resource \Ve are very provided since the We very satisfied satisfied with with the facilities facilities SUMEX SUMEX has provided the upgrade upgrade to the professional workstations t,he DEC DEC 2050. 2060. Other Other than than continued continued acquisition acquisition of of professional workstations that that can be shared by some of programs in this projects, we have no requests shared by of the more more mature mature progra.ms this set of of projects, for for additional addit,ional acquisitions acquisitions or or resource development development at this this time. time. 77 77 E. A. A. Feigenbaum Feigenbaum GUIDON/NEOMYCIN Project GUIDON/NEOhIYCIN Project 5P41 5P41 RR00785-11 II.A.1.2. GUIDON/NEOMYCIN Project II.A.1.2. GUIDON/NEOMYCIN Project GUIDON/NEOMYCIN Project GUIDON/NEOMYCIN Project William J. Ph.D. William J. Clancey, Clancey, Ph.D. Department Computer Science Department Computer Science Stanford University Stanford University Bruce Ph.D. Bruce G. G. Buchanan, Buchanan, Ph.D. Computer Science Computer Science Department Department Stanford University Stanford University I. I. SUM?vlARY OF PROGRAM SUMMARY OF RESEARCH RESEARCH PROGRAM A. A. Project Project Rationale Rationale The Project The GUIDON/NEOMYCIN GUIDON/NEOMYCIN Project is a research program program devoted devoted to the development development, of of a knowledge-based knowledge-based tutoring tutoring system system for for application application to medicine. medicine. This This work work derived That research gave way way to three three derived from from our our first first system, system, the MYCIN MYCIN program. program. That sub-projects sub-projects (EMYCIN, (EMYCIN, GUIDON, GUIDON, and ONCOCIN) ONCOCIN) described described in previous previous annual annual reports. reports. El\l)"CIN EMYCIN has haz been completed completed and its resources reallocated reallocated to other other projects. projects. GUIDON GUIDON and ONCOCIN ONCOCIN have become projects projects in their their own own right. right. The project is to program that The key key issue issue for for the GUIDON/NEOMYCIN GUIDON/NEOMYCIN project to develop develop a program that can provide provide advice by human advice similar similar in quality quality to that that given given by human experts, experts, modeling modeling how how they they structure problem solving procedures. The struct,ure their their knowledge knowledge as as well well as as their their problem solving procedures. The consultation consultation program using this knowledge is called NEOMYCIN. NEOMYCIN's knowledge base, program using this knowledge called NEOMYCIN. NEOMYCIN’s knowledge base, designed for use in a teaching application, will become the subject material used by a for use teaching application, will the subject material by family programs referred The problemfamily of of instructional instructional programs referred to to collectively collectively as GUIDON2. GUIDON2. The problemsolving procedures are developed by running NEOMYCIN and solving procedures developed by running test test cases cases through through NEOMYCIN comparing behavior. Also, NEOMYCIN as Also, we are using NEOMYCIN as a test test bed for for the the compa.ring them them to t,o expert expert behavior. explanation part of programs. explana.tion capabilities capabilities that that will will eventually eventually be part of our our instructional instructional programs. The purpose purpose of of the the current current contract, contract, now now in its sixth of of six six years, years, is to to construct construct an The its sixth intelligent tut,oring tutoring system system t.hat that teaches diagnostic diagnostic strategies strategies explicitSly. explicitly. By By strategy, strategy, we intelligent plans for for establishing establishing a set of of possible diagnoses, focusing focusing on and and confirming confirming mean plans individual diagnoses, gat.hering gathering data, data, and and processing processing new data. data. The The tut,orial tutorial program program will will individual capabilities to to recognize recognize these plans, as well well as to to articulate articulate strategies strategies in explanat,ions explanations have capabilities about how how to do diagnosis. diagnosis. The The strategies strategies represented represented in the program, program, modeling modeling a.bout techniqlles, and explanat.ion techniques are wholly separate from the knowledge base, so techniques, explanation techniques wholly separate from the knowledge base, can be used wit.h with many many medical medical (and (and non-medical) non-medical) domains. domains. That That is, the target target program program will be able t.o to be tested tested with with other other knowledge knowledge bases, bases, using using system-building system-building tools tools that that we will provide. provide. B. Medical Jvledical Relevance Relevance and and Collaboration Collaboration B. There is a growing growing realization realization that that medical medical knowledge, knowledge, originally originally codified codified for for the the There purpose of of computer-based computer-based consultations, consultations, may may be utilized utilized in additional additional ways ways that that are purpose medically relevant. relevant. Using Using the the knowledge knowledge to to teach teach medical medical students students is perhaps perhaps foremost foremost medically among t,hese. these, and NEOMYCIN NEOMYCIN continues continues to to focus on methods methods for for augmenting augmenting clinical clinical among knowledge in order order to to facilitate facilitate its its use in in a tutorial tutorial setting. setting. A A particularly particularly important important knowledge A. Feigenba.um Feigenbaum E. A. 78 GUIDON/NEOMYCIN Project Project GUIDON/NEOMYCIN 5P41 RR00785-11 RR00785-11 5Pll aspect of of this this work work is t,he insight insight that that has been gained gained regarding regarding the the need to to structure structure aspect knowledge differently, differently, and and in in more more detail, detail, when when itit is being being used for for different different purposes purposes knowledge (e.g., teaching teaching as opposed opposed to to clinical clinical decision decision making). making). ItIt was this this aspect aspect of of the the GUIDON GUIDON (e.g.3 resea.rch that t.hat led to to the the development development of of NEOMYCIN, NEOMYCIN, which which is an evolving evolving computational compu~ational research model of of medical medical diagnostic diagnostic reasoning reasoning that that we hope will will enable us to to better better understand understand model teach diagnosis diagnosis to to students. students. An An important important additional additional realization realization is that that these and teach st.ructuring methods methods are beneficial beneficial for for improving improving the the problem-solving problem-solving performance performance of of st,ructuring consultation progra,ms, progra.ms, providing providing more more detailed detailed and and abstract abstract explanations explanations to to consultation consultation consultation users, and and making making knowledge knowledge bases easier to to maintain. maintain. As we move move from from technological technological development development of of explanation explanation and and student student modeling modeling As the next next year year begin begin to to collaborate collaborate more more closely closely with with the the medical medical capabilities, we will will in the capabilities, communit.y to to design an effective, effective, useful useful tutoring tutoring program. program. Stanford Stanford Medical Medical School School community Dr. Maffly, Maffly, have shown shown considerable considerable interest interest in in this this project. project. A A research research fa.culty, such as Dr. fa.culty, fellow associated with with Maffly, Maffly, Curt Curt Kapsner, Kapsner, MD, MD, joined joined the the project project last last year year to to serve as fellow expert and liaison liaison with with medical medical students students at at Stanford. Stanford. medical expert medica. C. High.lights High.lights of of Research Research Progress Progress C. C.l Accomplish.ments This This Pad Past Ear rear C.l Accomplishments C.1.i Th.e NEOMYCIN NEOlwYCIN Consultation Consultation Program Program C.l.1 Th,e NEOMYCIN is distinguished distinguished from from other other AI AI consultation consultation programs programs by by its its uses uses of of an NEOMYCIN explicit set of of domain-independent domain-independent meta-rules meta-rules for for controlling controIling all all reasoning. reasoning. These rules rules explicit constitute the diagnostic procedure that we want to teach to students: the stages of constitute the diagnostic procedure that want to teach to students: the of diagnosis. how to focus on new hypotheses, and how to evaluate hypotheses. It has been a diagnosis, how hypotheses, and how to evaluate hypotheses. It major undertaking, separa.te problem of major undertaking, separate from from the the problem of representing representing disease disease knowledge, knowledge, to to design and test procedure. Such modifications test this t,his diagnostic diagnostic procedure. modifications require require changing changing our our conception partitioned For example, example, this this year year we partitioned conception of of how disease disease knowledge knowledge is organized. organized. For disease and "red disease findings findings into into "non-specific" “non-specific” “red flag" flag” (those requiring requiring explanation), explanation), augmenting procedure to augmenting the diagnostic diagnostic procedure to use this this information information for for focusing focusing on hypotheses. hypotheses. A program reason about A second cha.nge change is to to have the program about the disease disease process more more generally. generally. By by organ NEOMYCIN now primitive means to By associating associating symptoms symptoms by organ system, system, NEOMYCIN now has prirnitive to infer infer when a disease disease process began. It It also makes more more complete complete use use of of severity, severity, location, location, and progression information progression information to to discriminate discriminate among among hypotheses. hypotheses. During During this this past past year, year, we completely completely reworked reworked the program's program’s knowledge knowledge of of nonmeningitis cases. This is important if we wish to teach students to consider meningitis cases. This important if wish teach students consider the competitors of meningitis and how to discriminate among them. The goal is competitors meningitis discriminate among them. The is to prepare prepare the program order to test test the program for for presenting presenting these (or (or similar) similar) cases cases to students. students. In order modeling modeling component. component, it it is necessary to to ensure that that the program program has sufficient sufficient expertise expertise to recognize good student student beha.vior. behavior. All All data data that that might might be relevant relevant to solving solving a given given problem problem must must be be known known to the program. program. The The key key problem problem here is establishing establishing a base base of synonyms synonyms and knowledge knowledge about about classes classes of data. data. To To do this, this, we have been collecting collect,ing protocols protocols of student.s students solving solving problems, problems, requiring requiring them them to request request all by simple simple initial initial case case information. information. Student Student behavior behavior also suggests disease disease knowledge knowledge that that must must be be added to the knowledge knowledge base base that, that an expert expert might might not not consider, consider, but but which which the modeling modeling program program must must recognize in a student. In general, we find find that that students students carry carry out out a much much broader, broader, student. inefficient inefficient search, requesting requesting much much more information information than than an expert expert and a.nd drawing drawing fewer from the information that they receive. conclusions conclusions from t,he information that they The Teaching diagnosis involves involves recoglllzmg recognizing The Image Ima.ge Student Student Modeling Modeling Program Program -- Teaching the intent int,ent behind behind a student's student’s behavior, behavior, so that that missing missing knowledge knowledge can be distinguished distinguished from from inappropriate inappropriat,e strategies. strategies. The The teacher teacher interprets interprets behavior, behavior, critiques critiques it, it, and provides provides 79 E. A. Feigenbaum Feigenbaum 5P41 5P41 RR00785-11 RR00785-11 GUIDON/NEOMYCIN Project GUIDON/NEOMYCIN Project advice To do do this this successfully successfully and and efficiently efficiently in in aa complex complex advice about about other other approaches. approaches. To domain, the teacher benefits from multiple, complementary modeling strategies. domain, the teacher benefit,s from multiple, complementary modeling strategies. IMAGE IMAGE isis aa. student student modeling modeling program program that that uses uses NEOMYCIN's NEOMYCIN’s meta-rules meta-rules and and disease disease knowledge knowledge to to understand understand student student diagnostic diagnostic plans. plans. A student student isis presented presented with with aa problem problem to diagnose. diagnose. As the student student requests more problem problem data data (Le., (i.e., takes aa history history and and physical physical of of the patient), patient), IMAGE IMAGE looks looks for for regularities regularities in in sequences sequences of of his data data requests. IMAGE IMAGE contains contains aa body of of knowledge knowledge about about how to map such such sequences sequences of of behavior behavior onto onto aa strategical strategical interpretation interpretation of of what what the student student isis doing. The process isis heuristic heuristic in in nature nature because because the program program will will sometimes sometimes lose lose track track of what what the student student is is doing, because because he he is is being inconsistent inconsistent or using unexpected unexpected strategies. strategies. The I~!AGE IMAGE uses uses aa dual search strategy. strategy. The The program program first first produces produces multiple multiple predictions Focused, predictions of student student behavior behavior by aa model-driven model-driven simulation simulation of NEOMYCIN. NEOMYCIN. data-driven By By supplementing supplementing each other, other, these dst)a-driven searches then then explain explain incongruities. incongruities. met.hods lead to an efficient and robust plan understander. methods efficient robust plan understander. A model of student student strategies strategies in medical medical diagnosis must must disambiguate disambiguate the possible purposes and knowledge knowledge underlying underlying the student's student’s actions. actions. The The approaches approaches followed followed by other other plan p1a.nrecognizers recognizers and student student modelers modelers are not not sufficient sufficient here because: because: 1. 1. t.he t,he complex complex domain domain makes thorough thorough searches impractical, impractical, whether whether top-down top-down or bott.om-up; or bottom-up; 2. 2. we are not not modeling modeling only only facts and rules used in isolation, isolation, but but also the procedures for applying them; procedures for applying them; 3. everyone every one of of the student's student’s actions actions must must be monitored monitored in case case the teaching teaching module module decides to to interrupt; interrupt; 4. his behavior behavior must just explained; must be evaluated evaluated and not not just explained; and might not not have any any explicit explicit goal statements statements from from the student, student, so we expect expect 5. we might to rely rely only only on his queries for for problem problem data data as evidence for for his thinking. thinking. t,o The IMAGE IMAGE program program is a prototype prototype system system which which is now now being being extended. extended. Specifically, Specifically, a The more useful useful system system would would examine examine its its own own interpretations interpretations and and strive strive for for coherence. We \Ve more designing such a such a system system now, now, using using the the “blackboard "blackboard model” model" for for postding posting are designing interpretations that that. may may change over over time. time. The The levels of of this this blackboard blackboard are: 1) the the interpretations stUdent's data requests, requests, 2) a classification classification of of question question type type (e.g., triggered, triggered, follow-up, follow-up, student’ s data hypothesis-directed, general), general), 3) a strategic strategic interpretation interpretation in terms terms of of NEOMYCIN’ NEOMYCIN's hypothesis-directed, s diagnostic procedure procedure (ta.sks (tasks and and meta-rules). meta-rules). By incorporating incorporating a strategic strategic level level of of diagnostic By interpretation, this this program program can be expected expected to to make make significant significant contributions contributions to to our our interpretation, understanding and and use of of the the blackboard blackboard model model of of interpretation. interpretation. The The first first version version of of this this understanding program will will seek t,o to explain explain student student behavior behavior in in terms terms of of deletion deletion and and reordering reordering of of progra,m simple variations of disease knowledge (e.g., false proC'edural knowledge, plus procedural knowledge, plus simple variations of knowledge (e.g., data/hypothesis relations). relations). Study Study of of student student protocols protocols is now now suggesting suggesting what what kinds kinds of of dats.jhypothesis variations are common common that that we might might easily easily identify identify automatically. automatically. variations C.l.2 The The NEOMl%IN NEOMYCIN Ezplanation Explanation System System C.1.2 The initial initial explanation explanation system system of of NEOMYCIN, NEOMYCIN, now now completed, completed, enables the the user user t,o to The answer WHY \VHY and and HOW HOW questions questions during during a consultation. consultation. That That is, when when the the program program answer prompts the the user user for for new new data, data, the the user user may may ask ask WHY \VHY the the data data is is being being requested requested or or prompts HO\V some st,rategic strategic task task will will be be (or (or was) was) accomplished. accomplished. Unlike Unlike MYC’ MYCIN's explanation HO\V IN’s explanation E. A. A. Feigenbnum Feigenbaum E. 80 80 GUIDON/NEOMYCIN Project GUIDON/NEOMYCIN Project 5P41 RR00785-11 RR00785-11 system, NEOMYCIN are system, upon upon which which this this kind kind of of capability capability is patterned, patterned, explanations explanations in NEOMYCIN in terms plan, not just specific terms of of the diagnostic diagnostic plan, not just specific associations associations between between data data and diagnoses. The program can provide provide abstract The program abstract and concrete concrete paraphrases paraphrases of of strategy strategy rules (based on canned text). We have begun the next next phase, which which is to to answer answer WHY WHY questions questions by by text,). \Ve condensing condensing the entire entire line of of reasoning. reasoning. The The program program will will use use models of of the user's user’s disease disease and strategic strategic knowledge, knowledge, plus general general explanation explanation heuristics, heuristics, to select the task task and focus information Prototypic information that that is most most likely likely to to be of of interest. interest. Prototypic user models are now now implemented. Heuristics Heuristics have been designed and include: include: 1) mentioning mentioning the last last task task implemented. whose focus (or (or argument) argument) changed changed in kind kind (e.g., from from a disease disease hypothesis hypothesis to to a finding finding request): request): 2) never never mentioning mentioning tasks tasks that that are merely merely iterating iterating over over a list list of of rules, findings, findings, or or hypotheses; hypot,heses; and 3) only only mentioning mentioning tasks tasks with with a rule rule as as an argument argument to to programmers. programmers. Related Related to t,o our our explanation explanation condensations condensations is an effort effort to to teach the strategic strategic language of For For example, example, we will will have students students annotate annotate a of tasks tasks to to students. students. NEO!"IYCIN typescript NEOh,lYCIN typescript in terms terms of of tasks tasks and foci, to to help them them recognize good strategic strategic behavior. behavior. This This requires requires a common common language of of what what the tasks are, e.g., "grouping" “grouping” and "asking Rather just marking Rather than than just marking annotating annotating tasks, we seek the “asking general general question." question.” principles by principles by which which the tasks tasks could could be consistently consistently structured structured into into primitives primitives and auxiliary. These same principles auxiliary. principles could could be used by by the the explanation explanation system system for for choosing choosing tasks to mention. mention. Our Our current current theory theory is that that these primitive primitive or or "interesting" “interesting” operations operations correspond correspond to meta-rules meta-rules that that establish establish a new focus. C.LI for Teaching C.l.4 Graphics Graphics for Teaching vVe We are continuing continuing make make extensive extensive use use of of graphics graphics in our our programs. programs. For For example, example, we are implementing program that implementing a program that will will mostly mostly automatize automatize the protocol protocol collection collection process (though (though we are cautious cautious about about how how menus will will bias student student behavior, behavior, even when when lists lists are very very long and full full of of irrelevant irrelevant findings). findings). As part part of of our our series of of instructional instructional programs, GUIDON-WATCH is now being progra.ms, GUIDON-WATCH being implemented implemented as as a graphic graphic system system for for watching watching NEOMYCIN's reasoning. For NEOMYCIN’ s reasoning. For example, example, we can highlight highlight the the hypotheses hypotheses under under consideration consideration and show graphically graphically how how the program program "looks “looks up" up” its its hierarchies hierarchies before before refining is also mostly refining hypotheses. hypotheses. The The design or of GUIDON-ANNOTATE GUIDON-ANNOTATE mostly complete. complete. It It will NEOMYCIN's will allow allow a student student to to mark mark up a typescript typescript of of NEOMYCIN’ s behavior behavior using using the same language of of tasks tssks the t.he program program uses uses when when explaining explaining its own own behavior; behavior; iconic iconic menus are very very usef1ll useful to to avoid avoid natural natural language difficulties difficulties (though (though it it is clear clear that that the student student will will sometimes somet,imes need to to "talk “talk back"). back”). C.2 Research in Progress C.2 Research in Progress The The following following projects projects are active active as as of of June 1983 (see (see also near-term near-term plans plans listed listed in Section Section III.A): 111.‘~): 1. NEOMYCIN's disease 1. augmenting augmenting NEOhlYCIN’ s disease knowledge knowledge so we can fairly fairly evaluate evaluate the program's program’s focussing strategies strategies and evaluate evaluate IMAGE; IMAGE; 2. developing produce developing capability capability to to automatically automatically produce summary summary explanations explanations of of NEOMYCIN's NEOMYCIN’ s reasoning. reasoning. 3. development and GUIDON-ANNOTATE for development of of GUIDON-WATCH GUIDON-WATCH GUIDON-ANNOTATE for teaching teaching NEOMYCIN's NEOMYCIN’ s knowledge knowledge to to students. students. 4. developing developing new student student modeling modeling program program based on the blackboard blackboard model. 81 E. A. A. Feigenbaum Feigenbaum 5P41 5P41 RR00785-11 GUIDON/NEOMYCIN Project GUIDON/NEOMYCIN Project D. D. Publications Publications Since January January 1989 198.3 D., Clancey, Clancey, W.J., W.J., Rennels, G.: Strategic Strategic Consultation. C’onsultation. Int Int JJ Man-Machine Man-Machine Studies, in press. 1. 1. Hasting, Hasling, explanations explanations an in 2. 2. Clancey, Clancey, W.J.: W.J.: The advantages advantages of of abstract abstract control control knowledge knowledge in in expert expert 8y8tem system de8ign. design. Proceedings Proceedings of AAAI-83, AAAI-83, pages pages 74-78. 3. Clancey, Clancey, \V.J.: W.J.: Acquiring, Acquiring, representing, representing, and and evaluating evaluating a competence competence model model of OF of diagno.'li8. diagnosis. In Chi, Chi, Glaser, Glaser, and Fan Farr (Eds.), (Eds.), THE THE NATURE NATURE OF EXPERTISE. EXPERTISE. In preparation. preparation. HPP-84-2. HPP-84-2. 4. IN 4. Clancey, Clancey, W.J. W.J. and E. H. Shortliffe.:READINGS Shortliffe.:RELaINGS IN !vIEDICAL MEDICAL ARTIFICIAL ARTIFICIAL Il'VTELLIGENCE: THE Reading: IKTELLIGENCE: THE FIRST FIRST DECADE. DECADE. Reading: Addison-Wesley, Addison-Wesley, in press. 5. Clancey, Clancey, W.J.: W.J.: AAAI-84. AAAI-84. Classification Problem Classification Problem Solving. Soluing. HPP-84-7. HPP-84-7. Submitted Submitted to E. Funding Funding Support Support Cont,ract. Contract Tit.le: Title: "Explorat.ion “Exploration of of Tutoring Tutoring and Problem-Solving Problem-Solving Strategies" Strategies” Principal Principal Investigator: Investigator: Bruce Bruce G. Buchanan, Buchanan, Adjunct Adjunct Prof. Prof. Computer Computer Science Associate Associate Investigator: Investigator: William William J. Clancey, Clancey, Research Assoc. Computer Computer Science Agency: Agency: Office Office of Naval Naval Research and Army Army Research Institute Institute (joint) (joint) ID ID number: number: NOOOI4-79-C-0302 N00014-79-C-0302 Term: Term: March March 1979 to to March March 1985 Tot,al Tot(a.1award: a.ward: $583,892 $083,892 II. WITH THE RESOURCE II. INTERACTIONS INTERACTIONS WITH THE SUMEX-AIM SUMEX-AIM RESOURCE A. Medical Medical Collaborations Program Dissemination Dissemination via A. Collaborations and and Program uia SUAfEX SUMEX A NEOMYCIN has been shown by the ‘4 great great deal of of iuterest interest in GUIDON GUIDON and NEOMYCIN shown by medical mcdica.1 and a.nd computer computer science communities. communit,ies. We We are frequently frequently asked to to demonstrak demonstrate these programs programs to t,o Stanford St,anford visitors visitors or or at meetings meetings in this this cOllDtry country or or abroad. abroad. GUIDON GUIDON is available on the the SUMEX SUMEX 2020. Physicians Physicians have generally generally been enthusiastic enthusiastic about about these ava.ila.ble programs' potential potential and what what they they reveal reveal about about current current approaches approaches to to computer-based computer-based programs medical decision decision making. making. medical Perhaps our our most most significant significant project project to to disseminate disseminate our our research research via via SUMEX SUMEX in l.he the Perhaps past year year has been the completion completion of of a book, book, “Readings "Readings in Medical il,ledical Artificial Artificial Intelligence: Intelligence: past The First First Decade,” Dec:1de," edited edited by by Dr. Dr. Clancey Clancey and Dr. Dr. Shortliffe. Shortliffe. All All of of the significant significant The SUMEX-AI~vl products products of of the the past past decade are described described in this this collection. collection. Each Each cha.pter chapter is SUMES-,41&I preceded by by a. a one-age historical historical introduction. introduction. In addition, addition, opening opening and closing closing chapters chapters preceded by t)he the editors editors survey survey issues issues in the the field field and the the promise promise of of the the future. future. A A complete complete index index by should make make t,he t.he book book of of considerable considerable educational educational value. value. Preparation Preparation of of this this volume volume has should greatly aided aided by by use of of editing editing and and formatting formatting programs programs available available on on SUMEX-AIM. SUMEX-AIM. been greatly Royalties for for the the book, book, beyond beyond production production costs, will will be used to to sponsor sponsor an invited invited lecture lecture Royalties at a major major AI AI nationa. national conference, conference, such as AAAI. AAAI. at As mentioned mentioned disease knowledge knowledge of of disease July 1983). This This in July A. Feigenbaum Feigenbaum E. A. earlier, a physician physician joined joined our our group group this this year year to to help help us develop develop the the earlier, the program program (our (our first first collaborator, collaborator, Tim Tim Beckett, Beckett, MD, MD. died died of of cancer cancer the physician has found found the the convenience convenience of of accessing SUMEX SUMEX from from his physician 82 RR00785-11 5P41 RR00785-11 GUIDON/NEOMYCIN Project GUIDON/NEOMYCIN Project laboratory or or at at home to be extremely extremely important important for for finding finding time time to to test test NEOMYCIN hboratory NEOMYCIN and to communicate communicate with with us by by electronic electronic mail. mail. 83 Feigenbaum E. A. Feigenbaum 5P41 5P41 RR00785-11 RR00785- 1 I GUIDON/NEOMYCIN Project GUIDON/NEOMYCIN Project B. Projects B. Sharing Sharing and and Interaction Interaction with with Other Other SUA/EX-AlA! SUMEX-AIM Projects GUIDON/NEOMYCIN retains GUIDON/NEOMYCIN retains strong strong contact contact with with the the ONCOCIN ONCOCIN project, project, as as both both are siblings of the tvIYCIN parent. These projects regularly share programming expertise are siblings of the MYCIN parent. These projects regularly share programming expertise and and cont.inue continue to to jointly jointly maintain maintain large large utility utility modules modules developed developed for MYCIN. MYCIN. In addition, addition, the the centra.l centra,l SUMEX SUMEX development development group group acts acts as as an an important important clearing clearing house house for solving solving problems problems and and distributing distributing new new methods. methods. C. C. Critique Critique of of Resource Resource Management Management In the winter winter of 1984, 1984, the SUMEX SUMEX staff staff efficiently efficiently and effectively effectively shifted shifted our our operation operation away from from the center center of campus to aa professional professional office office building building adjoining adjoining the medical medica~l center. center. The placement placement and installation installation of LISP LISP workstations workstations proceeded smoothly. smoothly. After After aa.year with with Ed Pattermann Pattermann as as director director of SUMEX, SUMEX, we can report report that that the stability stability and a.nd excellence of the resource we have come to expect has has been completely completely maintained. maintained. Very Very import.ant important to us, us, the RAVEN RAVEN laser printer printer installed installed at our our new site not not only only provides provides excellent-quality output, but as a machine devoted to the Heuristic Programming excellent,-quality out,put, but as a machine devoted Heuristic Programming Project Project has has eliminated eliminated the dela.ys dela.ys we were experiencing experiencing aa year year a.go. a.go. \Vith continuing to experience a few With t,he the shift shift to personal personal machines, we are continuing The difficulties. The greatest. greatest, problem problem appears to be inadequately inadequately debugged software software from from difficulties. XEROX. In particular, SEROX. particular, Interlisp-D Interlisp-D relies heavily heavily on network network capabilities capabilities and must must be compatible compatible with with several operating operating systems. This This transition transition to new kinds kinds of hardware hardware and software softwa.re can be be expected expected to continue continue for for several years. Therefore, Therefore, we are extremely extremely reliant reliant upon upon the availability avai1abilit.y of experienced experienced systems support. support. We We believe believe that that additional additional SU~IEX SUhlEX staff staff is necessary to accommodate accommodate growing growing community community needs. needs ill. RESEARCH PLANS III. RESEARCH PLANS A. Project A. Project Goals Goals and and Plans Plans over the next next year year will will continue continue on several fronts, leading leading to to several several Research over several fronts, prototype instruct,ional instructional programs programs by by early early 1985. protot,ype 1. Continue Continue to to develop develop the knowledge knowledge base so the program program can understand understand and and anticipate any any reasonable reasonable approach approach to to the cases cases chosen for for teaching. teaching. ant.icipate Test st,udent student modeling modeling program program on these cases, cases, collecting collecting data data for for further further 2. Test development of of the program, program, as well well as exploring exploring about about the the range range of of student student development approaches to to diagnosis. diagnosis. approaches Extend the t.he explanation explanation system system to to do full full summaries. summaries. Incorporate Incorporate modeling modeling 3. Extend capabilities that t.hat relate relate inquiries inquiries to to a user model. model. Provide Provide explanations explanations t.ailored tailored capabilit,ies to this this interpretation interpretation of of t,he the motivation motivation behind behind the the user’ user's inquiry. to s inquiry. 4. Int,egra.te Integrate current current display display capabilities capabilities into into running running NEOMYCIN NEOMYCIN consultation consultation to to show how the space of diagnoses is explored and how diagnostic tasks are show how the of explored and how diagnostic tasks generated. Develop Develop these capabilities capabilities to to explore explore forms forms of of graphic graphic explanation explanation generated. useful in tutoring. (GUIDON-WATCH) useful tutoring. (GUIDON-WATCH) S. Extend Extend student student modeling modeling system system to to include include heuristics heuristics for for generating generating tests tests that that S. will confirm confirm and and extend extend the the model. model. Improve Improve the the model model to to include include analysis analYSis of of will pat,tans in model interpretations, including dependency-directed patterns including in model interpretat*ions, dependency-directed backt.racking in in the the belief belief system system and and some capability capability to to critique critique the the ” backtra.cking” mode lint:?; rules. rules. Relate Relate this this to to knowledge knowledge acquisition acquisition research. research. rnodrling II E. A. A, Feigenbaum Feigenbaum E. II 84 84 GUIDON/NEOMYCIN Project GUIDON/NEOMYCIN Project 5P41 5P-11 RR00785-11 6. 6. Work Work closely with with medical medical students students to package NEOMYCIN NEOMYCIN capabilities capabilities in aa "workstation" “workstation” for for learning learning medical medical diagnosis, determining determining what what mix mix of student student and program program initiative initiative isis desirable. desirable. B. Long Programs Long term term plans: plans: the GUIDON2 GUIDON2 Family Family of of Instructional Instructional Programs "Ve We sketch sketch here our our general conception conception of the research we plan plan for for 1984-87, specifically Our specifically the construction construction of instructional instructional systems that that use use NEOMYCIN. NEOMYCIN. Our ideas are strongly strongly based on recent recent proposals proposals by by JS JS Brown, Brown, particularly particularly his paper paper "Process “Process versus Product --A perspective perspective on tools for for communal communal and informal informal electronic electronic learning" learning” and Product, --A some related related papers that that he wrote wrote in 1983. 1983. The The plan plan is is to implement implement at least three three of these programs GUIDON-MANAGE, and GUIDONprograms (here called GUIDON-WATCH, GUIDON-WATCH, GUIDON-MANAGE, GUIDONANNOTATE). ANNOTATE). The key idea is that that NEOMYCIN NEOMYCIN provides provides aa language language by by which which aa program program can converse wit.h a student student about about strategies strategies and knowledge knowledge organization organization for for diagnosis. NEOtvIYCIN's NEOMYCIN’s tasks and structural structural terms terms provide provide the vocabulary vocabulary or or parts parts of speech; the meta-rules We will will construct construct different different meta-rules are the grammar grammar of the diagnostic diagnostic process. We graphic, graphic, reactive reactive environments environments in which which the student student can observe, describe, compare, compare, and improve improve diagnost,ic behavior behavior of of himself himself and others. others. There There are many many shared, underlying underlying ca.pabilities ca.pa.bi!it.ies that that wil! will be constructed constructed in parallel parallel and improved improved over over time. time. Our Our approach approach is is to delineate delineate clearly clearly different different kinds kinds of of interactions interactions that that a student st.udent might. might, have with wit,h a program pr0gra.m concerning concerning diagnostic diagnostic strategies. strategies. Thus, Thus, each instructional instructional system activity>, where syst.em (but (but one) has a name of the form form GUIDON- , where the name specifies what what the student student is doing doing (e.g., watching, watching, telling). telling). The The programs programs can be made arbitrarily by integrating arbitrarily complex complex by integrating coaches, coaches, student student models, and explanation explanation systems. "Ve We try try here to to separate separate out out these capabilities, capabilities, trying trying to to get at at the the minimum minimum interesting interesting activities provide for activities we might might provide for a student. student. GUIDON-WATCH The NEOMYCIN GUIDON-VVATCH The simplest simplest system system allows allows a student student to watch watch NEOMYCIN solve a.a. problem, problem. perhaps perhaps one supplied by the student. supplied by student. Graphics Graphics display display the evolving evolving search space, that that is, how how tasks, as as operators, operators, affect affect the differential differential (Differential (Differential ---(Question The ---(Quest ion X)--S)--- > Differential'). Differential’). The student student can step through through slowly slowly and replay replay the interaction. prosaic explanations program is intcract.ion. He can ask for for prosaic explanations and summaries summaries of of what what the program doing. program will The program will also indicate indicate its its task task and focus for for each data data request. request. This This doing. The introduces introduces the t.he student student to the idea that that the diagnostic diagnostic process has structllre structure and follows follows a certain kind kind of of logic. certain GUIDON-MANAGE In In this this system system the the student student solves a problem problem by by telling telling GUIDON-MANAGE NEO!\IYCIN what task task to to do at each step. Essentially, Essentially, the the student student provides provides the strategy strategy NEOXIY’ C’IN what and the program program supplies supplies the the tactics tactics (meta-rules) (meta-rules) and and domain domain knowledge knowledge to carry carry out out the a.nd strategy. The The program program will will in genera! general carry carry through through tasks tasks in in a logical logical way, way, for for example, example, strst.egy. proCt'cdiug to to t,est tcst a hypothesis hypothesis completely, completely, and not not “breaking” "breaking" on FINDOUT FINDOUT or or proceeding APPLYHULES (t.wo low-level low-level tasks tasks that that mainly mainly test test domain domain knowledge knowledge and not not strategy). strategy). APPLI’ RIJLES (two ThE' program program will will not not pursue pursue new hypotheses hypotheses automatically. automatically. However, However, the the student student will will The alwa.ys see see wha.t what questions questions a task task caused the the program program to to request, request, as well well as how how the alwa,ys differential changes. This activity activity leads the the student student to to observe the entailments entailments of of differential This strategies, helping helping him him become a better better observer observer of of his own own behavior. behavior. Here he shows that that strategies, knows the structural structural vocabulary vocabulary that that makes makes a strategy strategy appropriate. appropriate. he knows GUIDON-ANNOTATE This This system system allows allows the the student student to to annotate annotate a NEOMYCIN NEOMYCIN GUIDON-ANNOTATE typescript, indicating indicating the the task task and and focus associated with 'with each data data request. request. The The program program typescript, will indicate, indicate, upon upon request, request, where where the the student student is incorrect incorrect and and which which annotations annotations are will different from from NEOMYCIN’ NEOMYCIN's, but still still reasonable reasonable interpretations. interpretations. The The student student will will be different s, but 85 A. Feigenbaum Feigenbaum E. A. 5P41 RR00785-11 RR00785- 11 GUIDON/NEOMYCIN Project GUIDON/NEOMYCIN Project able to to- choose these tasks from from a menu menu of of icons, either either linearly linearly or or hierarchically hierarchically displayed, displayed, as NEOMYCIN will as he prefers. prefers. (Again, (Again, NEOMYCIN will annotate annotate its its own own solutions solutions upon upon request request and allow allow replaying.) This replaying.) This activity activity gets the the student student to to think think strategically strategically by by recognizing recognizing a good st.rategy. st,rategy. In In this t,his way, way, he learns to to recognize how how strategies strategies affect affect the the problem problem space. GUIDON-APPRENTICE This NEOMYCIN in which GUIDON-APPRENTICE This is a variant variant of of NEOMYCIN which the program program stops during during a consultation consultation and asks the student student to propose the the next next data data request(s). request(s). The The student student is asked to to indicate indicate the task task and focus he has in mind, mind, plus plus the differential differential he is operating NEOMYCIN would operating upon. The The program program compares compares this this proposal proposal to to what what NEOMYCIN would do. In In this this activit.y activit,y we descend to to the domain domain level and require require the student student to instantiate instant,iate a st.rategy st,rategy appropriately. a.ppropriately. GUIDON-DEBUG Here the student GUIDON-DEBUG student is presented presented with with a buggy buggy version version of of NEOM'(CIN goes through through the steps of of annotating annotating the buggy buggy NEOMY’ CIN and must must debug it. it. He goes consultation consultation session, session, indicating indicating what what questions questions are out out of of order order or or unnecessary, unnecessary, indicating indicating what properly, and then what tasks are not not being being invoked invoked properly, then trying trying out out his hypothesis hypothesis on a "rep:J.ired" predict what what will will be different, d.ifferent, then then allowed allowed to to observe “repaired” system. system. He is asked to predict what what happens. This This activity activity teaches the student student to recognize how how a diagnostic diagnostic solution solut,ion can be non-optimal, provides him non-optimal, further further emphasizing emphasizing the value value of of good strategy. strategy. It It also provides him with practice for problem behavior. with key key meta-cognitive meta-cognitive practice for criticizing criticizing and debugging debugging problem behavior. GUIDON-SOLVE This The The student student carries carries GUIDON-SOLVE This is the complete complete tutorial tutorial system. system. through plan recognizer through diagnosis diagnosis completely, completely, while while a plan recognizer attempts attempts to to track track what what he is doing doing and a coach interrupts interrupts to offer offer advice. Here annotation, annotation, comparison, comparison, debugging, debugging, and ('xplana.tion expiana.tion are all integrated int)egrated to illustrate illustrate to to the student student how his solution solution is non-optimal. non-optimal. For For example, example, the student student might might be asked to annotate annotate his solution solution after after he is done; this this will point out will point, out strategic strategic gaps in his awareness and provide provide a basis for for critique critique and improvement. A "curriculum" “curriculum” based on rrequent frequent student student faults faults and important important things things to improvement. A learn In learn will will drive drive the interaction. interaction. In this this activity, activity, the student student is on his own. own. Faced with with the must exercise his diagnostic diagnostic procedure procedure from from start start to to t.he proverbial proverbial "blank “blank screen," screen,” he must finish. finish. GUIDON-GAME Two GUIDON-GAME Two or or more more students students play play this this together together on a single machine. machine. They All They are given given a case case to solve together, together, and each student student requests data data in turn. turn. All st.lldents When students receive the requested requested information. information. When a student student is ready, ready, he makes a di3gnosis, diagnosis, indicated indicat)ed secretly secretly to to the program program while while the others others are not not watching. watching. He then then drops drops out out of of the questioning questioning sequence. sequence. However, However, he can re-enter re-enter later, later, but but of of course will will be penalized. Afterwards, score is based on the number number of of questions questions asked and use use of of penalized. Afterwards, good strategy. strategy. The The coach will will indicate indicate to to weak weak players players what what they they could could learn learn from from strong strong players, one players. encouraging encouraging them them to to discuss certain certain issues issues among among themselves. themselves. Variation: Variation: person solves while while one or or more more competing competing students students annotate annotate the solution solution and show where where it Variation: NEOMYCIN (and Variation: one team team introduces introduces a bug bug into into NEOMYCIN (and it, could could be improved. improved. predicts This predicts the effect) effect) and the the other other team team finds finds it it (as in SOPHIE). SOPHIE). This activity activity will will encourage students students to share their their experiences and talk talk to and learn learn from from each other other about about the diagnostic disgnostic process. C. Requirements Reqw:rements for for Continued Continued SUMEX SUMEX Use Although NEOMYCIN work Although most most of of the the GUIDON GUIDON and NEOMYCIN work is shifting shifting to to Xerox Xerox Dolphins Dolphins and Dandelions Dandelions (D-machines), (D-machines), the DEC DEC 2050 2060 and 2020 continue continue to to be key key elements elements in our our research plan. plan. Our Our primary primary use use of of the 2050 2060 will will be to develop develop the NEOMYCIN conSUltation access. NEOMYCIN consultation system, system, possibly possibly by by remote remote ARPANET ARPANET access. Because of of address a.ddress space limitations, limitations, the consultation consultation program program can be combined combined with with explanation explanation or or student student modeling modeling facilities, facilities, but but not not both, both, as is required required for for GUIDON2 GUIDON2 programs. programs. We We continue continue to use use the 2020 for for demonstrating demonstrating the original original GUIDON GUIDON program. program. As always, always, the 2050 2060 will will be essential essential for for work work at home, writing, writing, and electronic electronic mail. mail. E. A. Feigenbaum Feigenbaum 85 86 5P41 5P4 1 RROOi85-11 RR00785- 11 GUIDON/NEOMYCIN Project GUIDON/NEOMYCIN Project D. Requirements Requirements for for Additional Additional Computing Resources D. Computing Resources The D-machine's D-machine’s large address space is permitting permitting development development of of the large The program that that complex complex computer-aided computer-aided instruction instruction requires. requires. Graphics Graphics will will enable us to program methods for for presenting presenting material material to naive naive users. We We also plan plan to use use the Ddevelop new methods mn.chine as as a reliable, reliable, constant constant "load-average" “load-average” machine, machine machine, for for running running experiments experiments with with physicians and students. students. The development development of of GUIDON2 GUIDON2 on the the D-machine D-machine will will physicians The demonstrate the the feasibility feasibility of of running running intelligent demonstrate intelligent consultation consultation or or tutoring tutoring systems on sma.11,affordable affordable machines machines in physicians' physicians’ offices, offices, schools and other other remote remot,e sites. small, We cnrrently currentSly have access access to to 1 1/2 l/2 DOLPHINs. DOLPHINS. We We We expect that that 3 full full time time programmers will will need access access to to two two full full machines. machines. vVe We are keeping keeping logs so we can ca.n begin begin programmers understand patterns patterns of of ::t.ctivity activity and how how these "personal" “personal” machines machines can be effectively effectively to understand shared. shared. 87 E. A. Feigenbaum Feigenbaum GUIDON/NEOMYCIN Project GUIDON/NEOMYCIN Project RR00785-11 5P41 RR00785-11 E. Community and and Resource E. Recommendations Recommendations for for Future Future Community Resource Development Development shift our our development development of of systems systems to to personal personal LISP LISP machines, machines, such as as the As we shift DOLPHIN, it it becomes more more difficult difficult to to access access these programs programs remotely remotely for for access access from from DOLPHIN, our our homes (so that that we may ma,y work work conveniently conveniently during during the evenings and weekends) weekends) and from remote remote sites for for collaboration collaboration and demonstration. demonstration. from This This problem problem will will be partly partly ameliorated by by "dial-up" “dial-up” (modem) (modem) access access to these machines, machines, but but the use use of of bit-mapped bit-mapped ameliorated displays requiring requiring a high-bandwidth high-bandwidth makes the the phone lines inadequate inadequate for for our our purposes. displays Further Further technological technological development development of of networks, networks, probably probably involving involving access access over over cables, will will be necessary. computer resources become more distributed, distributed, the need for for a central central machine machine As computer does not not diminish. diminish. Programs Programs and knowledge knowledge bases bases continue continue to be shared, shared, requiring requiring highhighdoes speed network network connections connections among among computers computers and file servers. SUMEX-AIM's speed SUMEX-AIM’s role will will shiftly slightly slightly over over the the next next few years years to accommodate accommodate these needs, needs, but but its its identity identity as as a shiftly central will only only change in kind, kind, not not importance. importance. cent ral resource will Moreover, Moreover, sophisticated sophisticated printing devices, such as as the the Xerox Xerox RAVEN, RAVEN, must must necessarily necessarily be shared, again using a printing network. Maintenance Maint,enance of of this this network network and its its shared shared devices will will become a key key activity activity network. for for the SUMEX SUMEX staff. staff. Thus, Thus, while while computing computing resources will will be provided provided by by the the "outboard “outboard engines” of of personal personal machines, machines, the community community will will remain remain intricately intricately linked and engines" linked dependen dependent,t. on common, common, but but peripheral, peripheral, resources. From From t.his t,his perspective, perspective, future future resource development development should should focus on improving improving the capabilities of of networks, networks, file servers, and attached attached devices to respond respond to individua,l individual ca,pabilities For example, example, it it is now now common common for for 10% 10% of of a user's user’s time time at a personal personal machine machine requests. For spent waiting waiting for for a file server server or or printer printer to to process a request. Multi-processing to be spent Multi-processing necessity in such an environment, environment, so a request request can be honored, honored, while while the user becomes a necessity returns returns to continue continue his programming programming or or editing. editing. E. A. i4. Feigenbaum Feigenba,um 88 HPP HPP Core Core AI AI Research Research RR00785-11 5P41 RR00785-11 II.A.1.3. HPP HPP Core Core AI AI Research Research II.A.1.3. Heuristic Heuristic Programming Programming Project Project Principal Investigator: Investigator: Edward Edward A. A. Feigenbaum Feigenbaum Principal C-Principal Co-Principal Investigator: Investigator: Bruce Bruce G. G. Buchanan Buchanan Department Department of of Computer Computer Science Science Stanford Stanford University University SUMMARY OF OF RESEARCH RESEARCH PROGRAM PROGRAM I.I. SUMMARY A.IB. Rationale Rationale and and Medical Afedical Relevance Relevance A./B. Medicine and the the biological biological sciences are knowledge-intensive knowledge-intensive with with an exponential exponential Medicine rate of of growth growth in relevant relevant knowledge. knowledge. This This means that that problem problem solving solving of of all all sorts sorts is rate becoming increasingly increasingly complex complex in these disciplines. disciplines. Further, Further, most most problems problems are symbolic symbolic becoming nature rather rather than than amenable amenable to to mathematical mathematical formulation formulation and numerical numerical solution. solution. in nature Artificia.l Intelligence Intelligence (AI) (AI) methods methods have been focused on medical medical and biological biological problems problems Artificia.1 for over over a decade with with considerable considerable success. success. This This is because, of of all all the the computing computing methods methods for known, AI AI met,hods met.hods are the the only only ones that that deal explicitly explicitly with with symbolic symbolic information information and known, problem solving solving and with with knowledge knowledge that that is heurist,ic heuristic (experiential) (experiential) as well well as factual. factual. problem particularly fast-moving fast-moving area of of AI AI is expert. expert systems. An An expert expert system system is one One particularly performance level rivals rivals that that of of an human human expert expert because itit has extensive extensive domain domain whose performance knowledge (current,ly (currently usual!y usually derived derived from from an human human expert); expert); itit can reason rea..<;on about about its its knowledge knowledge knowledge to solve difficult difficult problems problems in the domain; domain; it it can explain explain its line of of reasoning reasoning much much as as an human human expert expert can; and it it is flexible flexible enough to incorporate incorporate new knowledge knowledge without Expert witholit reprogramming. reprogramming. Expert Systems Systems draw draw on the current current stock stock of of ideas in AI, AI, for for They are adequate adequate for for capturing capturing example, They example, about about representing representing and using using knowledge. knowledge. problem-solving Numerous high-performance, problem-solving expertise expertise for for many many bounded bounded problem problem areas. Numerous high-performance, expert expert, systems systems have resulted resulted from from this this work work in such diverse diverse fields fields as as analytical analytical chemistry, chemistry, medlcal cancer chemotherapy chemotherapy management, management, VLSI VLSI design, machine machine fault fault medical diagnosis, dia,gnosis, cancer of these programs programs rival rival human human experts experts in solving solving dia.gnosis. diagnosis, and molecular molecular biology. biology. Some of problems problems in particular particular domains domains and some are being being adapted adapted for for commercial commercial use. use. Other Other tools for for representing representing and utilizing utilizing projects projects have developed developed generalized generalized software software tools knowledge UNITS, EMYCIN, UNITS, AGE, AGE, MRS, MRS, GLISP) GLISP) as as well well as as comprehensive comprehensive knowledge (e.g., EMYCIN, publications Intelligence publications such as as the three-volume three-volume Handbook Handbook of of Artificial Artificial Intelligence and books summa.rizing summarizing lessons lessons learned learned in the DENDRAL DENDRAL and MYCIN MYCIN research projects. projects. But But the current current ideas fall fall short short in many many ways, necessitating necessitating extensive extensive further further basic research dforts. efforts. Our Our core research goals, as as outlined outlined in the next next section, section, are to analyze the limitations limitations of current current techniques techniques and to investigate investigate the nature nature of methods methods for for overcoming overcoming them. them. Long-term Long-term success success of of computer-based computer-based aids in medicine medicine and biology biology methods available available for for representing representing and using depend on improving improving the programming programming methods domain That That knowledge lcnowledge is inherently inherently complex complex .... it contains contains mixtures mixtures of domain knowledge. knowledge. symbolic symbolic and numeric numeric facts facts and relations, relations, many many of them them uncertain; uncertain; it it contains contains knowledge knowledge at different different levels of of abstraction abstraction and in seemingly seemingly inconsistent inconsistent frameworks; frameworks; and it it links links examples and exception exception clauses clauses with with rules of thumb thumb as as well well as as with with theoretical theoretical principles. principles. Current Current techniques t,echniques have been successful only only insofar insofar as as they they severely limit limit t.his t#his complexity. comp1exit.y. As the applications applications become more far-reaching, far-reaching, computer computer programs programs will will have to deal more effectively effectively with with richer richer expressions and much much more voluminous voluminous amounts amounts of knowledge. knowledge. This This report report documents documents progress on the basic or core research activities activities within within the 89 Feigenbaum E. A. Feigenbaum HPP HPP Core AI AI Research 5P 41 RR00785-11 5P41 Hruristic Heurist,ic Programming Programming Project Project (HPP), (HPP), funded funded in part part under under the SUMEX SUMEX resource as as well as This work work explores aa broad broad range of basic as by other other federal and industrial industrial sources. This research resea,rch ideas ideas in many many application application settings, settings, all of which which contributes contributes in the long term term to improved improved knowledge knowledge based based systems in biomedicine. biomedicine. C. Highlights Highlights of of Research Research Progress Progress In the last year, year, we made progress on several major major topics topics of research. The The style style of research t.hat that we believe is is most most productive productive at this this stage of development development of AI AI is the experimental experimental style. style. Thus, Thus, within within the HPP HPP we build build systems that that implement implement our our idea.s idea.s for for answering answering (or (or shedding shedding some light light on) fundamental fundamental questions; we experiment experiment with with those systems to drtermine determine the strengths strengths and limits limits of the ihe ideas; ideas: we redesign and test more; we attempt attempt to generalize generalize the ideas from from the domain domain of of implementation implementation to other other domains; domains; and we publish order to carry carry out out this this style style of research, then, then, publish details details of the experiments. experiments. In order we select. select specific specific problems problems to help focus the general questions. Many Many of these specific specific problem problem domains domains are medical medical or or biological. biological. In this this way way we believe the HPP HPP has made substantial just to the AIM substantial contributions contributions to core research problems problems of of interest interest not not just AIM community community but but. to AI AI in general. general. Progress Progress is reported reported below below under under each of of the major major topics topics of of our our work. work. are to HPP HPP technical technical reports reports listed listed in the publications publications section. section. Citations Citations How How can the knowledge knowledge necessary for for complex complex problem problem solving solving be represented represented for for its its most most effective effective use use in automatic automatic inference inference processes? processes? Often, Often, the knowledge knowledge obtained obtained from from experts experts is heuristic heuristic knowledge, knowledge, gained How can this this knowledge, knowledge, with with its its gained from from many many years years of of experience. experience. How inherent inherent vagueness and uncertainty, uncertainty, be represented represented and applied? applied? 1. n01lliedge representation: 1. 1-( Knowledge representation: \Vork NEOMYCIN systems VVork on the logic-based logic-ba.sed MRS MRS and the rule-based rule-based NEOMYCIN systems continues, Numerous copies cont.inues, attracting attracting wide wide interest interest within within the AI AI community. community. Numerous of of MRS MRS have been sent sent to collaborators collaborators elsewhere who who are experimenting experimenting with with it book on rule-based by Buchanan it on the own own machines. machines. The The book rule-based expert expert systems by Buchanan 8: & Shortliffe Short,liffe was completed completed in this this year. year. [Se(' [See HPP HPP technical t,echnical memos HPP-83-26, HPP-83-26, HPP-83-28, HPP-83-28, HPP-83-29, HPP-83-29, HPP-83-34, HPP-83-34, HPP-84-1] HPP-84- 1] What What kinds kinds of of software software tools tools and system architectures architectures can be constructed constructed to to make make itit easier to to implement implement expert expert syst,em programs with with increasing increasing complexity complexity and high high performance? performance? How How can we programs flexible control control structures structures for for powerful powerful problem problem solving solving programs? programs? design flexible 2. Advanced Adt1anced architectures architectures and and Control: Control: A major major effort, effort in exploring exploring and understanding understanding the the Blackboard Blackboard a.rchitecture architecture ha.s has A undrrtaken. A A new pilot pilot project project using using this this architecture architecture was st,arted started in been undcrbnken. domain of of protein protein chemistry chemistry (see (see description description of of Jardetzky Jardetzky & 8[ Buchanan Buchanan the domain pilot project). project). We \Ve have also begun begun investigating investigating Blackboard Blackboard systems systems as a way way pilot of organizing organizing expert expert systems systems to to exploit exploit concurrency. concurrency. Initial Initial work work has begun begun of using the HASP/AGE HASP/AGE systems systems as an application application example. example. using [See HPP t,echnical technical memos memos HPP-83-30, HPP-83-30, HPP-83-33, HPP-83-33, HPP-83-38, HPP-83-38, HPP-83-43, HPP-83-43, [See HPP HPP-83-H, HPP-84-4, HPP-84-4, HPP-84-61 HPP-84-6] HPP-83-44, Knowledge acquisition: acquisition: How How is knowledge knowledge acquired acquired most most efficiently”from efficiently"from 3. Knowledge human expert,s, experts, from from observed observed data, data, from from experience, experience, and and from from discovery? discovery? human How can a program program discover discover inconsistencies inconsistencies and and incompleteness incompleteness in its its How knowledge base? base? How can the the knowledge knowledge base be augmented augmented without without knowledge How perturbing t,he the established established knowledge knowledge base? base? perturbing A. Feigenba.um Feigenbaum E. A. gO 90 HPP HPP Core Core AI AI Research 5P41 RR00785-11 RR00785-11 \Ve We have continued continued to to make make progress on two two on-going on-going projects projects for for learning learning by by experience experience and learning learning by by analogy, analogy, and have initiated initiated work work on three three new systems Those Those three three are learning learning by by watching, watching, systems for for acquiring acquiring knowledge. knowledge. learning All learning from from text, text, and learning learning rules rules & meta-rules meta-rules inductively. inductively. All three three of of the t,he new systems systems use use medical medical problems problems as their their test-domains. test-domains. [Preliminary [Preliminary results results have been published published in HPP-83-27, HPP-83-27, HPP-83-36, HPP-83-36, HPP-84-2, HPP-84-2, HPP-84-8.] HPP-84-S.] 4. Knowledge utilization: By By what what inference inference methods methods can many many sources of of hr7owledge utilization: knowledge jointly and effiCiently knowledge of of diverse diverse types types be made to to contribute contribute jointly efficiently toward How can knowledge knowledge be used intelligently, intelligently, especially especially in toward solutions? solutions? How systems with with large knowledge knowledge bases, bases, so that that it it is applied applied in an appropriate appropriate manner manner at the appropriate appropriate time? time? These issues issues are being being explored explored in the the development development of of MRS MRS (Meta(MetaRepresentation Representation System) System) where where one of of the roles of of meta-knowledge meta-knowledge is to guide the effective They are also central central in the studies studies effective use use of of lower lower level knowledge. knowledge. They of of Blackboard Blackboard control control systems and their their use use in concurrent concurrent expert expert systems. [See [See HPP HPP technical technical memos membs HPP-83-26, HPP-83-26, HPP-83-28, HPP-83-28, HPP-83-30, HPP-83-30, HPP-83-33, HPP-83-33, HPP-83-38, HPP-83-38, HPP-84-1, HPP-84-1, HPP-84-2, HPP-842, HPP-84-6] HPP-84-61 5. Software Software Tools: Tools: How How can specific specific programs programs that that solve specific specific problems problems be generalized generalized to more more widely widely useful useful tools tools to to aid a.id in the development development of of other other programs programs of of the same class? class? We We have continued continued the development development of of new software software tools tools for for expert expert, system system construction construction and the distribution distribution of of packages that that are reliable reliable enough and documented documented so that that other other laboratories laboratories can use use them. them. These include include the old old rule-based rule-based EMYCIN EMYCIN system, system, MRS, MRS, and AGE. AGE. [See [See HPP HPP technical technical memos HPP-83-26, HPP-83-26, HPP-83-28, HPP-83-28, HPP-83-29, HPP-83-29, HPP-83-33] HPP-83-331 6. Explanation and How Explanation and Tutoring: Tutoring: How can the knowledge knowledge base base and the line of of reasoning reasoning used in solving solving a particular particular problem problem be explained explained to to users? users? What What constitutes constitutes a sufficient sufficient or or an acceptable acceptable explanation explanation for for different different classes classes of of users? How can ca.n knowledge knowledge in a system system be transferred transferred effectively effectively to students students users? How and trainees? tra.inees? The NEOMYCIN program comparison The NEOMYCIN program has undergone undergone preliminary preliminary comparison with with medical medical students' students’ protocols protocols to to understand understand the extent extent to to which which its medical medical concepts concepts match match those of of the students. students. Analysis Analysis of of experts' experts’ problem problem solving solving has also been done. NEOMYCIN’ NEOMYCIN's s explanation explanation capabilities capabilities have been improved. improved. NEOMYCIN in New work work on student student modelling modelling has started started in order order to to test test NEOMYCIN the cont.ext context of of tutoring. tutoring. [See [See HPP HPP technical technical memos HPP-83-41, HPP-83-41, HPP-83-42, HPP-83-42, HPP-84-2, HPP-84-2, HPP-84-7] HPP-84-71 7. Planning Planning and Design: and Design: planning planning and design? design? const.raints? const.raint,s? How How are What What are reasonable and effective effective methods methods for for How How can symbolic symbolic knowledge knowledge be coupled coupled with with numerical numerical constraints const,raints propagated propagated in design problems? problems? The The Palladio Palladia system system for for assisting assisting in the design of of VLSI VLSI circuits circuits has been demonstrated demonstrated and results results presented presented in major major publications publications and conferences. [See [See HPP HPP technical t,echnical memos HPP-83-31, HPP-83-31, HPP-83-39, HPP-83-39, HPP-83-45, HPP-83-45, HPP-83-46. HPP-83-46, HPP-83-47, HPP-83-47, HPP-84-3, HPP-84-3, HPP-84-5] HPP-84-51 8. Diagnosis: How Diagnosis: How can we build build a diagnostic diagnostic system system that that reflects reflects any any of of several 91 E. A. Feigenbaum Feigenbaum 5P41 RR00785-11 RR00785-11 HPP HPP Core Core AI AI Research diagnostic How use knowledge How can we use knowledge at different different levels of of diagnostic strategies? strategies? abstraction process? abstraction in the diagnostic diagnostic process? Research on using using causal models in a medical medical decision decision support support system system (NESTOR) was largely completed and will be published in the coming (NESTOR) largely completed will published coming year. year. A A ~econd program that physiology second medical medical diagnosis diagnosis program that uses uses causal models of of renal renal physiology (AIjMM) published soon. We (AI/MM) was also substantially substantially completed completed and will will be published We are investigating investigating the process of of diagnosis diagnosis in electronics electronics as as well well as as in medicine. medicine. The The major major thrust thrust of of this this work work has been integrating integrating causal models about, about, and the the structure body. structure of, a computer computer system system or or systems systems of of the the human human body. [See [See HPP HPP technical technica. reports: reports: HPP-83-32, HPP-83-32, HPP-83-37, HPP-83-37, HPP-83-40, HPP-83-40, 84-7] 84-71 D. Relevant Relevant Publications Publications D. HPP-83-28 HIT-83-20 Miehael “MRS Casebook", Casebook “, May May 1983. Michael R. Genesereth, Genesereth, ".MRS HPP-83-27 EDT’-83-27 Thomas Thomas HPP-83-28 HIT-83-28 Michael Meta-level Representation Representation System", Michael R. Genesereth,"A Genesereth, “A Meta-level System “, May Ma.y 1983. HPP-83-29 HIT-83-29 M. Grinberg, Installation Instructions Instruction8",“, May This Grinberg, "MRS “MRS Installation May 1983. This report purchased the software report available available only only to t,o those who who have purchased software system system MRS. MRS. HPP-83-30 HIT-83-30 Barbara Barbara HPP-83-31 HPP-83-31 Harold Tong, An Harold Brown, Brown, Christopher Christopher Tong, Gordon Gordon Foyster, Foyster, "Palladio: “Palladia: An Exploratory Exploratory Environment for for IC IC Design Design",“, June 1983. Environment HPP-83-32 HFP-83-32 John Kunz, Kunz, E.A.Feigenbaum, E.A.Feigenbaum, Bruce Bruce G. Buchanan, Buchanan, E.H. E.H. Shortliffe, Shortliffe, "Comparison of Decision hfaking Afuking “Comparison of Techniques Techniques of of Computer-Assisted Computer--4ssisted Decision in publication in the Pure and in Aledicine". Medicine “. Submitted Submitted for for publication Pure and Applied Applied Biostructure. World Biostructure. World Press, Singapore Singapore (1983). HPP-83-33 HIT-83-33 Nelleke Aiello, for Expert Expert Nelleke Aiello, "A “A Comparative Comparative Study Study of of Control Control Strategies Strategies for SY8tems: AGE Implementation Implementation of PUFF",“, June Systems: AGE of Three Yariations Variations of of PUFF 1983. HPP-83-34 HIT-83-34 Jock Mackinlay, Presentation: The Generation Problem for for Mackinlay, "Intelligent “Intelligent Presentation: Generation Problem User Interfaces Inter faces ",“, March March 1983. HPP-83-36 I-ET-83-30 Russell Greiner New'? A Semantic Greiner and ~tichael Michael R. Genesereth, Genesereth, "H-1wt's “What’s New? Semantic Definition of Novelty",“, June 1983. Definition of Novelty HPP-83-37 HIT-83-37 Robert About Time-dependent Behavior in Robert Joyce, "Reasoning “Reasoning About Time-dependent Behavior in a System System for Diagnosing Diagnosing Digital Digital Hardware Hardware Faults Faults",“, August for August 1983. 1983. HPP-83-38 Hpp-83-38 Barbara Blackboard Model A/odel of Barbara Hayes-Roth, Hayes-Roth, "The “The Blackboard of Control Control ",“, June 1983. HPP-83-39 mP-83-39 Jerry Expert System for .Jerry Yan, Yan, Gordon Gordon Foyster, Foyster, Harold Harold Brown, Brown, "An “An Expert Syste,m for Integrated Circuits", As.'Jigning A[ask Levels Levels to Interconnect Interconnect in Assigning Mask in Integrated Circuits “, October October 1983. E. A. Feigenbaum Feigcnbaum D. Dietterich and Ryszard Dietterich Ryszard S. S. Michalski, Michalski, Patterns in Patterns in Sequences of of Objects Objects If, “, May May 1983. "Discovering “Discovering Hayes-Roth, "The Blackboard Architecture: Architecture: A Hayes-Roth, “The Blackboard Framework for for Problem Problem Solvi1'g'?" Framework Solving?” May May 1983. 92 92 General General HPP HPP Core Core AI AI Research Research 5P411 RR00785RR00785-11 5P4 11 HPP-83-40 HPP-83-40 Benoit Mulsa.nt Mulsant and and Benoit A Daily Activity on A Daily Activity on HPP-83-41 HPP-83-41 (working paper) paper) Diane Diane Warner Warner Hasling, Hasling, “Strategic "Strategic Explanations Explanations for for aa (working Diagnostic Consultation Consultation System System “,", in in MA1 AAAI Proceedings Proceedings 1983 1983 pp. pp. Diagnostic David Servan-Schreiber, Servan-Schreiber, “I s been improved improved with with a region region based window window system system which which increases both both speed and flrxibility. The The region region based window window system, system, the register register input input devices, and the flexibility. formatting formatt,ing language la.ngua.ge interface interface (that (that describes how how forms forms should should be presented presented on the display) display) have been generalized generalized to to be usable by by other other portions portions of of the ONCOCIN ONCOCIN project project (notably (notably the OPAL OPAL knowledge knowledge acquisition acquisition interface interface described described below). below). C.2.2 Development of program to physician inter face to C.2.2 Development of new new program to connect connect the the physician interface to the the reasoning portion of program reasoning portion of .,the the program The The ONCOCIN ONCOCIN system system uses uses a special design that that allows allows the Interviewer Interviewer program program a.nd program to operate and the reasoning reasoning section section of of the program operate independently. independently. In order order to coordinate coordinate the activities programs, a special communication a,ctivities of of these two two programs, communication program, program, called called the t,he Interactor 1nteract)or was drsigned designed and built. built. The t,eractor program program provides provides a message The In Interactor message passing facility facility between between two two or or more more Interlisp-D processes (sub-programs Interlisp-D processes (sub-programs that that can run run at the same time). time). The The form form of of the messages The messa.ges are specified specified by by the programmer. programmer. The system system further further allows allows messages messages to to processes machines processes running running on different different machines via via the computer computer network network called called the ETHERNET. This ETHERNET. This will will allow allow moving moving components components of of a large program program from from one to to several machines The Interactor Interactor also has the machines in a way way invisible invisible to the programs programs themselves. The ability ability to rind find other other Int.eractors Interactors on the local communication communication network. network. C.2.3 Detleiopment of programs to C.2.9 Development of new new programs to improve improve the the efficiency efficiency and and capabilities capabilities of NCO CIN 0f o olvc~oc~Iiv 109 E. A. A. Feigenbaum Feigenbaum ONCOCIN ONCOCIN Project Project 5P41 RR00785-11 In order order to speed speed up both both versions versions of the ONCOCIN ONCOCIN system, we have written written aa simple simple rule and control control block block compiler compiler for for ONCOCIN ONCOCIN that that converts converts rules and control control blocks into into Interlisp Interlisp programs, programs, and then then into into compiled compiled Interlisp. Interlisp. This This helps to alleviate alleviate aa memory version memory space space problem problem we have had in the Interlisp-lO Interlispversion of the system system as as well well as as the workstation version of the program. give us increased speed in speed workstation version program. Another Another systems level 3.'>pect aspect of our our work work is in the creation creation and access access of efficient efficient patient patient record record data data files. To To this this end, we have implemented implemented a machine machine independent independent hash file system system (special data data record record format) format) that that allows access access to the data data base base via via memory memory from and Interlisp-D The system system is is compatible compatible with with both both Interlisp-lO InterlispInterlisp-D and from disk files. The allows sharing sharing of files between between the two two systems. Its Its format format is is also machine machine independent independent enough to allow allow access access from from other other lisps on other other computers. computers. It It is currently currently accepted by XEROX XEROX as as a standard standa.rd for for the D machines machines and has been used by them them to bring bring up programs programs of use use by all D machine machine users. users. Along Along the same lines, we have experimented experimented with wit,h solutions solutions to the problems problems of having having portions portions of text text easily easily accessible accessible by key key from from a file in a machine machine independent independent way. way. C.2.4 Reorganization and programs for for improved C.2.4 Reorganization and recoding recoding of of existing ezisting programs improved efficiency efficiency The The reasoning reasoning portion portion of the t.he ONCOCIN ONCOCIN program program is being reprogrammed reprogrammed to increase speed and to benefit from the special capabilities of the Interlisp benefit from capabilities Interlisp workstation. workstation. We are also re-writing parts of the program that were borrowed We re-writing parts of program that borrowed from from other other expert expert systems developed developed by our our group. group. \Ve We have reorganized reorganized the system syst,em into into logical logical subsystems subsystems that that are of a manageable manageable size. This This consisted consisted of cat.egorizing categorizing all the system system functions functions (portions (portions of of the program) program) that that are necessary for putting each in an appropriate The for the Reasoner to to run run and putting appropriate file. file. The Reasoner now runs runs in stand stand alone mode independently independently of of which which system system it it is on. programming part part for \Ve We are now now in the the process of of cleaning cleaning up the specific specific programming for each of of the subsystems. both style subsystems. This This entails entails making making various various enhancements enhancements for for both style and efficiency, efficiency, adding and further breaking down breaking down functionally functionally adding comments comments and documentation, documentation, further independent parts of the system. independent parts of system. We portions of We have transferred transferred portions of our our EMYCIN EMYCIN utilities utilities (based on the MYCIN MYCIN expert both Interlisps. expert syst.em) and rewritten rewritten those utilities utilities to to make make them them work work in both Interlisps. We We have removed program that removed from from the stand-alone stand-alone Reasoner sections sections of of the the program that depended on specific hardware hardware of of the the DEC-2060 DEC-2050 mainframe mainframe computer computer and now now have versions versions of of the specific ONCOCIN on the 2060 2050 and D-machines D-machines that that are identical, identical, being being generated generated from from the same ONCOCIN program text,. t.ext. This This step also included included the the use of of the the new new hash file syst,em system (described (described program D-machines. above) on the D-machines. C.2.5 System Sy.'1tem support support for for the the reorganization reorganization C.2.5 implemented a program program called called Graphcalls Graphcalls which which allows allows programmers programmers on \Ve have implemented the D-ma.chines D-machines to to visually visually graph graph the structure structure of of the the programs programs they they have written. written. One t,he examine the use of of each of of the the functions functions on the graph graph as well well as examine examine and can also examine variables they they access. access. It provides visual visual tracing tracing and and dynamic dynamic control control of of a change the variables It also provides program in in execution. execution. ItIt has been used daily daily since its its creation creation by by both both our our project project and pr0gra.m members of of the the SUMEX SUMEX community. community. members C.B Goal: Goal: To To modify modify the the 2020 Clinic Clinic Version Version of of ONCOCIN ONCOCIN in in response response to to user user C.9 feedback feedback During the the last last year, year, we have added added a number number of of new new options options to to ONCOCIN ONCOCIN for for use During by the the fellows fellows in the the clinic. clinic. These include: include: a special special option option to to request request that that a test test be by A. Feigenbaum Fdgenbaum E. A. 110 ONCOCIN ONCOCIN Project Project 5P41 5P41 RR00785-11 RR00785- 11 ordered for entering entering reasons reasons for for treatment treatment ordered STAT STAT (immediately), (immediately), special menus for modifications modifications (these are used used when there is a disagreement between the ONCOCIN ONCOCIN recommended therapy physician's therapy and the physician’ s treatment treatment plan, in order order to to gather gather data data about about why why the physician physician has decided to override override the system), and the option option to to request a copy of a patient’ patient's s flowsheet flowsheet be printed printed out out on the clinic clinic line-printer. line-printer. We We have also streamlined streamlined the met.hods met,hods that that the various various forms forms are created created by by ONCOCIN. ONCOCIN. C.4 for use the used C.4 Goal: Goal: To encode encode and and implement implement for use by by ONCOCIN ONCOCIN the COfnmonly commonly used chemotherapy protocols from from our chemotherapy protocols, our oncology oncology clinic clinic We have pursued two two approaches to to increasing the number number of of prot.ocols protocols known known by by our The first first approach is to to use use the existing existing software software to to implement implement active active our system. The protocols not protocols not encoded encoded at the time time of of our our last report. report. The The second second approach has been been to develop new software software that that is able to to dramatically dramatically speed speed up the entry entry of of protocols protocols by by providing graphically-oriented providing graphically-oriented forms forms to to be filled filled out out on the computer computer that that follow follow the basic outline protocol documents. outline of the protocol In t.he protocol to t,he past, adding adding a new protocol to the ONCOCIN ONCOCIN knowledge base base has been been a tedious process process in which programmer sit which an oncologist oncologist and a programmer sit down and translate translate the oncologist's All oncologist’s knowledge about about the protocol protocol into into rules accessible accessible to ONCOCIN. ONCOCIN. All the rules pertaining pertaining to the new protocol protocol are written process must written at at that that time, time, and this this process must be repeated for protocol that This for every new protocol that is added to ONCOCIN. ONCOCIN. This method method is rather rather inefficient inefficient since many many of of the rules are similar similar between protocols, protocols, differing differing only only in their their data To To speed speed the process process of knowledge acquisition, acquisition, a program program is being data content. content. developed whereby whereby a doctor doctor could sit sit down at a terminal terminal and fill fill in a series series of of forms forms containing containing appropriate appropriate questions about about a new protocol. protocol. The The information information entered would would take care of the large number number of general rules pertaining pertaining to the protocol protocol and allow allow the doctor doctor and programmer programmer to concentrate concentrate on the special cases. cases. The The program program will will have two two levels, the first first of of which which is the program program that that will will interact interact directly directly with with the doctor. doctor. This This program program runs on Xerox Xerox D machines which which have extensive graphics Sections of the display graphics capabilities. capabilities. display screen screen (called windows) windows) are organized in a way way that that emulates the physician's physician’s patterns patterns of thought thought when thinking thinking about about the prot.ocoi. protocol. Other Other graphical gra.phical entry entry devises devises have been been used used to to encourage encourage pointing pointing at the answer rather rat.her than than text text entry. entry. These methods are able to display display all of the possible choices choices in a compact compact and comprehensible way. way. The The first first phase phase of of this this program program has been been complrted complet.ed and h;JS has been been examined and approved approved by our our oncology collaborators. collaborators. InFormat.ion Informat.ion entered in the top top level program program will will be be converted converted to t,o an intermediate intermediat,e data data. structure structure which which will will be be used used by the second second level of the program program to make new rules for the ONCOCIN ONCOCIN knowledge base. base. Eventually, Eventually, this this process process will will also work work in the opposite opposit,e direction direclion so that that information informa,tion about about a previously previously entered protocol protocol can be copied or or modified modified by the t,he physician physician for the new protocol. protocol. This This "similar “similar to" to” option option will will also extend t·o t.o chemotherapies chemot,herapies and drugs, so that that when the doctor doctor enters a chemotherapy chemotherapy or or drug drug that that the system systt=rn knows about, about, pertinent pertinent information information will will be filled filled in for the doctor doctor to copy or or modify. modify. The The "similar “similar to" to” capably capably along with with the use use of of graphical graphical input input devices to speed speed the process process of entering entering a new protocol protocol and will will also reduce errors errors and duplications. duplications. When When this this project project is completed completed the total total time time needed needed to to enter enter a new protocol protocol should be greatly greatly reduced and more effort effort will will be concentrated concentrated on fine tuning tuning the rules to handle special situations. situations. C.5 Evaiuat'ions of Performance (2.5 Eualuations of ONCOCIN's ONCOCIN'S Performance Data Data collection collection and analysis for all three ONCOCIN ONCOCIN evaluations evaluations are now complete, complete, results results were presented at the annual annual meeting meeting of of the Society Society for Medical Medical Decision Making, Making, and we expect to have formal formal reports reports published published during during the next next year. 111 111 E. A. Feigenbaum Feigenbaum ONCOCIN ONCOCIN Project Project 5P41 RR00785-11 RR00785- 11 Study Study 1, overseen by by Dr. Dr. Robert Robert Carlson Carlson of of the Division Division of of Oncology, Oncology, is an evalua,tion of of the program's program’s impact impact on the attitude attitude of of the the oncology oncology fellows fellows towards t,owards evaluation computers All All physicians physicians were administered administered computers in general genera.1 and ONCOCIN ONCOCIN in particular. particular. questionnaires and structured structured interviews interviews in the Spring Spring of of 1981 1981 before before ONCOCIN ONCOCIN was questionnaires introduced. intQoduced. The The same questionnaires questionnaires were distribut.ed distributed to them them again after after they they had used system for for over over a year. year. Follow-up Follow-up interviews interviews were also undertaken. undertaken. This This study study was the system repeated again a.ga.in during during 1983 to to determine determine the trends trends over over time. time. The The results results of of this this study study repeated are presently presently being being prepared prepared as a formal formal report. report. \Ve We are also revising revising this this study study in preparation preparation for for the integration integration of of the workst.ation version version of of ONCOCIN ONCOCIN into into the clinic. clinic. To To maintain maintain some consistency consistency in the workstation evaluation process, the original original questions questions from from Study Study 1 will will be given given and analyzed analyzed as as evaluation before along with with new questions. questions. Several Several of of these new survey survey and interview interview questions questions will will before serve as as a "baseline" “baseline” for for evaluating evaluating any any perceived perceived improvements improvements that that will will come with with the introduction intZroduction of of the professional professional workstations workstations in the clinic. clinic. Study 2, overseen by by Dr. Dr. Daniel Daniel Kent Kent of of the Division Division of of General General lilternal Internal Medicine, Medicine, Study is an evaluation evaluat.ion of of the program's program’s impact impact on the the completeness completeness and accuracy accuracy of of flowsheet flowsheet data Research programmers data recorded recorded with with and without without ONCOCIN. ONCOCIN. programmers wrote wrote routines routines to to formally analyze a.naiyze on-line on-line flow flow sheets for for completeness completeness and accuracy. accuracy. Pre-ONCOCIN Pre-ONCOCIN flow flow formally then entered entered into into the system system exactly esactly as as they they were were originally originally recorded by by the the sheets were then recorded physicinn. The same analytic analytic routines routines were used to to analyze analyze these pre-ONCOCIN pre-ONCOCIN flow flow physician. The sheet,s. The The pre- and postpost- ONCOCIN ONCOCIN data data were compared. compared. sheet.s. Results Results indicat.e that that ONCOCIN has had a statistically statistically significant significant beneficial beneficial impact impact on the completeness completeness of of data data ONCOCIN recording, the ordering ordering of of required required tests, and the accuracy accuracy of of the data data recorded. recorded. A A formal formal recording, report report of of the results results is in preparation. preparation. Finally, Study Study 3 is examining examining the comparison comparison between between ONCOCIN's ONCOCIN’s therapeutic therapeutic Finally, treatment decisions made by by oncology oncology fellows fellows in the same setting. setting. The The advice and the treatment study was coordinated coordinated by by Dr. Dr. David David Hickam, Hickam, formerly formerly of of our our Division Division of of General General Internal Internal study hledicine and now now on the the faculty faculty at the University University of of Oregon Oregon in Portland. Portland. Medicine Expert Expert evaliuators rated rated treatment treatment plans plans without without knowing knowing whether whether the recommendation recommendation was that tha.t evaluators of ONCOCIN ONCOCIN or or one of of the clinic clinic physicians. physicians. Over of Over 200 flow flow sheets were evaluated evaluated by by Stanford lymphoma lymphoma experts, experts, and the resulting resulting data data have been fully fully analyzed analyzed by by Dr. Dr. Stanford Hick-am. The The results results indicate indicate that that the experts experts were unable unable to to fault fault the recommendations recommendations Hickam. by ONCOCIN ONCOCIN relative relative to those of of experienced experienced oncology oncology fellows fellows treating treating patients patients made by with wit,h lymphoma. lymphoma.. A A paper pa,per describing describing the results results is in preparation. preparation. A stlldy study was made ma.de of of all all of of the cases cases run run by by phYSicians physicians in the clinic clinic to determine determine A st.atist.ics when they they chose chose to to override override ONCOCIN's ONCOC’I~N’s therapy therapy recommendation. recommendation. The The stat istks about. when result#s showed that that approximately approximately 75% 75% of of the time time they they agreed completely. completely. \Vhen When there there resl1lts disagreements, 15% 15% were about about individual individual drug drug doses. doses. This were disagreements, This study study pointed pointed out out a number of of situations situations where where ONCOCIN ONCOCIN needs more more knowledge, knowledge, and snd where where our our expert expert nnmber clarificntion from from the Principal Principal Investigators Investigators of of the particular particular protocol. protocol. needed clarificar.ion As a result.. a meeting meet,ing was wa.s held (7/12/83) (7/12/83) with with some of of the Faculty Faculty in charge of of the Hodgkin's Hodgkin’s result. prot,ocols to discllsS discllss issues issues arising arising from from this this study. study. protocols C.6 Documentation C.6 Documentation An extensive extensive effort effort to to document document the ONCOCIN ONCOCIN system system was completed completed during during this this An year. Many Many aspects of of the ONCOCIN ONCOCIN program program and its its programming programming environment environment are last year. now \vritten writ,ten and available available for for project project members' members’ use. use. The now The increase in documentation documentation has significantly reduced reduced the start-up start-up time time for for new researchers researchers working working with with the project. project. In significantly addition, we have published published several papers papers and prepared prepared several several technical technical reports reports addition, describing the system. syst,em. describing E. A. Feigenbaum Feigenbaum 112 112 ONCOCIN ONCOCIN Project Project 5P41 RR00785-11 RR00785-11 C.7 Hypothesis Assessment Assessment C.7 Hypothesis As mentioned mentioned above, largely largely through through the the efforts efforts of of Curtis Curtis Langlotz, Langlotz, we have continued continued to to develop modifications modifications to ONCOCIN ONCOCIN that that will will permit permit it it to to function function as as an "observer" physician's “observer” of of the physician’ s own own decisions rather rather than than as as a primary primary source of of advice. By By permitting the physician physician to plan on the flowsheet, permitting to enter enter his or or her her own own therapy therapy plan flowsheet, we can acknowledge acknowledge the oncologist's oncologist’s ability ability to to reach reach appropriate appropriate therapeutic therapeutic decisions for for most most patients. ONCOCIN physician's plan with patients. ONCOCIN will will simply simply compare compare the physician’ s plan with what what it it believes is the proper therapy. If physician, or If the system system agrees with with the the physician, or determines determines that that small small proper therapy. differences differences are clinically clinically insignificant, insignificant, no advice from from the computer computer will will be necessary. If If significant significant disagreements disagreements occur, occur, on the other other hand, hand, ONCOCIN ONCOCIN will will need to respond respond with with warnings plan may warnings and explanations explanations for for why why it it feels that that an alternate alternate therapy therapy plan may be preferable. Our preferable. Our experience with with ONCOCIN ONCOCIN since its its clinic clinic implementation implementation suggests that that tllis preferred by physicians. It this mode of of interaction interaction will will be preferred by the clinic clinic physicians. It will will require require minimal minimal changes decision chnnges to ONCOCIN's ONCOCIN’s decision making making approach, approach, but but the the determination determination of of what what differences differences are clinically clinically significant, significant, and the optimal opt,imal method method for for explaining explaining their their importance physician, are exciting problems. importance to the physician, exciting challenges and important important theoretical theoretical problems. An International Journal Journal An initial initial report report, describing describing this this work work appeared appeared during during 1983 in the International of Al11tl-Afachine Studies, plan to continue of Man-i\lachine Studies, and we plan continue enhancing enhancing the the system's system’s critiquing critiquing and explanation Mr. Mr. Langlotz Langlotz presented presented this this work work in the 1983 Society Society for for explanation capabilities. capabilities. Computer Computer Applications Applications in Medical Medical Care Care Conference Conference Student Student Paper Paper Competition, Competition, and was a finalist finalist in the competition. competition. The The approach approach will will not not be used in the clinic, clinic, however, however, llDtiI until ONCOCIN professional workstations, ONCOCIN has been transferred transferred to to professional workstations, hopefully hopefully in about about two two years. C.7 Rule Analysis Analysis C.? Query Query System System and and Rule Shako Shoko Tsuji Tsuji has completed completed her work work on the development development of of a query query system system to permit permit easy access access to the large ONCOCIN ONCOCIN knowledge knowledge base. Once we had encoded several several hundred hundred rules. rules, it it became unwieldy unwieldy for for system system builders builders to work work from from large hard-copy hard-copy listings listings of of the t,he knowledge knowledge base, base, and we anticipate anticipate that that physicians physicians will will also require require direct direct access access to the program's progra.m’s knowledge. knowledge. The The query query system system permits permits this this kind kind of of access. access. Rather Rather than than dealing dea.ling with with natural natural language understanding understanding by by computer, computer, we are designing designing ways ways that that menu menu selection select,ion and the high-speed high-speed interface interface can be used to to permit permit access access to the information information that that is needed by by a physician physician or or system system builder. builder. A A paper paper describing describing the early early work A,A.MSI Congress work was presented present.ed last year year (May (May 1983) at the A4MSI Congress 83 8.Y in San Francisco. Francisco. In previous previous reports reports we also described described the work work of of Dr. Dr. Motoi Motoi Suwa Suwa who who developed developed programs to 3:"sist programs a.?sist. in determining determining knowledge knowledge base base consistency consistency and completeness. His paper on this AI Magazine. Afagazine. However, paper t,his subject subject appeared appeared in late 1982 in the AI However, the programs programs that. t,hnt, he wrote wrote were never never formally formally linked linked to to our our system system for for writing writing rules and modifying modifying otller parts of Noble spent ot)her p;Lrts of tlle t,he knowledge knowledge base. base. As a result, result, Mr. Mr. Robert Robert Noble spent time time during during the last few years modifying modifying Suwa's Suwa’s code so that that it it would would operate operat,e as 3s an integral integral part part of of ONCOCIN. These changes ONCOCIN. cha.nges have now now been implemented implemented so that that a new rule rule can be dynamically dynnmicslly cOlllpared compared to the rest of of the knowledge knowledge base base during during the process of of knowledge knowledge entry. Noble is currently entry. Mr. Mr. Noble currently considering considering how how such a program program might might be implemented implemented on a workstation workstation in order order to take take advantage advantage of of the newly newly available available graphical graphical capabilities capabilities of of these macllines. machines. C.g Encoding of Additional Fkotocols Protocols C.9 Encoding of Additional As was indicated to a indicated above, we have emphasized emphasized transfer transfer of of ONCOCIN ONCOCIN professional workstation rather of professional workstation rather than than the implementation implementation of additional additional protocols. protocols. However, However, the oncologist oncologist in charge of of breast breast cancer treatment treatment at Stanford Stanford had expressed great great interest interest in adding adding those treatment treatment protocols protocols to to the system system as as soon as as possible. We We 113 E. A. Feigenbaum Feigenbaum 5P41 RR00785-11 RR00785-11 ONCOCIN Project Project ONCOCIN accordingly encoded and and thoroughly thoroughly tested tested the the treatment treatment plans plans for for adjuvant adjuvant therapy therapy have accordingly of breast breast carcinoma carcinoma (CMF (CMF and and CMFVP CMFVP treatment treatment plans) plans) and and released them them for for regular regular of the spring spring of of this this year. year. Encoding Encoding of of the the CMF CMF treatment treatment plan plan required required encoding encoding of of use in the special rule rule types. types. In In order order to to represent represent these treatment treatment plans plans special special methods methods were were special created for for looking looking back back to to previous previous cycles to to compa.re current current laboratory laboratory results results to to created previous va.lues. values. This This allows allows the the development development of of treatment treatment recommendations recommendations based upon upon previous past experience experience with with the the patient. patient. A A number number of of other other protocols protocols were were added added to to the the past ONCOCIN system system in order order to to keep the the system’ system's knowledge about about Hodgkin’ Hodgkin's and ONCOCIN s knowledge s and These included Lymphoma protocols protocols current. current. included new Lymphoma Lymphoma protocols protocols with with very very Lymphoma complex alternating alternating chemotherapies chemotherapies (M-HOP/B-Cepp/HD-MTX (M-HOP jB-CeppjHD-MTX and M-BACOD/HDM-BACODjHDcomplex MTX), and new Hodgkin’ Hodgkin's protocols (alternating (alternating MOPP/ABVD). MOPP j ABVD). hlTX), s protocols C.l0 Strategic Strategic Therapy Therapy Planning Planning C.10 As mentioned mentioned above, we have begun begun a new new research research project project to to study study the the therapy therapy As planning process, and how how strategies strategies which which are used to to plan plan therapy therapy in difficult difficult cases cases pla.nning might be represented represented on a computer. computer. This This project, project, which which we call call the the ONYX ONYX project, project, has might as it,s its goals: to to conduct conduct basic research research into into the the possible representations representations of of the the therapy therapy planning process; to to develop develop a computer computer program program to to represent represent. this this process: process; and eventually eventually planning t.o interface interface the the planning planning program program with with ONCOCIN. ONCOCIN. The The project project members members (Fagan, (Fagan, to Bischoff, Williams, Williams, Langlotz, Langlotz, and Rennels) Rennels) have spent spent many many hours hours meeting meeting with with Dr. Dr. Sikic Sikic Bischoff, trying to to understand underst.and how how he plans plans therapy therapy for for patients patients whose special special clinical clinical situation situation trying precludes following following the the standard standard therapeutic therapeutic plan plan described described in the the prot.ocol protocol document. document. In In precludes March of of t)his this year, year, the the group group spent spent two two days at at Xerox Xerox Palo Palo Alto Alto Research Center Center binrch (PAHe), working working with with Mark Mark Stefik, Stefik, Daniel Daniel Bobrow Bobrow and Sanjay MiHal of of PARC PARC on (PARC), Sanjay Mittal possible representations representations for for the knowledge knowledge structures structures and how how such a program program might might run run We hope to to have a prototype prototype of of this this using the LOOPS LOOPS knowledge knowledge programming programming system. system. We system system running running this this year. year. D. Publications Publications Since January 1989 D. Since January 1983 Expert systems systems research: research: modeling modeling the the Expert In: An to Monitoring medical An Integrated Integrated Approach Approach to Afonitoring medical decision decision malcing making process. In: (J.S. Gravenstein, Newbower, A.K. N.T. Smith, Gravenstein, R.S. Newbower, A.K. Ream, and N.T. Smith, eds.), pp. 183-:!OO. 183-200, \Voburn, Woburn, MA: MA: Butterworth's, Butterworth’s, 1983. 1. (*) (*) Shortiiffe, Shortiiffe, E.H. E.H. and Fagan, Fagan, L.M. L.M. 2. 2. Duda, Dudn, R.O. R.O. and Shortliffe, Shortliffe, E.H. E.H. Expert systems research. Expert Science, Science, 220:251-258 220:261-268 (1983). Adapting a consultation consultation system system to 3. (*) (*) Langlotz, La,nglot,z, C.P. C.P. and Shortliffe, Shortliffe, E.H. E.H. Adapting International Journal of Man-Machine Studies, critique International Journal of ]t.lan-Afachine Studies, critique user plans. 19:479-495 19:479-496 (1983). 4. Graphical access access to the knowledge knowledge base base of of a 4. (*) (*) Tsuji, Tsuji, S. S. and Shortliffe, Shortliffe, E.H. E.H. Graphical Proceedings of Congress 89, medical medical consultation consultation system. Proceedings of AAAfSI AAMSI Congress 83, pp 551-555. San Francisco, Francisco, CA, CA, May May 1983. 1983. 5. (*) (*) Bischoff, Bischoff, M.B., M.B., Shortliffe, Shortliffe, E.H., E.H., Scott, Scott, A.C., A.C., Carlson, Cnrlson, R.W. R.W. and Jacobs, C.D. C.D. Integratio:l Integration of a computer-based computer-based consultant consultant into into the clinical clinical setting. setting. Proceedings 7th Symposium on Applications in Proceedings 7th Annual Annual Symposium on Computer Computer Applications in Afedical Medical Care, October 1983, 1983, Baltimore, Baltimore, Maryland. Maryland. Care. pp. 149-152. October 6. Mulsant, Rlulsant, B. and Servan-Schreiber, Servan-Schreiber, D.: Knowledge Knowledge engineering: engineering: a daily daily activity activity on a hospital and Re. ion of confounding confounding time time intervals, intervals, or or statistical statistical control. control. The The Study Study Module Module then then determines detrrmines an appropriate appropriate statistical statistical method, method, using knowledge knowledge stored stored as as production production rules. \10st. 1lost studies studies have used a longitudinal longitudinal design involving involving a multiple multiple regression model model applied applied to individual individual pat.ient pat,ient records. records. Results Results across patients patients are combined combined using weights weights based on the t,he precision precision of the estimat.ed estimat,ed regression coefficient coefficient for for each patient. patient. B \/edical Relevance B..•,Weciical Relevance and and Collaboration Collaboration As a t.est t,est bed for for system syst,em development development our our focus of of attention attention has been on the records records of patients patients with with systemic systemic lupus lupus erythematosus erythematosus (SLE) (SLE) contained contained in the Stanford Stanford port.ion port,ion of the ARAMIS ARAMIS Da.t.a Da,t,a Bank. Bank. SLE SLE is a chronic chronic rheumatologic rheumatologic disease disease with with a broad broad spectrum spectrum of manifestations. manifest.ations. Occasionally Occasionally the disease disease can cause cause profound profound renal renal failure failure and lead to an early early dt'ath. death. \Nith With many many perplexing perplexing diagnostic diagnostic and therapeutic therapeutic dilemmas, dilemmas, it it is a disea.se of considerable medical interest. disease of considerable medical interest. In the fut.lIre project users of future we anticipate anticipate possible collaborations collaborations with with other other project of the TOO National Stroke Northern California TOD Syst.em System SllCh such 3,S as the National Stroke Data Data Bank, Bank, the Northern California Oncology Oncology Gro1lp, Group, and the t,he St,anford St,a.nford Divisions Divisions of of Oncology Oncology and of of Radiation Radiation Therapy. Therapy. \Ve project is broadly broadly applicable We believe that tha.t this t,his research resea.rch project applicable to the entire entire gamut gamut of of chronic bulk of chronic diseases diseases that that constitute constit,ute the bulk of morbidity m0rbidit.y and mortality mortality in the United United States. States. Consider Consider flY\" five major major diagnost.ic diagnost,ic categories categories responsible responsible for for approximately approximately two two thirds thirds of of the two two million million deaths deat,hs per year year in the United United States: myocardial myocardial infarction, infarction, stroke, stroke, cancer, hypertension, Therapy hypertension. and diabetes. diabetes. Therapy for for each of of these diagnoses is fraught fraught with with controversy concerning concerning the balance balance of of benefits benefits versus costs. controversy l. Myocardial Myocardial Infarction: Infarction: Indications Indications for for and and efficacy efficacy of of coronary coronary artery artery bypass 1. graft vs. medical medical management management a,lone. alone. Indications for for long-term long-term grsft Indications antiarrhythmics ..... . long-term long-term anticoagulants. anticoagulants. Benefits Benefits of of cholesterol-lowering cholesterol-lowering antia.rrhythmics diet.s, exercise, et,c. etc. diets, Strokr: Efficacy Ef'ficacy of of long-t,erm long-term anti-platelet anti-platelet agents, long-term long-term anticoa.gulation. anticoagulation. 2. Stroke: Indications for for revascularization. revascularization. Indications Cancer: Relative Helative efficacy efficacy of of radiation radiation therapy, therapy, chemotherapy, chemotherapy, surgical surgical excision excision 3. Cancer: singly or or in combination. combination. Optimal Optimal frequency frequency of of screening screening procedures. procedures. - singly Prophylactic thera.py. therapy. Prophylactic 4. Hypertension: Hypertension: Indications Indications for for chronic an tihypertensive drugs. chronic antihypertensive drugs. arteriography in in work-up. work-up. arteriography therapy. Efficacy Efficacy versus versus therapy. Role of of various various diagnostic diagnostic adverse effects effects of of tests such as renal tests renal Diabetes: Influence Influence of of insulin insulin administration administration on microvascular microvascular complications. complications. 5. Diabetes: of ora. or::!,l hypoglycemics. hypoglycemics. Role of Feigenbaum E. ‘A. 4. Feigenbsum -.. 1'>'> 132 RADIX RADIX Project Project 5P4l 5PA 1 RR00785-11 RR00785- 11 Despite the expenditure expendit,ure of billions billions of dollars dollars over recent recent years for randomized randomized controlled controlled trials trials (RCT's) (RCT’s) designed to answer these and other other questions, answers have been been slow in coming. coming. ReT's RCT’s are expensive of funds and personnel. The therapeutic therapeutic questions in clinical clinical medicine medicine are too numerous numerous for each each to be be addressed by its own series series of RCT's. RCT’s. On t.he t,he other ot,her hand. hand, the data data regularly regularly gathered gathered in patient patient records records in the course of the t.he normal normal performance performance of health health care delivery delivery are a rich rich and largely largely underutilized underutilized resource. The of these data The ease ease of accessibility accessibility and manipulation manipulation data afforded afforded by computerized computerized clinical clinical databases holds out out the possibility possibility of a major major new resource for for acquiring acquiring knowledge knowledge on the evolution evolution and therapy therapy of chronic chronic diseases. diseases. The The goal of the research that that we are pursuing pursuing on SUMEX SUMEX is to increase the reliability reliability of knowledge knowledge derived derived from from clinical clinical data data banks with with the hope of providing providing a new tool tool for augmenting augmenting knowledge knowledge of diseases diseases and therapies therapies as as a supplement supplement to knowledge knowledge derived the incorporation derived from from formal formal prospective prospective clinical clinical trials. trials. Furthermore, Furthermore, incorporation of knowledge knowledge from from both both clinical clinical data data banks banks and other other sources into into a uniform uniform knowledge knowledge base base should should iflC'rease increase the ease ease of of access access by individual individual clinicians clinicians to this this knowledge knowledge and thereby thereby facilitate facilitate both both the t,he practice pract.ice of of medicine medicine as as well well as as the investigation investigation of of human human disease disease processes. processes. C. of C. Highlights Highlights of Research Research Progress Progress C.l C.1 1 .May May 1988 198.9 to to 1 May May 1984 198;5 Our Our primary primary accomplishments accomplishments in this this period period have been the following: following: 1) complete complet,e modifications modifications to RADIX RADIX to accommodate accommodate the one hundred-fold hundred-fold increase patients, in the size of of our our database to to 1700 patients, 2) carry prednisone on serum carry out out the study study of of the effect effect of of prednisone serum cholesterol cholesterol on the new database. database. 3) publish publish results patient prednisonejcholesterol prednisone/cholesterol study, results of of the 1700 patient study, 4) publish publish the description by us to this description of of a two-stage two-stage regression method method adapted adapted by this stlldy. study, 5) complete preparation for complete System System Programmer's Programmer’s !v1:1nuals Manuals and User's User’s Manual Manual in preparation fo1 transfer tra.u?fer to to outside olltside sites, and 5) begin transfer transfer of of RADIX RADIX to to Xerox Xerox D-Machine D-Machine personal personal work work stations. stations. 6) C.l.l Modifications j'IIodifications to to RADIlr’ RADIX the enlarged enlarged database database C.l.l for for the Extensive modifications modifications to to RADIX RADIX were were required required to to deal with with the the 100-foid Extensive lOO-fold in the the size of of the database. database. The The modifications modifications necessary to to run run the study study automatically on the prednisone/cholesterol prednisone/cholesterol study study were completed completed this this year. year. automatically increase module module C.i.E Prednisone/chlouterol Prednisone/chlosterol study study on on enlarged enlarged database database C.1.2 \Ve have carried carried out out the t,he automated automated study study of of the the effect effect of of prednisone prednisone on serum serum We cholesterol using using the the new new 1700 patient patient database. database. ItIt has strongly strongly confirmed confirmed the the effect effect cholest,erol previously observed obst'rved in the the 50-patient 50-patient SLE SLE database. da.tabase. In In addition, addition, we are examining examining the the previously effect in non-SLE non-SLE patients patients and and in in other other patient patient subsets. We We are also examining examining alternative alternative effect pharmacokinetic models models for for the the prednione prednione effect effect using using the the newly newly available available data. data. pharmscokinetic C.1.3 Publish Publish results results of of predn.isone/cholesterol prednisone/cholesterol study study C.1.9 123 A. Feigenbaum Feigenbaum E. A. 5P41 RR00785-11 RR00785- 11 RADIX RADIX Project Project The paper reporting The paper reporting these results results is in draft draft form. form. It It will will be submitted submitted for for publicat.ion shortly. publication shortly. C.LI Publish description of method C.1.4 Publish description of 2-stage !&stage regression regression method A publicat.ion. A description description of of the 2-stage Z-sta.ge regression method method has been submitted submitted for for publicat,ion. C.l.5 Documentation C.l.5 Document&ion A A two-volume two-volume System System Programmer's Programmer’s Manual Manual and a User's User’s Manual Manual describing describing implementation, implementation, maintenance maintenance and use of of the the system system at at Stanford Stanford has been completed. completed. In addit.ion, addit,ion, a complete complete set of of the files needed for for on-line on-line demonstrations demonstrations has been created, created, separating separating them them from from the working working versions. versions. C.1.6 RADIX to D-Machines CJ.6 Tran.ser Transer of of RADIX to D-Machines Prl.'liminary Preliminary work work on implementing implementing RADIX RADIX on D-Machines D-Machines has begun. This This will will cont.inue continue in coming coming years. years. C.l.7 C. 2.7 Other Other accomplishments accomplishments \Ve have present.ed the results results of of our our research at at several several conferences during during the year. publications for publications. year. Additional Additional publications for the year year are noted noted in the section section on publications. C.2 Research in Progress C.2 Research in Progress \Ve patient We are currently currently completing completing additional additional studies studies on subsets of of the the 1700 patient database. These include automated analysis of the prednisone/ cholesterol effect in nondatabase. include automated analysis of prednisonel cholesterol effect SLE patients and subsets of patients, and fitting pharmacokinetic models SLE pa.tients of SLE SLE patients, fitting alternative alternative pharmacokinetic of prednisone/cholesterol effect. of the prednisone/cholesterol effect. This This work work should should be completed completed shortly. shortly. We We will will then then return below in the ret,urn to to the more more AI-oriented AI-oriented aspects of of RADIX, RADIX, as as described described below the section section on Resea.rch Resea.rch Plans. Plans. D. Publication8 Publications D. 1. Blum. Regression: Application Application to Clinical Blum. R.L.: R.L.: Two Two Stage Stage Regression: to a Time-Oriented Time-Oriented Clinical Database. (Submitted Dcltabnue. (Submitted Epidemiology.) Epidemiology.) for for publication publication to to the American American Journal Journal of of R.L.: Prednisone Elevates Elevates Cholesterol: An Automated Study R.L.: Prednisone Cholesterol: An Automated Study of of Longitudinal Clinical Data. (Manuscript preparation.) Longitudinal Clinical Data. (Manuscript in preparation.) 2. 9. Blum, Blum, 3. Minimycin: AA !vfiniature Rule-Based System nnd Walker, Walker, M.G.: M.G.: Minimycin: Miniature Rule-Ba8ed System 3 . Blllm, Bliim, R.L R.L.... and (Submitted publication to (Submitted for for publication to M.D.Computing) M.D.Computing) R.L.: A/odding and encoding clinical causal R.L.: Modeling and encoding dinid CCJu8al relationships. relationships. Proceedings Proceedings of of SCAMC, SCAMC, Baltimore, Baltimore, MD, MD, October, October, 1983. 4. Blum, Blum, 5. Blum, Representation of derived Blum, R.L.: R.L.: Representation of empirically empirically deriued causal causal relationships. relationships. IJCAI, IJCAI, Karlsruhe, Ksrlsruhe, \Vest West Germany, Germany, August, August, 1983 . R.L.: representation of Machine representation 0 / clinical clinical R.L.: !v/achine MEDINFO MEDINFO 83, Amsterdam, Amsterdam, August, August, 1983. causal relationships. causal relationships. R.L.: decision making aboard Clinical decision making aboard the the R.L.: Clinical Chairman's paper, Session on Artificial Intelligence Chairman’s paper, Session Artificial Intelligence Making, hlaking, AAMSI, AAMSI, San Francisco, Francisco, May, May, 1983. Starship Enterprise. Starship Enterprise. and Clinical Clinical Decision Decision 5. G. Blum. Blum. 7. Blum, Blum, 8. Blum. hypotheses on Studying hypotheses on a time-oriented time-oriented Blum, R.L. R.L. and Wiederhold, Wiederhold, G.: Studying E. A. Feigenbaum Feigenbaum 124 RADIX RADIX Project Project 5P41 RR00785-11 RR00785-11 database: .An An overvie'w of RX project. project. Proc. overview of the the Rx Proc. Sixth Sixth SCAMC, SCAMC, IEEE, IEEE, database: \Vashington D.C., October, 1982. \Vashington D.C., October, 9. Blum, Induction of from a time-oriented clinical Blurn, R.L.: R.L.: Induction of causal causal relationships relationship8 from time-oriented clinical database: An overview project. Proc. database: An overview of of the the R:'<: RY project. Proc. AAAI, AAAI, Pittsburgh, Pittsburgh, August, August, 1982. 1982. 10. Blum, Automated induction of from a timeinduction of causal causal relationships relationships from timeBlum, R.L.: R.L.: Automated oriwted RX project. Proc. oriented clinical clinical database: database: The The RXproject. Proc. AMIA AMIA San Francisco, Francisco, 1982. Discovery and Representation of Relationships from from a 11. 11. Blum, Blum, R.L.: R.L.: Discovery and Representation of Causal Causal Relationships Clinical Database: The RX Project. Project. IN Large Time-oriented IN D.A.B. D.A.B. Lindberg Lindberg Large Time-oriented Clinical Database: The RX and P.L. NOTES IN P.L. Reichert.z Reichertz (Eds.), (Eds.), LECTURE LECTURE NOTES IN MEDICAL MEDICAL INFORMATICS, INFORMATICS, SpringerVerlag, 1982. Springer-Verlag, 1982. Discovery, confirmation, and incorporation 12. Blum, Blum, R.L.: R.L.: Discovery, confirmation, and incorporation relationships from a large time-oriented clinical database: The relationships from large time-oriented clinical database: The Computers April, Computers and Blomed. Biomed. Res. Res. 15(2):164-187, 15(2):161-187, April, 1982. 1982. of causal of causal RX project. RX project, 13. Discovery and of from a 13. Blum, Blum, R.L.: R.L.: Discovery and representation representation of causal causal relationships relationships from large time-oriented clinical database: The RX project project (Ph.D. large time-oriented clinical database: The RX (Ph.D. thesis). thesis). Computer Computer Science and Biostatistics, Biostatistics, St.anford Stanford University, University, 1982. 1982. R.L.: Displaying clinical data from a time-oriented clinical data from time-oriented R..L.: Di8playing Computers 1981. Computers in BioI. Biol. and Med. 11(4):197-210, 11(4):197-210, 1981. 14. 14. Blum. Blum. database. database. 15. Automating the study of hypothe,'3es 15. Blum, Blum, R.L.: R.L.: Automating the study of clinical clinical hypotheses on on a time-oriented time-oriented databa,'le: R.Y project. project. Proc. database: The The RY Proc. MEDINFO MEDINFO 80, Tokyo, Tokyo, October, October, 1980, pp. 4,56-4130. 456-460. (Also (‘41~0 STAN-CS-79-816) STAN-CS-79-816) 16. Inferring knowledge from clinical data jerring knowledge from clinical data 16. Blum, Blum, R.L. R.L. and Wiederhold, Wiederhold, G.: In utilizing techniques from artificial intelligence. Proc. banks technique8 from artificial intelligence. Proc. Second banks utilizing SCMvlC, November, 1978. SCAh3C:. IEEE, IEEE, Washington, Washington, D.C., D.C., November, 1978. Blum. R.L.: R.Y 17. .Blum. R.L.: The The RY clinical data clinical data banking banking University University Ph.D. Ph.D. thesis thesis project: AA medical consultation system integrating project: medical consultation system integrating and intelligence methodologies, Stanford and artificial artificial intelligence methodologies, Stanford proposal, August, proposal, August, 1978. 1978. 18. 18. Kuhn, Kuhn, Ingeborg, Ingeborg, Gio Gio Wiederhold, Wiederhold, Jonathan Jonathan E. Rodnick, Rodnick, Diane Diane M. M. RamseyRamseyKlee, Automated Ambulatory Ambulatory Medical Medical Klee, Sanford Sanford Brnctt, Benet,t, and Donald Donald D. Beck: Beck: Automated published by by Springer-Verlag, Record Sy.5tems Springer-Verlag, 1983, in Record System8 in in the the U.S., U.S., to be published Information Inforrna.tion Systems Systems for for Patient Patient Care, Care, B. Blum Blum (ed.), Section Section III, III, Chapter Chapter 14. 19. Walker, \Valker, M.G., Lisp Tutorial. (Submitted publication to M.G., and Blum, Blum, R.L.: R.L.: AA Lisp Tutorial. (Submitted for for publication to M.D.Computing.) M.D.Computing.) 20. Wiederhold, Knowledge and Database Management, A1anagement, IEEE W’iederhold, Gio: Gio: Knowledge and Database IEEE Software Software Premier pp.63--73. Premier IsslJe, Issue, Ja11.1984, Jan.1984, pp.63--73. 21. Wiederhold,Gio: \Viederhold,Gio: Networking Networking of Data Information, Information, of Data \Vorkshop on the Role of Computers Workshop of Computers in Cancer Cancer Institu tes of Health, June 1983, pp.113-11 9. Institutes of Health, 1983, pp.L13-119. National National Clinical Clinical Cancer Cancer Trials, Trials, Institute Institute National National 22. \Viederhold, Database Design Design (in Wiedrrhold, Gio: Gio: Database (in the Computer Computer Science Series) McGrawMcGrawHill NY, May Hill Book Book Company, Company, New York, York, NY, May 1977, 678 pp. Second edition, edition, Jan. 1983, 1983, 758 768 pp. 125 125 E. A. A. Feigenbaum Feigenbaum RADIX Froject Project RADIX 5P41 RR00785-11 RR00785-11 5P41 23. Wiederhold, Wiederhold, G.: G.: IN IN D.A.B. D.A.B. Lindberg Lindberg and and P.L. P.L. Reichertz Reichertz (Eds.). (Eds.), Databases Databases lor for 23. Health Cure, Lecture Notes in Medical Informatics, Springer-Verlag, 1981. Health Care, Lecture Notes in Medical Informatics, Springer-Verlag, 1981. Wiederhold, G.: G.: Database Database technology technology in, in health health care. care. J. J. Medical Medical Systems Systems 24. Wiederhold, 5(3):175-Hlt), 1981. 5(3):175-196, E. Funding Funding Support Support Statue Status E. Representation and and Use of of Causal Causal Knowledge Knowledge for for Inference Inference from from 1) Representation Databases DatabEes Robert L. L. Blum, Blum, M.D., M.D., Ph.D.: Ph.D.: Principal Principal Investigator Investigator Robert Total award: award: $89,597 $89,597 (direct (direct ++ indirect) indirect) Total Term: hIarch l\.'larch 15, 1984 through through March March 14, 1986 Term: DeriVing Knowledge Knowledge from from Clinical Clinical Databases Databases 2) Deriving Gio C. C. M. M. Wiederhold, Wiederhold, Ph.D.: Ph.D.: Principal Principal Investigator Investigator Gio National Library Library of of Medicine Medicine National Total award: award: $291,192 $291,192 (direct) (direct) Total Term: May May 1, I, 1984 through through November November 30, 3D, 1986 Term: II. II. INTERACTIONS WITH WITH THE THE SUMEXA.IM SUMEX-AIM: RESOURCE RESOURCE INTERACTIONS A.. Collaborations -4. Collaborations During During the the last last year year we have completed completed System System Programmer's Programmer’s Manuals Manuals and and a User's User’s Manual hlsnual as as steps steps towards towards making making the the system system available available to to outside outside collaborators. collaborators. We We have had preliminary preliminary discussions Edward Shortliffe Shortliffe and and Lawrence Lawrence Fagan Fagan discussions with with Drs. Drs. Edward concerning project. Once Once the the RADIX RADIX concerning use of of components components of of RADIX RADIX in the the ONCOCIN ONCOCIN project. program progrs.m is developed, developed, we would would anticipate anticipate collaboration collaboration with with some of of the the ARAMIS ARAMIS project project sites in the further further development development of of a knowledge knowledge base pertaining pertaining to to the chronic chronic arthritides. arthritides. The The ARAMIS ARAlMIS Project Project at at the Stanford Stanford Center Center for for Information Information Technology Technology is used by by a number number of of institutions institutions around around the the country country via via commercial commercial leased lines to store store and process their their data. data. These institutions institutions include include the University University of of California California School School of of Medicine, hledicine, San Francisco Francisco and Los Angeles; Angeles; The The Phoenix Phoenix Arthritis Arthritis Center, Center, Phoenix; Phoenix: The The University University of Cincinnati Cincinnati School of Medicine; Medicine; The The University University of of Pittsburgh Pittsburgh School of Medicine; Kansas University; and The University of Saskatchewan. All All of of the The University Saskatchewan. hledicine; Kans,as University: rheumatologists at these sites have closely collaborated with the development of ARAMIS, rheumatologists closely collaborated with development of ARAMIS, and their their interest interest in and use use of the RADIX RADIX project project is is anticipated. anticipated. We We hasten hasten to mention mention thaI, we do not expect SUMEX to support the active use of RADIX as an on-going that not expect SUMEX support active use RADIX as on-going service to this this extensive extensive network network of arthritis arthritis centers, centers, but but we would would like to be able to allow allow the national nationa. centers centers to participate participate in the development development of the arthritis arthritis knowledge knowledge base base and to test that that knowledge knowledge base base on their their own clinical clinical data data banks. banks. B. Interactions EX-AIM Projects Interactions with with Other Other SUM SUA4ELX-AIM Projects Several Several of of the conce!}ts concepts incorporated incorporated into into the design design of of the RADIX RADIX Project Project have been by other The RADIX RADIX knowledge knowledge base base is is similar similar to been inspired' inspired .by other SUMEX-AIM SUMEX-AIM Projects. Projects. The the Units The The production production rule inference inference Units Package of of the MOLGEN MOLGEN PROJECT. PROJECT. mechanism mechanism used used by us us is is similar similar to that that in in the MYCIN MYCIN Project. Project. Several Several programs programs developed developed by by the the MYCIN MYCIN group group are are regularly regularly used used by by RADIX. RADIX. These These include include disk disk hash hash file file facilities, facilities, text text editing editing facilities, facilities, and and miscellaneous miscellaneous LISP LISP functions. functions. E. E. A. ,A. Feigenbaum Feigenbaum 126 126 RADIX RADIX Project Project 5P41 5P4 1 RR00785-11 RR00785- 11 Regular Regular communication communication on programming programming details details isis facilitated facilitated by the on-line on-line mail mail system. C. Critique Critique of of Resource Resource Alanagement Management The The DEC DEC System System 20 continues continues to provide provide accepta.ble acceptable performance, performance, but but it it is is frequently frequently heavily heavily loaded at peek hours. The The SUMEX SUMEX resource management management continues continues to be accessible accessible and cooperative. cooperative. ID. III. RESEARCH RESEARCH PLANS PLANS A. Project Project Goals and and Plans The The overall overall goal of the RADIX RADIX Project Project is to develop a computerized computerized medical medical information system capa.ble of accurately extracting medical knowledge information system cspa.ble accurately extracting medical knowledge pertaining pertaining to the therapy therapy and evolution evolution of of chronic chronic diseases diseases from from a database consisting consisting of of a collection collection of of stored stored patient patient records. records. SHORT-TERM SHORT-TERM GOALS GOALS -For For the last two two years we have concentrated concentrated more heavily heavily on publishing publishing and presen tat ion of our presentation our earlier earlier AI AI results, results, on acquisition acquisition of a 1700 patient patient database, on medical medical studies studies based on the enlarged enlarged database, and on reporting reporting the medical medical results results and statistical statist.ica.1 techniques techniques arising arising from from our our research. This This is is in concert concert with with the long-term long-term goal of ensuring ensuring that that the work work of of the SUMEX SUMEX / Artificial Artificial Intelligence Intelligence in Medicine Medicine community community be disseminated he disseminated and applied applied in the general general medical medical community. community. During coming two two years years we will will concentrate concentrate much much more more on the artificial artificial During the coming intelligence intelligence aspects of of RADIX. RADIX. We We were successful this this year year in obtaining obtaining funding funding from from the National Library National Science Foundation pursue this National Library of of Medicine Medicine and the National Foundation to pursue this work. work. In In particular, we will particular, will be deeply deeply concerned concerned with with the representation representation of of causal, temporal, temporal, and 11antitative medical qqusnt!it,st,ive medical knowledge. knowledge. It It has become clear clear that that t,hese these types types of of knowledge knowledge are crurial crucia.1 for for the RADIX RADIX tasks of of automated automated discovery discovery of of medical medical knowledge knowledge and the provision of provision of intelligent intelligent, automated aut.omsted assistance to to clinical clinical researchers, in addition addition to their t,heir generally perceived value generally perceived value in other other medical medical expert expert systems systems applications. applications. LONG-RANGE GOALS GOALS -- There There are two two inter-related inter-related long-range long-range goals of of t,he the LONG-RANGE RADIX Project: Project: 1) automatic automatic discovery discovery of of knowledge knowledge in a large time-oriented time-oriented database database RAD1.X and 2) provision provision of of assistance to to a clinician clinician who who is interested interested in testing testing a specific specific a.nd hypothe:ois. These t.asks tasks overla,p overla.p to to the extent extent that that some of of the algorithms algorithms used for for hypothesis. discovery are also used in the process of of testing testing an hypot,hesis. hypothesis. discovery \Ve hope to to make make these broad range of hypotheses broad of hypotheses and records. records. algorithms sufficiently SUfficiently algorithms over a broad spectrum over broad spectrum robust that that they they will will work work robust of data distributions in the of data distributions the over a over patient patient B. Justification Justification and and Requirements Requirements for for Continued Continued Use of of SUMEX SUMEX B. Computerized clinical clinical data data banks banks possess possess great great potential potential as tools tools for for assessing assessing the the Computerized effica.cy of of new diagnostic diagnostic and and therapeutic therapeutic modalities, modalities, for for monitoring monitoring the the quality quality of of health health efficacy delivery, and for for support support of of basic medical medical research. research. Because of of this this potential, potential, many many care delivery. clinical da.ta data banks banks have recent,ly recently been developed developed throughout throughout the the United United Stat,es. States. However, However, clinical the init)ial initial problems problems of of data data acquisition, acquisition, storage, storage, and retrieval retrieval have been dealt dealt with, with, once the there remains remains a. a. set of of complex complex problems problems inherent inherent in the the task task of of accurately accurately inferring inferring t.here 127 A. Feigenbaum Feigenbaum E. A. RADIX RADIX Project F’roject 5P41 5P41 RR00785-11 RR00785- 11 medical medics.1 knowledge knowledge from from aa collection collection of of observations observations in in patient patient records. records. These These problems problems concern concern the the complexity complexity of of disease disease and and outcome outcome definitions, definitions, the the complexity complexity of of time time relationships, relationships, potential potential biases biases in in compared compared subsets, subsets, and and missing missing and and outlying outlying data. data. The The major major problem problem of of medical medical data data banking banking isis in in the the reliable reliable inference inference of of medical medical knowledge knowledge from from primary primary observational observational data. data. \Ve see see in in the the RADIX RADIX Project Project aa method method of of solution solution to to this this problem problem through through the the utilization of knowledge engineering techniques from artificial intelligence. The utilization of knowledge engineering techniques from artificial intelligence. The RADIX RADIX Project, Project, in in providing providing this this solution, solution, will will provide provide an an important important conceptual conceptual and and technological technological link link to aa large community community of medical medical research groups groups involved involved in the treatment treatment and study study of the chronic chronic arthritides arthritides throughout throughout the United United States and Canada, Canada, who who are presently presently using the ARArvns Data Bank Ba.nk through through the CIT CIT facility facility via via TELENET. TELENET. ARAMIS Data Beyond Beyond the arthritis arthritis centers which which we have mentioned mentioned in this this report, report, the TOD TOD (Time-Oriented Data (Time-Oriented Data Base) User Group Group involves involves aa broad broad range of university university and community community medical medical institutions institutions involved involved in the treatment treatment of cancer, stroke, stroke, cardiovascular cardiovascular disease, disease, nephrologic nephrologic disease, disease, and others. others. Through Through the RADIX RADIX Project, Project, the opportunity opportunity will will be be provided provided to foster foster national national collaborations collaborations with with these research groups groups and to provide provide aa major major arena in which which to demonstrate demonstrate the utility utility of artificial artificial intelligence intelligence to clinical clinical medicine. medicine. C. Recommendation8 for Re80urce Recommendations for Resource Development Development The The on-going on-going acquisition acquisition oi oi personal personal work-station work-station Lisp Lisp processors is a very very positive positive step. step, as a.s these provide provide an a.n excellent excellent environment environment for for program program development, development, and can serve as as a vehicle vehicle for for providing providing programs programs to to collaborators collaborators at other other sites. Continued Continued acquisitions acquisitions are very very desirable. desirable. Another Another resource that that would would be highly highly desirable desirable is a faster faster and more more reliable reliable means for programs interactively between SUMEX for transferring tra.nsferring data data and programs interactively between SUMEX and the CIT CIT IBM IBM 370. The The addition addition of of a reliable reliable local network network facility facility would would greatly greatly facilitate facilitate our our ability ability to to transfer patient files from transfer patient from CIT CIT to to SUMEX. SUMEX. E. A. A. Feigenbaum Feigenbaum E. 128 National National AIM AIM Projects Projects 5P41 51’41 RROOi85-11 RR00785-11 ll.A.2. AIM II.A.2. National National AIM Projects Projects The following following group group of projects projects isis formally formally approved approved for access access to the AIM AIM aliquot aliquot of the SUMEX-AIM Their access access isis based based on review review by the AIM AIM Advisory Advisory SUMKX-AIM resource. Their Group Group and approval approval by the AIM AIM Executive Executive Committee. Committee. In addition addition to the progress reports reports presented presented here, abstracts abstracts its individual users are submitted on a separate Scientific individual users submitted Scientific Subproject Subproject 129 for each project project and Form. Form. A. Feigenbaum Feigenbaum E. A. CADUCEUS CADUCEUS Project Project 5P41 5P41 RR00785-11 RR00785- 11 ll.A.2.1. CADUCEUS Project II.A.2.1. CADUCEUS Project CADUCEUS CADUCEUS Project Project J. D. D. Myers, Myers, M.D. M.D. and and Harry Harry E. E. Pople, Pople, Jr., Jr., Ph.D. Ph.D. University University of of Pittsburgh Pittsburgh Decision Decision Systems Systems Laboratory Laboratory Pittsburgh, Pittsburgh, Pa., Pa., 15261 15261 I. Su.M:MARY SUMMARY OF OF RESEARCH RESEARCH PROGRAM PROGRAM A. A. Projcct Project rationalc rationale The The principal principal objective objective of this this project project is the development development of a high-level high-level computer computer diagnostic diagnostic program program in the broad broad field field of internal internal medicine medicine as as an aid in the solution solution of complex To be effective, effective, the program program must must be complex and complicated complicated diagnostic diagnostic problems. problems. To ~apable capa.ble of multiple multiple diagnoses (related (related or independent) independent) in aa given patient. patient. A A major major achievement achievement of this this research undertaking undertaking has been the design of of a program along with progra.m called INTERNIST-I, INTERNIST-I, with an extensive extensive medical medical knowledge knowledge base. base. This This program program has been been used used over over the past past decade to analyze analyze many many hundreds hundreds of of difficult difficult diagnost.ic diagnost,ic problems problems in t.he t,he field field of internal internal medicine. medicine. These problem problem cases cases have included included cases journals (particularly Case Records of the Massachusetts Massachusetts cases publisht"cl published in medical medical journals (particularly Records of General CPCs, and unusual unusual problems problems General Hospital, Hospital, in the New England England Journal Journal of Medicine), Medicine), CPCs, of patients in our Medical Center. In most instances, but by no means all, INTERNIST-I of patient,s our Medical Center. In most but by all, INTERNIST-I has performed high-lighted performed at the level of the skilled skilled internist, internist, but but the the experience has high-lighted several areas (or for improvement. improvement. B. Medical Afcdical Releoance Relevancc and and Collaboration B. Collaboration The progra.m program inherently inherently has direct direct and substantial medical relevance. relevance. The substantial medical The institution institution of of collaborative collaborative studies with other other institutions institutions has been deferred deferred The studies with pending completion completion of of t,he the programs programs and knowledge knowledge base enhancements enhancements required required for for pending CADUCEUS. The The insta.llation installation of of our our own, own, dedicated dedicated VAX VAX computer computer can be expected expected to to CADUCEUS. considerably any any future future collaboration. collaboration. aid considerably HI~ghlights of of Research Rcsearch Progress Progrcss C. Highlights ----Accomplishments this this past past year year ----Accomplishments In a previous previous progress progress report report the the concept concept of of “facets” "facets" of of diseases diseases was was introduced introduced In and the the need of of CADUCEUS CADUCEUS to to proceed proceed from from broad broad pathophysiopathophysio- logical logical and and pa.thopathoand biochemical concepts t,o specific disease processes was emphasized. The need for better biochemical concepts t,o specific emphasized. The for better represent,ation of of a.natomical anatomical information information and and for for better better time time representation representation were were pointed pointed represent,ation out. out. Drs. hliller rvliller and and Myers Myers have have continued continued in in the the development development of of a new new format format for for the the Drs. CADUCEUS knowledge knowledge base. major goal goal in in making making the the transit,ion transition from from the the CADLiCEUS AA major INTEH0:IST-I knowledge knowledge base to to that that of of CADUCEUS CADUCEUS ha.s has been been to to insure insure that that there there is INTERNIST-I continuity between between the the two: two: The The CADUCEUS CADUCEUS knowledge knowledge base base will will be be derived derived from from the the cont.inuity information in in the the INTERNIST-I INTERNIST-I knowledge knowledge base, base, with with significant significant additions additions made made as as information necessary. necessary. A. Feigenbaum Feigenbaum E. A. E. 130 CADUCEUS CADUCEUS Project Project 5P41 5P41 RR00785-11 RRO0785-11 A screen-oriented screen-oriented editor editor program program for entering entering and manipulating manipulating the knowledge knowledge base was written writt.en in Franz Franz Lisp. Using the editor, editor, aa total total of 52 52 diagnosis nodes have been been base crea.ted and aa total total of 282 282 findings findings have been defined. Due to the more complex complex nature nature of created finding, the 282 282 findings findings represent represent over over 600 old INTERNIST-I INTERNIST-I style style manifestations. manifestations. aa. finding. In the CADUCEUS CADUCEUS knowledge knowledge base, base, the basic unit unit of observational observational in- formation formation is is Unlike an INTERNIST-I INTERNIST-I manifestation, aa finding finding can be be assigned aa called aa. finding. Unlike manifestation, finding. status within within aa given given patient, patient, either either "normal" “normal” or any number number of forms forms of of abnormal. abnormal. For For status example, example, the status status of the finding finding "heart “heart murmur" murmur” can be either either absent (normal) (normal) or Various qualifiers qualifiers are allowed allowed to modify modify aa finding. finding. For the finding finding "heart “heart present. For present. Various patient, aa user might might specify specify that that it is is heard at the second left left murmur", murmur”, in aa specific patient, interspace. systole only, only, that that it it is blowing, blowing, that that interspace. that that it it is is systolic, systolic, that that it it is heard heard in early early systole is crescendo- decrescendo. For For findings findings whose values vary vary it it is is grade gra,de 2 of 6, and its shape is numerically, numerically, (e.g. SGOT-blood) SGOT-blood) the units units of measurement measurement and the normal normal range are specified so so that that aa. user may may simply simply enter enter aa number number as as the result result of the test. test, specified The The concept concept of a disease disease profile profile has been carried carried over over from from the INTERNIST-I INTERNIST-I knowledge knowledge base. bsse. However, However, there there are three three separate separate diagnostic diagnostic node types types represented represented in the CADUCEUS CADUCEUS knowledge knowledge base: base: the disease, disease, the facet, and the subdivision. subdivision. A A disease disease is ent,ity whose presence should should be reported reported if if detected detected in a patient, patient, and conceptually conceptually an entity corresponds correeponds to the diseases diseases mentioned mentioned in the separate chapters chapters of standard standard medical medical text.bnoks. A A subdivision subdivision is either either a specific specific subtype subtype of of a disease disease (e.g. hepatitis hepatitis B is a textbooks. subt.ype subtype of acute viral viral hepatitis, hepatitis, a disease) disease) or or a major major specific specific organ organ system system involvement involvement by a multisystem multisystem disease disease (e.g. lupus lupus nephritis nephritis and lupus lupus cerebritis cerebritis are subdivisions subdivisions of of the disease system system lupus lupus erythematosus). erythematosus). disease The The internal internal organization organization of of disease, disease, facet facet and subdivision subdivision profiles profiles is identical. identical. Apart Apart from from links links to other other nodes, there there are nine essential components components to to each profile: profile: disea.:5e disea.se para.meters para.meters (e.g. prevalence prevalence of of disease, disease, specific specific sites it it effect,s); effects); demographic demographic information about about patients patients with with the disease; disease; general general predisposing predisposing factors factors (which (which are information interciepencient. interdependent, only only one of of which which is likely likely to to be present); present); independent independent risk risk factors factors (which (which often often co-exist co-exist, synergist.ically); synergistically); general general findings findings caused by by the illness; specific specific findings findings which which relatively unique unique to the disease disease process; characteristic characteristic findings findings (e.g. a positive positive throat throat are relatively culture culture for for beta beta hemolytic hemolytic streptococcus streptococcus in streptococcal streptococcal pharyngitis); pharyngitis); academically aca.demically known known but but clinically clinically contraindicont,raindi- cated findings findings (e.g. one should should not not do a renal renal biopsy biopsy in patients patients wit,h renal renal leptospirosis, lept,ospirosis, but but we know know what what the biopsy biopsy will will show if if it it is done anyway); anyway); and with manifestations whose presence make make the the diagnosis diagnosis untenable untenable (e.g. male sex makes makes manifest,ations pregnancy pregnancy an invalid invalid consideration). consideration). In In addition addition to to the aforementioned aforementioned work work in internal internal medicine. medicine. Drs. Drs. Gordon Gordon Banks Banks and a.nd John John Vries Vries have beeu been working working on the development development of of a neurological diagnostic diagnostic component component for for CADUCEUS. CADUCEUS. Dr. Banks Banks has developed developed a neurological Dr. neurnanatomic 1,000 neuroanatomic neuroanat.omic database database which which contains contains spatial spatial descriptors descriptors for for nearly nearly 1,000 neuroanatomic st,ructures structures and contains contains information information as to to their their blood blood supply, supply, and function. function. This This database database will allow allow a.na.tomic anatomic loca.lization localization of of neurologic neurologic lesions. Some of of this this work work for for the the peripheral peripheral will nervous nervous system system has been done previously previously by by students students in our our laboratory. laboratory. The The approach approach to to central nervous nervous syst.em system hss has been to to design a set of of “symbolic "symbolic coordinates”. coordinates". In the cent.ral const0uct ing the neuroanat,omic constructing neuroanatomic database, database, the human human body, body, including including the the nervous nervous system, system, conceptually partitioned partitioned into into a set of of cubes (boxes). The The largest largest cube, containing containing the the is conceptually ent,ire body. body. is 2.187m 2.187m on a side. This This cube is divided divided into into 27 smaller smaller cubes, each 729mm 729mm on a side. Each Each of of the the smaller smaller cubes is likewise likewise subdivided subdivided until until finaliy finally cubes that that are each lmm 1mm on a side are reached. reached. Thus Thus any any cube has neighbors neighbors (of (of equal size) rostra], rostral, caudal, caudal. vent,ral, ventral. dorsal, dorsal, left, left, and right right of of it, it, as well well as a “pa,rent” "parent" cube which which contains contains it, it, and “dsught,er” "dallghter" cubes which which itit contains. contains. Each Each of of these cubes has the the potential potential for for being being represented inside inside the the computer computer program program with with a unique unique name (known (known as an atom atom in LISP. LISP. represrnted which the the dat,abase database is programmed). programmed). Attached Attached to to each cube LISP LISP atom atom the language in which 131 A. Feigenbaum Feigenbaum E. A. 5P41 5P4 1 RR00785-11 RR00785- 1 I CADUCEUS CADUCEUS Project Project a.re are lists of all of the anatomic anatomic structures structures that that are completely completely and partially partially contained contained within within the cube, as as well well as as the blood supply supply to the region. This This structure structure facilitates facilitates rapid rapid retrieval retrieval of the location location of aa given anatomic anatomic structure structure as as well well as as rapid rapid localization localization of possible areas areas of involvement involvement when there isis evidence of dysfunction dysfunction of one or more neural neural systems. The The hierarchical hierarchical arrangement arrangement of of the nested cubes ensures rapid rapid convergence during during searches, searches, because because if the sought sought object object is is not not found found in a parent parent cube, there there is no need to search for for it it in any of the patient's patient’s children children cubes. cubes. The The addition addition of anatomic anatomic reasoning reasoning may may allow allow parsimonious parsimonious explanation explanation of multiple multiple manifestations manifestations arising arising from from a single lesion. lesion, or allow 3,110~the program program to query query the user regarding regarding the presence of manifestations manifestations of involvement involvement of areas that that might might be be expected to be affected affected by whatever whatever clinical clinical state st.ate the program pr0gra.m has under under current current consideration. consideration. Dr. Vries Vries has developed an imaging imaging system system using "octree “octree encoding" encoding” to reconstruct reconstruct n-dimensional n-dimensional images of of the database as as well well and images of of patients patients acquired acquired by by CT, CT, NMR, NMR, and a.nd other other neuroimaging neuroimaging techniques. techniques. Combining Combining the database with with the imaging ima.ging system correlation system may may open new areas a,rea.s of of research, including including clinical-pathological clinical-pathological correlation of of imaged lesions with with symptoms, symptoms, signs, and affected affected structures, structures, automated automated reading reading of of images, etc. Dr. Dr. r-.mler Miller in the last last year year completed completed work work on a sub-project sub-project of of CADUCEUS, CADUCEUS, called called CPCS. CPCS. He received received support support for for this this work work from from the National National Library Library of Medicine Medicine New Investigator lnvest.igator Program. Program. The The original original objective objective of the project project was to create a program, program, CPCS CPCS (for (for Computer-based Comput,er-based Patient Patient Case Simulator), Simulator), to aid in the teaching teaching of diagnosis to medical knowledge medical students. st,udents. The The INTERNIST-I/CADUCEUS INTERNIST-I/CADUCEUS knowledge base base was to be used as as the source of program's of the program’ s medical medical expertise. expertise. This This overall overall goal has been accomplished, accomplished, and the program CPCS AX-II/780 using Franz t.he program CPCS exists and runs runs on our our V VAX-h/780 Franz Lisp. Lisp. The The CPCS CPCS project was a feasibility project feasibility study study to demonstrate demonstrate that that it it is possible to to construct construct a general general case The project has been successful in that program has been The project that the CPCS CPCS program case simulator. simulator. written, and runs quite well in its small test domain. But there is room writt)en, runs quite well its small test domain. But there room for for the the future future development development. of of CPCS. CPCS. Further Further construction construction of of the CADUCEUS CADUCEUS knowledge knowledge base, in areas beyond the current beyond current set of of liver liver diseases, diseases, will will significantly significantly improve improve the utility utility of of the CPCS CPCS program. As additional a.ddit.ional capabilities capabilities are added to CADUCEUS, CADUCEUS, the corresponding corresponding changes program. will will be made in CPCS. CPCS. The medical knowledge continued to grow both both in the incorporation of The medical knowledge base has continued to grow incorporation of modification of of diseases diseases already already profiled profiled so as to to include include recent recent diseases and the modification new diseases medical knowledge. knowledge. The The knowledge knowledge base of of 3/l/84 3/1/84 includes includes 591 591 individual individual advances in medical disease profiles, profiles, 4,040 manifestations manifestations of of disease, disease, and and about about 3,500 ”"links" or disease links” or interrelationships among among diseases diseases as well well as a myriad myriad of of miscellaneous miscellaneous pieces of of int,errelnt,ionships informat.ion which which are essential essential for for the the correct correct operation operation of of the the system. system. Twenty Twenty new information profiled during during the the past past year year and the the pediatrics pediatrics knowledge knowledge base base has diseases have been profiled diseases continued to to grow. grow. continued Recently the the medical medical knowledge knowledge base (but (but not not yet yet the the diagnost.ic diagnostic program) program) has been Recently made available ava.ilable on line for for use of of the medical medical house staff staff at at Presbyterian-University Presbyterian-University ma.de Hospit.al, our our main ma.in t,eaching t.eaching hospital hospital in internal internal medicine, medicine, and and at at an affiliated affiliated community community Hospital, hospit.al in Pit.tsburgh, Pit.tsburgh, the t.he Shadyside Shadyside Hospital Hospital which which operates operates a residency residency program program in hospit,al internal medicine. medicine. Preliminary Preliminary reports reports indicate indicate that that the the residents residents find find the the knowledge knowledge base internal useful. useful. ----Research in progress progress ----Research There are five five major major components components to to the the continuation continuation of of this this research research project: project: There A. Feigenbaum Feigenbaum E. A. 132 CADUCEUS Project Project CADUCEUS 5P41 5Pdl RR00785-11 RR00785-11 1. refinement and testing testing of of the the extensive extensive 1. The The enlargement, enlargement, continued continued updating, updating, refinement medical medical knowledge knowledge base base required required for for the operation operation of of INTERNIST-I. INTERNIST-I. 2. ?. The The completion completion and implementation implementation of of the improved improved diagnostic diagnostic consulting consulting program, which CADUCEUS, which has been designed to to overcome overcome certain certain program, CADUCEUS, performance problems performance problems identified identified during during the the past past years years of of experience with with the the origina.l original INTERNIST-I INTERNIST-I program. program. 3. Institution Institution of of field field trials trials of of CADUCEUS CADUCEUS on the clinical clinical services in internal internal medicine medicine at the Health Health Center Center of of the the University University of of Pittsburgh. Pittsburgh. 4. Expansion Expansion of of the clinical clinical field field trials trials to other other university university health health centers centers which which ha.ve have expressed interest interest in working working wit,h with the system. system. 5. Ad:tptation program and data Adaptation of of the diagnostic diagnostic program data base base of of CADUCEUS CADUCEUS to SI\ bserve educational pl1rposes and the evaluation of clinical clinical performance performance and subserve educational purposes evaluation of competence. compet,ence. Current Current, activity activity is devoted devoted mainly mainly to to development of the medical knowledge development of medical knowledge diagnostic program. diagnostic consulting consulting program. the first first base, base, and two two the of of these, namely, namely, implementation implementation of of the the continued continued improved improved D. List List of publications D. of relevant relevant publicationa Harry E.:Knowledge-based Expert Expert Systems: The Buy Buy or Build Harry E.:Knowfedge-based Systems: or Build Derision IN Reitman INTELLIGENCE Derision IN Walter Walter Reitman (Ed.), (Ed.), ARTIFICIAL ARTIFICIAL INTELLIGENCE APPLICATIONS FOR Proceeding NYU Symposium. APPLICATIONS FOR BUSINESS. BUSINESS. Proceeding of of the NYU Symposium. Ablex Ablex Pub. Pub. Corp., Corp., May May 1983, 1983, pp. 23-40. 1. I. Pople, 2. Artificial Intelligence Intelligence and Medical Education Education The 3. ~1yers, Myers. J.D.: J.D.: Artificial and MedicaI The Medical Medical Journal, Journal, St. Joseph Hospital, Hospital, Houston. Houston. Vol. Vol. 18, December December 1983, pp. 193-202. E. Fwnding Funding support support E. 1. Clinical Decision Clinical Decision Systems Research Research Resource Resource Harry Harry E. Pople, Pople, Jr., Jr., Ph.D. Associate Associate Professor Professor of of Business Business Jack D. Myers, Myers, M.D. University Professor University Professor (MediCine) (Medicine) University Uni versi ty of of Pittsburgh Pittsburgh Division Division of of Research Research Resources Resources National Institutes National Institutes of of Health Health 5 R24 R24 RROllOI-07 RROllOl-07 07/01/80 - 06/30/85 - $1,607,717 $369,484 07/01/83 - 06/30/84 - $369,484 2. CADUCEUS: CADUCEUS: A A Computer-Based Computer-Based Diagnostic Diagnostic Consultant Consultant Harry Harry E. Pople, Pople, Jr., Jr., Ph.D. Associate Associate Professor Professor of of Business Business Jack D. Myers, Myers, M.D. H.D. University Professor University Professor (Medicine) (Medicine) University of University of Pittsburgh Pittsburgh National Library National Library of of Medicine Medicine National Institutes of National Institutes of Health Health 133 E. A. Feigenbaum Feigenbaum 5P41 RR00785-11 RR00785- 11 CADUCEUS CADUCEUS Project Project R01 LM03710-04 5 ROI o;r/ol/ao 06/30/85 $817,884 07/01/80 -- 06/30/85 -- $817,884 07/01/83 C6/30/84 -- $196,710 $196,710 07/01/83 -- 06/30/84 3. Neurologic Consultation Computer Program Neurologic Consultation Gordon E. Banks, Banks, M.D. Assistant Assistant Professor Professor of of Medicine Medicine National Library of of Medicine Medicine - New New Investigator Investigator National Library National Institutes of Health Health of National Institutes 5 R23 LM03889-02 Lb+03889-02 04/01/82 - 03/31/85 03/31/85 - $107,675 $107,675 04/01/82 04/oi/a3 03/31/84 $35,975 04/01/83 -- 03/31/84 -- $35,975 04/oi/a4 03/31/85 $35,975 04/01/84 -- 03/31/85 -- $35,975 ll. WITH II. INTERACTIONS INTERACTIONS WITH THE THE SUMEX-AIM SUMElX-AIM RESOURCE RESOURCE A,B. Collaborations and and Program Ka SUMEX SUMEX A,B. Medical Afedical Collaborations Program Dissemination Dissemination Via CADUCEUS CADUCEUS remains remains in a stage of of research and development. development. As noted noted above, we are continuing cont,inuing to to develop develop better better computer computer programs programs to operate operate the diagnostic diagnostic system, system, and a.nd t.he t,he knowledge knowledge base base cannot cannot be used very very effectively effectively for for collaborative collaborative purposes purposes until until it it has critical stage of of completion. completion. factors have stifled stifled collaboration collaboration via via reached a critical These factors SUI\!EX SUMEX up to this this point point and will will continue continue to to do so for for the next next year year or or two. two. In In the the meanwhile, through through the SUMEX SUMEX community community there there continues continues to to be an exchange of of meanwhile. information and states states of of progress. Such interactions interactions particularly particularly take take place at at the information annual AIM AIhI Workshop. Workshop. annual Critique of of Resource C. Critique Resource Management Jvfanagement sur-.mx Our SUhlEX has been an excellent excellent resource for for the development development of of CADUCEUS. CRDUCEUS. Our large program program is handled handled efficiently, efficiently, effectively effectively and accurately. accurately. The The staff staff at SUMEX SUMEX have uniformly supportive, supportive, cooperative, cooperative, and innovative innovative in connection connection with with our our project's projectA’s been uniformly needs. needs. III. RESEARCH RESEARCH PLANS PLANS ill. A. Goals and and Plans A. Project Project Goals Plan8 Continued Continued effort effort to complete complete the medical medical knowledge knowledge bases bases in internal internal medicine medicine and pediatrics pedintrics will will be pursued pursued including including the incorporation incorporation of of newly newly described described diseases diseases and new or or altt'n'd alt,ered medical medical information information on "old" “old” diseases. diseases. The The latter latter two two activities activities have proven proven to to Profiles of of added diseases diseases plus other other be more formidable formidable than t,ha.n originally originally conceived. conceived. Profiles informat,ion information is first first incorporated incorporated into into the medical medical knowledge knowledge base base at SUMEX SUMEX before before being being transferred into This tratlferred into our our newer newer information information structures structures for for CADUCEUS CADUCEUS on the t,he VAX. VAX. This sequence retains retains the operative operative capability capability of of INTERNIST-I INTERNIST-I as as a computerized computerized "textbook “textbook of of sequence medicine" medicine” for educational educstionsl purposes. Justification and Requirements for for Continued B. Justification and Requirements Continued SUMEX SUME. Use Our A..X. Our use use of of SUMEX SUMEX will will obviously obviously decline decline with with the installation installation of of our our V VAX Nevertheless. the excellent Nevertheless, excellent facilities facilities of of SUMEX SUMEX are expected to to be used for for certain certain developmental at SUMEX development,al work. work. It It is intended intended for for the present present to keep INTERNIST-I INTERNIST-I SUMEX for for Our team team hopes to remain remain as as a compara.tive compsra,tive use use as as CADUCEUS CADUCEUS is developed here. Our component, of of the SUMEX SUhlEX community community and to to share experiences and developments. developments. component. E. A. Feigenbaum Feigenbaum 134 CADUCEUS CADUCEUS Project Project 5P41 RR00785-11 RR00785-11 C. Need.~ Need.~ and Plans for for Other Re80urces Beyond Beyond SUAfEX-ALM and Plans Other Computing Computing Resources SUMEX-AIM Our predictable needs in this by the dedicated ~'C computer Our predictable this area will will be met met by dedicated V VAX computer recently recently installed. inst,alled. D. Recommendations Recommendations for for Future Future Community Resource Development Development D. Community and and Resource \\llether a program program like better operated Whether like CADUCEUS, CADUCEUS, when when mature, mature, will will be better operated from from centralized, personal computers centra.lized, larger larger computers computers or or from from the developing developing self contained contained personal computers is difficult predict. For difficult to t,o predict. For the foreseeable future future it it would would seem that that centralized, centralized, advanced advanced fa,cilit·ies like SUtvlEX will be important in further program development fscilit,ies like SIJMEX will important further prograrn development and refinement. refinement. 135 135 E. A. A. Feigenbaum Feigenbaum CLIPR CLIPR -- Hierarchical Hierarchical Models Models of of Human Human Cognition Cognition 5P41 5P41 RR00785-11 RR00785- I I II.A.2.2. CLIPR Models II.A.2.2. CLIPR -- Hierarchical Hierarchical Models of of Human Human Cognition Cognition Hierarchical Hierarchical Models Models or of Human Human Cognition Cognition (CLIPR (CIJPR Project) Project) Walter Kintsch and Walter Kintsch and Peter Peter G. G. Polson Polson University University or of Colorado Colorado Boulder, Colorado Boulder, Colorado I. OF PROGRAM I. SUM:MARY SUMMARY OF RESEARCH RESEARCH PROGRAM A. Project Project Rationale Rationale The The two two CLIPR CLIPR projects projects have made progress during during the last year. year. The The prose comprehension project has completed one major project, and is designing comprehension project has completed major project, designing aa prose comprehension comprehension model that that reflects reflects state-of-the-art state-of-the-art knowledge knowledge from from psychology psychology (van (van Dijk Dijk &. & Kintsch, Kintsch, 1983) and artificial artificial intelligence. intelligence. During During the last year, Polson, Polson, in collaboration collaboration with with Dr. Dr. David David Kieras Kieras of the University University of Arizona Arizona have h'ave have continued continued work work on a project project studying st,udying the psychological psychological factors factors underlying underlying device complexity complexity and the difficulties difficulties that that nontechnically nontechnically trained trained individuals individuals have in learning learning to use use devices like like word word processors. They They have develop formal formal representations representations of of aa user's user’s knowledge knowledge of of how how of of operate operate a device and of of the user-device interface interface (Kieras (Kieras & & Polson, Poison, in Press) and have completed completed several experiment.s experiment.s evaluating evaluating their their theory. theory. B. Tech.nical Goals Tech.nical Goals The first, the the text text comprehension comprehension The CLIPR CLIPR project project consists of of two two subprojects. subprojects. The The first, project, is headed by \Valter Kintsch and is a continuation of work on understanding of project, by Walter Kintsch continuation of work understanding of connected laboratory for for several years. connected discourse that that has been underway underway in Kintsch's Kintsch’s laboratory several years. The second, the device complexity complexity project project is headed by by Peter Peter Polson Polson in collaboration collaboration with with The Dayid Kierss Kieras of of the University University of of Arizona, Arizona, Tucson. Tucson. They They are studying studying the the learning learning and and David problem solving processes involved involved in the utilization utilization of of devices like like word word processors or or problem solving processes complex computer computer controlled controlled medical medical instruments instruments (Kieras (Kieras & &. Poison, Polson, in Press) Press) complex The goal of of the prose comprehension comprehension project, project is to to develop develop a computer computer system system The capable of of the meaningful meaningful processing processing of of prose. This This work work has been generally generally guided guided by by the the ca,pa,ble comprehension model model discussed by by Kintsch Kintsch and van van Dijk Dijk (1978) (1978), although although our our prose comprehension programming efforts efforts have identified identified necessary clarifications clarifications and and modifications modifications in that that programming model (Miller &. Kintsch, 1980, 1981; Kintsch & Miller, 1981; Miller, 1982). In general, model (Miller S: Kintsch, Kintsch Miller, Miller, 1983). In general, this research research has emphasized emphasized the the importance importance of of knowledge knowledge and and knowledge-based knowledge-based processes this in comprehension, compn'hension, and and we are accordingly accordingly working working with with the the AGE AGE and and UNITS UNITS groups groups at at SUMEX toward the development of a knowledge-based, blackboard model of prose SU~IEX toward the development of knowledge-based, blackboard model of comprehension. \\‘ \Ve to be able to to merge the the substantial substantial artificial artificial intelligence intelligence research research comprehension. e hope to systems with with psychological psychological interpretations interpretations of of prose comprehension, comprehension, resulting resulting in in a on these systems computational model model that that is also psychologically psychologically respectable. respectable. computational The goal goal of of the the device device complexity complexity project project is to to develop develop explicit explicit models models of of the the useruserThe device interactsion. interaction. They They model model the the device device as a nested nested automata automata and and the the user user as a device production system. system. These These models models make make explicit explicit kinds kinds of of knowledge knowledge that that are are required required to to production operate different different kinds kinds of of devices devices and and the the processing processing loads loads imposed imposed by by different different operate implementations of of a device. device. We \Ve feel that that tools tools being being developed developed at at SUMEX--in SUMEX--in particular particular implementations AGE and and the the UNIT UNIT package--will package--will dramatically dramatically facilitate facilitate our our abilities abilities to to generate generate such such AGE models of the user-device interface. models of the user-device interface. E. A. A. Feigenbaum Feigenbaum E. 136 SP41 5P41 RR00785-11 RROO785-11 CLIPR CLIPR - Hierarchical Hierarchical Models Models of of Human Human Cognition Cognition C. Medical Afedical Relevance Relevance and and Collaboration Collaboration The The text text comprehension comprehension project project impacts impacts indirectly indirectly on medicine, medicine, as as the medical medical profession is no stranger problems of profession stranger to to the problems of the information information glut. glut. By By adding adding to to the research on how how computer computer systems systems might might understand understand and summarize summarize texts, texts, and determining deterrnining ways ways by by which which the readability readability of of texts texts can be improved, improved, medicine medicine can only only be helped by research on how people understand understand prose. Development Development of of a more more thorough thorough understanding processes responsible underst,anding of of the various various processes responsible for for different different types types of of learning learning problems problems in children children and the the corresponding corresponding development development of of a successful remediation remediation strategy strategy would would also be facilitated facilitat,ed by by an explicit explicit theory theory of of the normal normal comprehension comprehension process. Note that particularly relevant Not,e that, our our goal of of a blackboard blackboard model model is particularly relevant to to the understanding understa.nding of of learning learning difficulties. difficulties. One important important aspect of of a blackboard blackboard model model is the separation sepa.ra.tion of of cognitive cognitive processes processes into into a set of of interacting interacting subprocesses. Once such subprocesses have been identified identified and constructed, constructed, it it would would be instructive instructive to to observe the model's performance when processes are facilitated model’s performance when certain certain of of these processes facilitated or or inhibited. inhibited. Many Many researchers have shown shown that that there there are a variety variety of of cognitive cognitive deficits deficits (insufficient (insufficient shortshortterm poor long-term can lead to that to term memory memory capacity, capacit,y, poor long-term memory memory retrieval, retrieval, and such) that reading problems. Having power of resding problems. Having a blackboard blackboard model model in which which the power of individual individual components components could could be ma.nipulated ma,nipulated would would be a significant significant step in determining determining the the nature nature of of such ren.ding problems. reading problems. The project has two The device complexity complexity project two primary primary goals: the development development of of a cognitive cognitive theory theory of of user-device user-device interaction interaction in including including learning learning and performance performance models, and the t,he development development of of a theoretically theoretically driven driven design process that that will will optimize optimize the relationships relationships between between device functionality functionality and ease ease of of learning learning and other other performance performance factors factors (Polson (Polson & Kieras, Kieras, 1983). The The results results of of this this project project should should be directly directly relevant relevant to to the t.he design of of complex, complex, computer computer controlled controlled medical medical equipment. equipment. We We are currently currently using using word word processors to t,o study study user-device user-device interactions, interactions, but but principles principles underlying underlying use use of of such devices should should generalize generalize to to medical medical equipment. equipment. Both Both the text text comprehension comprehension project project and the device complexity complexity project project involve involve the t#he development processes; cognitive development of of explicit explicit models models of of complex complex cognitive cognitive processes: cognitive modelling modelling is a stated NIMH. stated goal of of both both SUMEX SUMEX and research supported supported by by NIMH. The The on-going on-going development development of of the prose comprehension comprehension model model would would not not be possible without without, our our colla.boration colla.borat*ion with with the AGE AGE and UNITS UNITS research groups. groups. We We look look forward forward to a continued Several continued collaboration, colla.boratlion, with, with, we hope, mutually mutually beneficial beneficial ,results. results. Several other other psyC'hologists h3\'e psychologists ha,ve either eit,her used or or shown shown an interest interest in using an early early version version of of the prose romprehension comprehension model. model, including including Alan Alan Lesgold Lesgold of of SUMEX's SUMEX’s SCP project, project, who who is exporting exporting the system system to the LRDC LRDC Vax. Vax. \Ve We have also worked worked with with James Greeno Green0 -- another another member project -- on a project project that rnember of of the SCP SCP project that will will integrate integrate this this model model with with models of of problem solving problem solving developed developed by by Greeno Green0 and others others at the University University of of Pittsburgh. Pittsburgh. :\'eedless Needless to sa.y, say, all of of this this interaction interaction has been greatly greatly facilitated facilitated by by the local and network-wide network-wide communication communication systems supported supported by by SUMEX. SUMEX. There There has ha.s been considerable considerable commnnication communica.t.ion bet.ween bet,ween members members of of the prose comprehension comprehension and AGE/UNITS AGE/UNITS groups groups as as program bugs have been discovered program discovered and corrected; corrected; the presence of of a mail mail system system has made this this process infinitely infinitely easier than than if if telephone telephone or or surface surface mail mail messages messages were required. required. The The mail professional contacts mail system, system, of of course, has also enabled us to to maintain maintain professional contacts established established at conferences conferences and other other meetings, meetings, and to to share and discuss ideas with with these contacts. contacts. 137 137 E. A. A. Feigenbaum Feigenbaum CLIPR CLIPR - Hierarchical Hierarchical Models of Human Human Cognition Cognition 5P41 RR00785-11 D. D. Progress Progress Summary Summary The prose comprehension comprehension project project has compieted compieted an initial initial version version of aa model of prose comprehension 1980). comprehension (Miller (Miller 8l & Kintsch, Kin&h, 1980). This This model has been applied applied to aa large number of texts, and has yielded quite reasonable predictions of reca.ll number yielded quite predictions recall and readability. readability. Psychologists from other universities have used this system to derive Psychologists from other universities this system derive reading reading time time and recall recall predictions predictions for for their their own experimental experimental materials. materials. We We are currently currently using the AGE AGE and UNITS UNITS packages to extend extend this this model model toward toward one that that can make use use of world world knowledge knowledge in its analyses; this this model model is discussed in Miller Miller and Kintsch Kintsch (1981) and Miller Miller (1982). It It is further further developed in van Dijk Dijk and Kintsch Kintsch (1983) has been applied applied to the domain domain of word word arithmetic arithmetic problems problems in our our most most recent recent work work (Kintsch (Kintsch and Greeno, in Press). The The device complexity complexity project project is in it's it’s third third year. year. "We We have developed an explicit explicit model for for the knowledge knowledge structures structures involved involved in the user-device user-device interaction, interaction, and we are developing Our Our preliminary preliminary theoretical theoretical results results are described described in developing simulation simulation programs. programs. Kieras Kieras &, & Polson Polson (in (in Press). We We have also completed completed several experiments experiments evaluating evaluating the theory. theory. E. List List of of Relevant Relevant Publications Publication8 1. 1. Kieras, Kieras, D.E. D.E. and Polson, P.G.: P.G.: An An outline outline of of a theory theory of of the the user u8er complexity complezity of of devices and sy.'Jtems. sy.stem.9. Working Working Paper Paper No.1, No. 1, Device Complexity Complexity Project, Project, Universities Iiniversities of of Arizona Arizona and Colorado, Colorado, May, May, 1982. 2. Kieras, formal analysis Kieras, D.E. D.E. and Polson, Poison, P.G.: P.G.: The formal analyclis of of user u8er complexity. complexity. Int. ht. J. Man-f\.lachine Studies, In Press. Man-Machine Studies, In 3. Kintsch, tiint,sch, "V. W. and a.nd van van Dijk, Dijk, T.A.: T.A.: Toward Toward a model model of of text tezt comprehension comprehension and and production. Psychological production. Psychological Rev. 85:363-394, 1978. 4. Kintsch, J.G.:Understanding and solving word word arithmetic arithmetic Kintsch, W. W. and Greeno, Greeno, J.G.:Underatanding and solving problems. Psychological problems. Psychological Review, Review, In In Press. S. Readability and short prose prose passages: passages: and recall recall of of short 5. l'vliller, Miller, J.R. J.R. and Kintsch, Kintsch, W.: W.: Readability .4. Human -4 theoretical theoretical analysis. analysis. J. Experimental Experimental Psychology: Psychology: Human Learning Learning and f\.lemory hiemory 5:33.5-354, 61335-354, 1980. Miller, J.R. J.R. and and Kint,sch, Kintsch, W.: "V.: Readability Readability and and recall recall of of short short prose prose passages. passages. 6. h?iller, Text 1:21S-232, Text 1:215-232. 1981. J.R.: A Knowledge-based Knowledge-based Model Model of of Prose Prose Comprehension: Comprehension: Miller, J.R.: 7. hiiller, A Applications toto Expository Expository Text. Text. IN IN B.K. B.K. Britton Britton and J.B. J.B. Black Black (Eds.), (Eds.), Appficationu UNDERSTANDING EXPOSITORY TEXT. Erlbaum, Hillsdale, NJ, 1982. UNDERSTANDING EXPOSITORY TEXT. Erlbaum, Hillsdale, NJ, 8. Polson, Polson, P.G. P.G. and Kieras, Kieras, D.E.: D.E.: Theoretical Theoretical foundations foundations of of a design design process process for the minimization of user complexity. \Vorking Paper No.3, guide guide for the minimization of user complezity. \Vorking Paper No. 3, Project on ou User User Complexity, Complexity, Universities Universities of of Arizona Arizona and and Colorado, Colorado, June, 1983. Project Polson, P.G. P.G. and and Kieras, Kieras, D.E.: D.E.: A A formal formal description description of of users’ users' knowledge knowledge ofof 9. Poison, how to to operate operate a device device and and user user complexity. complexity. Behavior Behavior Research Methods Methods how Instrumentation. and Instrument,ation. van Dijk, Dijk, T.A. T.A. and and Kintsch, Kintsch, W. \v.:STRATEGIES 10. van : STRATEG IES COiHPREHENSION. Academic Academic Press, New New York, York, 1983. 198:3. CO~\IPREIIENSIO.V A. Feigenbaum Feigenbaum E. A. 138 0OF F DISCOURSE DISCOURSE 5P41 RR00785-11 RR007851 1 CLIPR CLIPR - Hierarchical Hierarchical Models Models of of Human Human Cognition Cognition Funding Support F. Funding Support Status Status 1. Text Text Comprehension Comprehension and Memory Memory \Valter Walter Kintsch, Kintsch, Professor, Professor, University University of of Colorado Colorado National Institute of Mental Health 5 Rol Nat,ional Institute of Mental Health Rol MHI5872-14-16 MH15872-14-16 7/1/81 7/l/81 - 6/30/84: 6/30/84: $281,085 7/1/83 7/l/83 - 6/30/84: 6/30/84: $69,878 $69,878 2. Understand problems Understand and solving solving word word arithmetic arithmetic problems \Valter Walter Kintsch, Kintsch, Professor, Professor, University University of of Colorado Colorado National Science Foundation National Foundation 8/1/83 8/l/83 - 7/31/86: 7/31/86: $200,000 $200,000 3. User Complexity Complexity of of Devices and Systems David David Kieras, Kieras, Associate Associate Professor, Professor, University University of of Arizona Arizona Peter Pet.er G. Polson, Polson, Professor, Professor, University University of of Colorado Colorado International International Business Machines Machines Corporation Corporation 1/1/82 ] 2/31/84: $364,000 l/1/82 12/31/84: $364,000 1/1/84 l/l,‘84 - 12/31/84: 12/31/84: $145,000 IT. WITH RESOURCE II. INTERACTIONS INTERACTIONS WITH THE THE SUMEX-AIM SUMEX-AIM RESOURCE A. Sharing Interactions with Projects .Q. Sharing and and Interactions with Other Other SUAfEX-AIAf SUMEX-AIM Projects Our primary interaction Our primary interaction with with the SUMEX SUMEX community community has been the work work of of the projects at prose comprehension comprehension group group with with the AGE AGE and UNITS UNITS projects at SUMEX. SUMEX. Feigenbaum Feigenbaum and Nii Nii have visited visited Colorado, Colorado, and one of of us (Miller) (Miller) attended attended the AGE AGE workshop workshop at at sur"mx. Both of these meetings have been very valuable in increasing our understanding SUMG.X. Both of meetings very valuable increasing our understanding of problems might by the various of ilow how our our problems might best be solved solved by various systems systems available available at at SUMEX. SUMEX. \Ve also hope that We that our our experiments experiments with with the AGE AGE and UNITS UNITS packages have been helpful helpful to the development projects. development of of those projects. \Ve \Ve should should also mention mention theoretical theoretical and experimental experimental insights insights that that we have received received from project. The from Alan Alan Lesgold and other other members members of of the SUMEX SUMEX SCP project. The initial initial comprehension by Dr. comprehension model model (Miller (Miller & & Kintsch, Kintsch, 1980) has been used by Dr. Lesgold and other ot,her researchers at a.t tht' the University University of of Pittsburgh, Pittsburgh, as as well well as as rest'archers researchers at at Carnegie-Mellon Carnegie-Mellon Univt'rsity, Universit’y, the University University of of Manitoba, Manitoba, Rockefeller Rockefeller University, University, and the University University- of of Victoria. Victoria. Re.'3ource Management Afanagement B. Critique C’ritique of of Resource The The SU!\IEX-AIM SIJ’llEX-AIM resource is clearly clearly suitable suitable for for the current current and future future needs of of our project. \Ve our project. We have found found the staff staff of of SUMEX SUMEX to be cooperative cooperative and effective effective in dea.ling dealing wit.h The wit,h special requirements requirements and in responding responding to to our our questions. The facilities facilities for for communicat.ion on the ARPANET have also facilitated collaborative communicat.ion ARPANET fa.cilitated collaborat.ive work work with with investigators investigators throughout throughout the country. country. ill. III. RESEARCH RESEARCH PLANS PLANS A. Long Range Range Projects Projects Goals Plans A. Long Goals and and Plans The by the prose comprehension The use use of of SUMEX SUMEX by comprehension group group was greatly greatly reduced reduced in the 139 E. A. A. Feigenbaum Feigenbaum CLIPR CLIPR - Hierarchical Hierarchical Models Models of of Human Human Cognition Cognition 5P41 RR00785-11 RR00785-11 two years, because because the the focus of of the work work during during that that period period was on experimental experimental work work two book writing, writ.ing, rather rather than than computer computer simulation. simulation. This This will will change in the the fall fall of of 1984, and book when a new research associate will will join project whose primary primary responsibility responsibility will will be in when join the project cont,inuing the modelling modelling work work started started in previous previous years years with with J. Miller Miller (who (who is no longer longer continuing with us). Thus, Thus, we expect expect a level of of activity activity comparable comparable to to previous previous years years next next associated with fall. fall. E. A. Feigrnbaum Feigpnbaum 140 5P41 5P41 RR00785-11 RR00785-11 CLIPR CLIPR -- Hierarchical Hierarchical Models Models of of Human Human Cognition Cognition The The primary primary goal goal of of the the device device complexity complexity project project isis the the development development of of aa theory theory of of the the processes processes and and knowledge knowledge structures structures that that are are involved involved in in the the performance performance of of routine routine cognitive cognitive skills skills making making use use of of devices devices like like word word processors. processors. We We plan plan to to model model the the useruserdevice interaction by representing the users processes and knowledge as a production device int,eraction by representing the users processes and knowledge as a production system system and and the t.he device device as as aa nested automata. automata. We We are are also also studying studying the role of of mental mental models in in learning learning how to use use them. B. Justification and Requirements Justification Requirementa for for Continued Continued SUMEX SUMEX Use Use The The research of of the prose comprehension comprehension project project isis clearly clearly tied to continued cont.inued access access to t.he the AGE AGE and UNITS UNITS packages, which which are simply simply not available available elsewhere. We hope that that our our continued continued use use of these systems will will be be offset offset by the input input we have been been and will will continue continue to provide provide to those projects: projects: our our relationship relationship has has been been symbiotic, symbiotic, and we look forward forward to its continuation. continuation. C. Needs Keeds and and Plans Plans for for Other Other Computational Computational Resources Resources \Ve currently currently use use two two other other computing computing systems located at the University Universit,y Colorado. Colorado. One is is the Department Department of of Psychology's Psychology’s VAX VAX 11/780, 11/780, which which is used used primarily primarily run run real-time rea.l-t,ime experiments experiments to be modeled on SUMEX. SUMEX. The The second is is the University University Colorado's Colorado’s CDC CDC 6400, 6400. which which is is used used for for various various types types or of statistical statistical analysis. of to of When When the ARPA-sponsored ARPA-sponsored Vax/Interlisp Vax/Interlisp project project is completed, completed, we would would be most most int,erested in experimenting experimenting with with becoming becoming a remote remote AGE/UNITS AGE/UNITS site. It It would would seem that that this this sort sort of dev('lopment devciopment is the ultimate ultimate goal of of the package projects, projects, and this this type type of int.eract.ion, intera,ct.ion, once it it becomes feasible, would would be a logical logical extension extension of of our our association association wit.h with the SUMEX SUhIEX fa.cility. fa.cilit,y. D. Recowlwzendations Recommendations for for Future Future Community Resource Development Development and Resource D. Community and Our prima.ry recommendation Oiur primary recommendation for for ruture future development development within within SUMEX SUMEX involves involves (a) t.he t,he cont,inued cont,inurd support support of of INTERLISP, INTERLISP, which which is needed for for AGE AGE and for for other other work work we have underway underway on SUMEX SUMEX and (b) (b) the continued continued development development of of the the AGE AGE and UNITS UNITS projects. In In particular, particular, we would would like like to to see see an extension extension of of AGE AGE to to include include a wider wider projects. variety of cont.rol structures so that our psychological models would not be confined to variety of control structures that our psychological would not confined to one particular part iclliar view view of of knowledge-based knowledge-based processing. The The limited limited physical physical capacity capacity of of SUr-,fEX, both both in terms terms of of address space and overloading, overloading, is, as before, before, a major major problem. problem. SUhlES, The prose comprehension comprehension group group can no longer longer use the the publicly publicly released AGE/UNITS AGE/UNITS The system due to to its its severely severely limited limited address space, and and has had had to to build build a personal personal AGE AGE system system from from a stripped-down stripped-down version version of of Interlisp Interlisp and and a selected subset subset of of AGE AGE and system UNITS. We We heartily heartily endorse the the plans plans underway underway to to obtain obtain more more computing computing capacity capacity for for UNITS. the SUMEX SUl'vlEX project. project. the Given our our acquisition acquisition of of a VAX, VA.X, . we particularly particularly support support the the ongoing ongoing and and continued continued Given development of of INTERLISP INTERLISP for for the the VAX, VAX, so that that local local use of of AGE AGE and and UNITS UNITS would would be development possihle. Since we, as well well as other other psychologists, psychologists, need the the real-time real-time capability capability of of possible. VAX/VMS to to run run on-line on-line experiments, experiments, we hope that that the the INTERLISP INTERLISP system system to to be VAX/VhIS developed will will be compatible compatible with with VMS. VMS. Note Note that that this this need for for real-time real-time work work coincides coincides developed with real-world applications of SUMEX programs, in which a VAX might be devoted to with real-world applications of SUMEX programs, in which VAX might devoted to both real-t.ime real-time patient patient monitoring monitoring and and diagnostic diagnostic systems systems such such as PUFF PUFF or or MYCIN. MYCIN. both 141 E. A. A. Feigenba.um Feigenbaum E. 5P41 5P41 RR00785-11 RR00785- 11 Rutgers Rut,gers Research Research Resource Resource II.A.2.3. Rutgers II.A.2d. Rutgers Research Research Resource Resource Rutgers in Biomedicine Rutgers Research Research Resource--Computers Resource--Computers Biomedicine Principal Saul Principal Investig.ators: Investigators: Saul Amarel Amarel [1982-83], [1982-831, Casimir Casimir Kulikowski, Kulikowski, Sholom Sholom Weiss Weiss [1983-84]. [1983-841. Rutgers Rutgers University, University, New New Brunswick, Brunswick, New New Jersey Jersey I. SUMMARY SUMMARY OF OF RESEARCH RESEARCH PROGRAM PROGRAM A. Goals Goals and and Approach Approach The The fundamental fundamental objective objective of the Rutgers Rutgers Resource is to develop a computer computer based framework framework for significant significant research in the biomedical biomedical sciences sciences and for for the application application of resea,rch resea.rch result,s results t.o t.o the solution solution of important important problems problems in health health care. The The central central concept concept is in artificial is t.o to introduce int,roduce advanced advanced methods methods of computer computer science science - particularly particularly artificial intelligence int,elligence into into specific specific areas of biomedical biomedical inquiry. inquiry. The The computer computer is used as as an integral integral part part of the inquiry inquiry process, both both for for the development development and organization organization of knowledge knowledge in a domain domain and for its utilization utilization in problem problem solving solving and in processes processes of of experimentation experimentation and theory theory formation. forma.tion. At At, present, present., the total total number number of investigators investigators who who participate participate in scientific scientific activities activities of t.he outside investigators investigators t.he Resource is 83, of of these, 20 have Rutgers Rutgers appointments, appointments, 21 are outside who mainly located located at at Rutgers, Rutgers, and who participate particips.te in collaborative colla,borative research projects projects that that are mainly 42 are investigators from colla.borative national AIM projects that are located different investigators from collaborative national AIM projects that located in different parts members in Administrative, Administrative, parts of of the country. country. In addition, addition, the the Resource has 12 other other members Computer functions. Thus, Thus, Computer Systems/Operations Systems/Operations and general general programming programming and secretarial secretarial functions. Rutgers Resource community community numbers numbers at at present present a total total of of 95 participants. participants. the Rutgers activities include include research projects projects (collaborative (collaborative research research and and core Resource activities res(~arch) training/dissemination training/dissemination projects,projects, and computing computing services in in support support of of user rese;r.rch) projects. projects. B. Medical Afedical Relevance Reiwanre and and Collaborations Collaborations B. In 1883-84 1983-84 we continued con tinned the the development development of of several several versatile versatile systems systems for for building building In and t,esting testing consult,ation consult.ation models models in biomedicine. biomedicine. The The EXPERT EXPERT system system ha.s has had had many many of of a.nd its ca.psbilities capabilities enhanced in the course of of collaborative collaborative research research in in the the areas of of its rheumatology, ophthalmology, ophthalmologs, and clinical clinical pathology. pathology. rheumatology, In ophthnlmology opht.halmology we have developed developed a knowledge knowledge representation representation scheme for for In treatment pb,nning which is both natural and efficient for encoding the strategies for treatment planning which both natural and efficient for encoding the strategies for choo:3ing among competing and cooperating treatment plans. This involves a ranking of choosing a.mong competing cooperating treatment plans. This involves ranking of treatments according according to to their their characteristics characteristics and and desired desired effects effects as well well as treatments contra indications. Kastner Kastner has generalized generalized the the scheme so that that itit is now now being being used used for for a cont,raindications. number of of reasoning reasoning models: models: infectious infectious eye disease, primary primary eye care, care, and and rheumatology rheumatology number m;1nagement. Our Our main main collaboration collaboration continues continues to to be with with Dr. Dr. Chandler Chandler Dawson Dawson of of the the management. Proct.or Foucdation, Foundation, UCSF. UCSF. Proctor In rheumatology, rheumatology, our our collaboration collaboration with with Drs. Drs. Donald Donald Lindberg Lindberg and and Gordon Gordon Sharp Sharp In at the the Universit’ University of Missouri-Columbia Missouri-Columbia has has continued continued at at a very very active active level. level. The The model model at y of which now now includes includes detailed detailed diagnostic diagnostic criteria criteria for for 26 26 major major for rheumstologicsl ri1ellmatological diseases which for E. A. A. Feigenbaum Feigenbaum E. 142 Rutgers Rutgers Research Research Resource Resource 5P41 5P11 RH00785-11 RROOi85- 11 diseases, diseases, had had the the management mana.gement advice advice and and treatment treatment planning planning developed developed further. further. Dr. Dr. Sha.rp's Sha.rp’s group group continues continues to to develop develop the the knowledge knowledge base base in in this this area, area, with with formalization formalization of of the the knowledge knowledge carried carried out out in in conjunction conjunction with with Dr. Dr. Lindberg's Lindberg’s group group and and the the Medical Medical Expert The Expert Systems Systems Group Group at at Rutgers. Rutgers. The Resource Resource researchers researchers have have developed developed new new representational representational elements elements for for EXPERT EXPERT in in response response to the needs needs of of the rheumatology rheumatology research, and and Politakis Politakis has has developed developed aa coordinated coordinated system system called called SEEK SEEK (System (System for for Empirical Empirical Experimentation Experimentation with with Expert Expert Knowledge.) Knowledge.) which which provides provides interactive interactive assistance assistance to the human human expert expert in in testing, t,esting, refining refining and and updating updating aa knowledge knowledge base haze against against aa data dat,a. base bsse of of trial trial cases. ca.ses.SEEK SEEK has has been been tested and and extended extended during during the past year. In clinical Robert Galen clinical pathology pathology our our main main collaboration collaboration has has been with with Dr. Robert (Cleveland Clinic Foundation), with whom we have developed the serum serum protein protein (Cleveland Clinic Foundation), with whom electrophoresis electrophoresis model which which isis incorporated incorporated into into an instrument instrument - the scanning scanning densitometer This densitometer manufactured manufactured by Helena Laboratories. Laboratories. This instrument instrument with with interpretive interpretive reporting reporting capabilities capabilities has now been been on the market market for over aa year, isis located at over 100 100 clinical clinical sites, and represents represents the first first known known spin-off spin-off of AI AI expert expert systems research in the t.he field of laboratory instrumentation. 'vVe continue laboratory instrumentation. We continue to refine refine the representational representational mecha.nisms model. mecha.nisms used used for for this this kind kind of model. In biomedical biomedical modeling modeling applications applications we are experimenting experimenting with with several prototype prototype models for for giving giving advice on the interpretation interpretation of of experimental experimental results results in the field field of enzyme kinetics. kinetics. in conjunction conjunction with with Dr. Dr. David David Garfinkel. Garfinkel. His PENNZYME PENNZYME program program has been linked linked to a model in EXPERT, EXPERT, which which allows allows the user to interpret interpret the progress of the model analysis. C. Highlights of Research Research Progress Progress Highlights of Expert Expert Medical Medical Systems Systems (C. (C. Kulikowski, Kulikowski, S. Weiss) Weiss) Research has continued problems of continued on problems of representation, representation, inference inference and control control in expert systems. Emphasis has been placed this year on problems of knowledge expert Emphasis placed this year problems of knowledge base acquisition, acquisition, empirical empirical testing testing and refinement refinement of of reasoning reasoning (the (the SEEK SEEK system), system), and treat.ment planning planning st,ra.tegies strategies over over time. time. From From a technological technological point point of of view view the the market, market trea.t.ment availability of of the int.erpretive interpretive reporting reporting version version of of a scanning scanning densitometer, densitometer, and the the availability development of of modeis models for for eye care consultation consultation that that run run on microprocessor microprocessor systems systems development (Apple IIc, lIe, IBM-PC) IBM-PC) represents represents an important important achievement achievement for for AIM AIM research in showing showing its its (Apple practical impact imp:1ct in medical medical applications. applications. This This was recognized recognized by by the award award of of a scientific scientific practical exhibit prize prize at. at, the Academy Academy of of Ophthalmology Ophthalmology Annual Annual Meeting Meeting in November November 1983. exhibit SEEK: A A System System for for Empirical Empirical Experimentation Experimentation with with Expert Expert Knowiedge Knowledge 1.1) SEEK: SEEK is a~t system syst.em which which has been developed developed t.o to give give interactive interactive advice advice about about rule rule SEEK refinement daring the design of an expert system. The advice takes the form of refinement during the of expert system. The advice takes the form of suggestions for possible experiments in generalizing and specializing rules in an expert suggestions for possible experiments in generalizing and specializing rules in expert model that that has been specified specified based on on reasoning reasoning rules rules cited cited by by aa. human human expert. expert. Case model ('xperience, in the the form form of of stored stored cases cases with with known known conclusions, conclusions, is used to to interactively interactively experience. guide the the expert expert in refining refining the the rules rules of of a model. model. The The clesign design framework framework of of SEEK SEEK consists consists gl.;idtx of a tabular tabular model model for for expressing expressing expert-modeled expert-modeled rules rules and and a general general consultatiou consultation system system of for a.pplying applying a model model to to specific specific cases. cases. This This approach approach has proven proven particularly particularly valuable valuable in in for assisting the the expert expert in in domains domains where where the the logic logic for for discriminating discriminating two two diagnoses is difficult difficult assisting to specify; specify; and and we have have benefited benefited primarily primarily from from experience experience in in building building the the consultation consultation to system in in rheumatology. rheumatology. system Treatment Planning Planning 1.2) Trea.tment The ranking ranking and and selection selection strategies strategies developed developed as as aa stand-alone stand-alone system system last last year year The 143 E. A. A. Feigenbaum Feigenbaum E. Rutgers Rut,gers Research Resource 5P41 RR00785-11 have been Capabilities Capabilities for expressing been incorporated incorporated into into the EXPERT EXPERT framework. framework. reasoning over time time have been been added, so stored stored chart chart reviews can be be carried carried out out automatically, a.ut~omat.ically. summarizing summarizing various various patterns patterns of findings findings over over time, time, and abstracting abstracting the major major features features of interest interest for prognostic prognostic advice or treatment treatment recommendations. recommendations. Applications Applications have been in infectious infectious eye eye disease disease modeling, modeling, rheumatology rheumatology treatment, treatment, and sequential sequential advice in interpretation interpretation and sequencing of cardiac cardiac enzyme tests (e.g. CPK/LDH CPK/LDH isoenzymes). 1.3) Technology Technology Transfer Transfer Important Import.ant technology technology transfer transfer milestones milestones have also been achieved this this year: year: the instrument EXPERT inst.rument interpretation interpretation EXPERT program program for for serum serum protein protein has been widely widely disseminated disseminated after after being made available available by Helena Laboratories, Laboratories, based on the prototype prototype program program developed by us; us; and we have succeeded succeeded in transferring transferring a large knowledge knowledge base base in rhel1matology hypotheses and 1000 rules) rules) onto onto a rheuma.tology (about (about 1000 findings, findings, 400 hypotheses mi<'roprocessor microprocessor (rvtotorola (Motorola 68000) based system system - the WICAT WICAT - which which is well well within within the means of clinical clinical researchers and practitioners. practitioners. This This system system has been on site at the University University of r-.lissouri Missouri during during the last last year year for for testing testing and refining refining of of the knowledge knowledge base. base. 1.4) 1.4j Learning Learning with with Prior Prior Structural Structural Knowledge Knowledge This This approach approach to knowledge knowledge acquisition acquisition and representation representation has as as its goal to allow allow the expert just the elements expert to specify specify just elements that that are to enter enter into into the reasoning reasoning model, model, with with a few causal and taxonomic taxonomic relations. relations. These should should then then be be sufficient sufficient to guide a learning learning program which operat,es on a data base of cases with known program which operates data base of cases with known end-points. end-points. Such an approach approach wonld woiiid he be w3eful ii,seful in situations sit:uations where where the expert expert either either has little little time time to explicitly explicitly formulate formulate decision rules, or program [Drastal or finds finds it it difficult difficult to do so. so. Our Our program [Drastal and Kulikowski, Kulikowski, 1982] 19821 uses uses aa. blackboard blackboard representation, representation, with with multiple multiple knowledge knowledge sources to to handle handle the different different conclusions, pertain to conclusions, and the formation formatsion of of rules from from the data data that that pertain to them. them. We We have tested tested this this scheme in the areas of of glaucoma glaucoma and rheumatology, rheumatology, and shown shown that that there there are some int.eresting provided by interesting tradeoffs tradeoffs between between the degree of of a-priori a-priori structure structure provided by the expert, expert, and snd the t.he complexity complexity of of rule rule generation. generation. In relat ion to relation to alternative alternn.tive means mea.ns of of of of additional a.dditional medical medical a system preprocessing or system like like SEEK, SEEK, this this approach approach represents represents a preprocessing or developing prototype model. We developing the prot.otype We are now now investigating investigating the role semantic semantic constraints constraints on the strategies strategies of of rule rule generation. generation. 2) Artificial Artificial 1ntelligence:Expertise Intelligence:Expertise Acquisition Acquisition and Problem Problem Reformulation Reformulation (S. Amarel) Amarei) The main main research research activity activity in this this area is concerned concerned with with improvements improvements in problem problem The solving expertise expertise via via shifts shifts in problem problem representation, representation, i.e., via via reformulation. reformulation. solving this research, we have concentrated concentrated on the developmental developmental processes processes that that lead to to In this formation of of specialized specialized high high performance performance procedures procedures in sub-domains SUb-domains of of a problem problem the forma.tion ci::lss. Theory Theory formation formation is a, a key key task task in these processes; processes; and we are now now studying studying several several CIXS. approaches to this this task task - both both top-down, top-down, model model guided, guided, approaches approaches and ‘'bottom-up' approaches bottom-up’ methods that that are based on detailed detailed analysis analysis of of individual individual cases. cases. methods D. lip-to-Date Up-to-Date List List ofof Publications Publications D. The following following is an update update of of publications publications in the the Rutgers Rutgers Resource for for the the period period The 198·1 (only (only publications publications not not listed listed in previous previous SUMEX SUMEX annual annual reports reports are 1983 and 198.1 presented here). presented Wei5s, S.M. S.M. and and Kulikowski, Kulikowski, C.A. C.A. A A Practbcal Practical Guide Guide to Designing Designing Ezpert Expert 1. Weiss, Systems, Rowman Rowman a.nd and Allanheld, Allanheld, 1984. Syatemu. Feigenbaum A. Feigenbaum E. A. 144 Rutgers Rutgers Research Research Resource Resource 5P41 5P41 RR00785-11 RR00785- 11 2. 2. Kulikowski, Kulikowski, C.A. C.A. contributor contributor to to the the Knowledge Knowledge Acquisition Acquisition chapter chapter edited edited by by B. B. Buchanan Buchanan in in the the book book Building Building Expert Expert Systems Systems (F. (F. HayesHayes- Roth, Roth, et et aI., al., cds) eds) Addison-\o\Tesley, Addison-Wesley, 1983 1983 (in (in press). press). Strategies of 3. 3. 1'30, l’a.0, Y. Y. and and Kulikowski, Kulikowski, C.A., C.A., "lYfultiple ’ Multiple Strategies of Reasoning Reasoning for for Expert Expert Sy.'3tems", Systems “, Proc. Proc. Sixteenth Sixteenth Hawaii Hawaii International International Conference Conference on Systems Sciences, Sciences, pp. pp. 510-514 510-514 ,, 1983. 1983.** 4. 4. Kulikowski, Kulikowski, C.A. C.A. "Progress “Progress in in Expert Expert AI Al Medical Medical Consultation Consultation Systems: Systems: 1980 1980 -- 1983 1983 If,“, Proc. Proc. MEDINFO MEDINFO '83 ‘83 , pp. pp. 499-502, 499-502, Amsterdam, Amsterdam, August August 1983.* 1983.t 5. 5. Kastner, Kastner, J.K., J.K., \Veiss, Weiss, S.M., and Kulikowski, Kulikowski, C.A., C.A., "An “An Efficient Efficient Consultation Systems", Proc. MEDINFO Time-Dependent Time-Dependent Consultation Systems I, MEDINFO '83, ‘83, 1983. * 1983.’ Scheme for for pp.619-622, 6. 6. K1Jlikowski, Kulikowski, C.A. C.A. "Expert “Ezpert Medical Medical Consultation Consultation Systems", Systems “, Journal Journal of Medical Medical Systems, v.7. pp. 229-234, 1983.* 7. Weiss, S.M., Kulikowski, Kulikowski, C.A., C.A., and Galen, R.S., "Representing “Representing Expertise Expertise in in a Computer Computer Program: Program: The Serum Serum Protein Fbotein Diagnostic Diagnostic Program", Program”, Journal Journal of Clinical Clinical Laboratory Laboratory Automation, Automation, v.3, pp. 383-387, 1983.* 8. Kastner, for Kastner, J.K., J.K., Weiss, S.M., and Kulikowski, Kulikowski, C,A., C.A., "” lin An Expert Expert System System for Front-line Front-line Health Health Workers Workers in in Primary Primary Eye Care", Care “, Proc. Proc. Seventeenth Seventeenth Hawaii Hawaii International 1nt.ernationn.i Conference Conference on Systems Sciences, Sciences, pp. 162-166, 1984. 1984.** 9. Kulikowski, Kuiikowski. C.A. C.A. "Knowledge “K,nowledge Acquisition Acquisition and and Learning Learning in in EXPERT", EXPERT”, Proc. Proc. 1983 'Workshop IIIinois,Champaign-Urbana Workshop on Machine Machine Learning, Learning, Univ. Univ. of of Illinois,Champaign-Urbana 1983. credit. Indicat.e Indicate by by an a,n asterisk asterisk (*) (*) that that the resource was given given credit. E. Funding Funding Support Support Since December December 1983, the Rutgers Rutgers Research Resource on Artificial Artificial Intelligence Intelligence in funded under under grant grant RR RR 02230-01 from from the Division Division of of Research Resources, Medicine is funded Medicine Biotechnology Resources Program. Program. Principal Principal Investigators Investigators are Casimir Casimir A. A. Kulikowski, Kulikowski, Biotechnology Proffssor of of Comput,er Computer Science and Chairman Chairman of of the Department Department of of Computer Computer Science Profrssor Dr. Shoiom Sholom M. M. Weiss, \Veiss, Associate Associate Research Professor Professor of of Computer Computer Science. [1984-87]' and Dr. [1984-871, The tota. total direct direct costs for for the the period period 1983-87 is The current period (December I, 1983 November current period (December 1, November 30, 1984) $3,198,075, with with the the total total for for the the $3,198,075, being $ 989,276. being The Rutgers Rutgers Resource was funded funded until until December December 1983 through through an NIH NIH grant grant The entitled"“Rut.gers Rut,gers Research Resource on on Computers Computers in in Biomedicine” Biomedicine" -- number number P41RR643. P41RR643. entitled The Co-Principa.1 Co-Principal Investigators Investigators were were Dr. Dr. Saul Saul Amarel, Amarel, Professor, Professor, Chairman Chairman of of the the The Department of of Computer Computer Science, and Director Director of of the the Laboratory Laboratory for for Computer Computer Science Department Research, and Dr. Dr. Casimir Casimir Kuiikowski, Kulikowski, Professor Professor of of Computer Computer Science at at R.utgers. Rutgers. Kesezrch, II. INTER4CTIONS INTERA.CTIONS WITH WITH THE THE SUMEX-AIM SUMEX-AIM RESOURCE RESOURCE II. A. hledical Aledical Collaborations Collaborations and and Dissemination Dissemination A. The SUMEX-hIM SUMEX-AIM facility facility provides provides a backup backup node node where where some of of our our medical medical The collaborators can access programs developed at Rutgers. The bulk of the medical collaborators ca.n access programs developed at Rutgers. The bulk of the medical collaborative work work outlined outlined in in LB. LB. above above is centered centered at at the the Rutgers Rutgers facility facility (the (the RutgersRutgerscollaborative AIM node). node). AIM 145 A. Feigenbaum Feigenbaum E. A. E. RR00785-11 5P41 RR00785-11 Rutgers Rutgers Research Resource Dissemination Dissemination activities activities continue continue to to be an important important responsibility responsibility of of the Rutgers Rutgers Resource within wit)hin t,he t,he AIM AIM community. community. The The following following activities activities took took place in the last year: year: 1. Ninth AIM 1. Ninth ,4IM \Vorkshop Workshop (1983): Organized Organized by by Dr. Dr. Casimir Casimir Kulikowski, Kulikowski, it it was held in Baltimore, Baltimore, in conjunction conjunction with It It consisted consisted of of a series of of working working group group with the SCAMC SCAMC 83 meeting. meeting. discussions followed followed by by summary summary presentations presentations by by members members of of the AIM AIM community community on their their conclusions. conclusions. 2. Hawaii Hawaii International International Conference Conference On On Systems Sciences: Sciences: presented a paper paper on the expert Dr. Dr. \Veiss Weiss presented expert system system for for front-line front-line health health workers, workers, and Dr. Dr. Kulikowski Kulikowski chaired chaired a session session on knowledge knowledge based medical medical systems. 3. VII-pn,n-American VII-Pan-America,n Congress on Rheumatology: Rheumatology: Dr. presented the Dr. Sharp Sharp present,ed the rheumatology rheumatology knowledge knowledge base and consultation consultation program at this program this meeting. meeting. 4. At Amarel was elected member member of of the Executive Executive At the AAAI-82 AAAI-82 meeting, meeting, S. Amarel Council Council of of AAAI. AAAI. He is also General General Chairman Chairman of of IJCAI-83 IJCAI-83 which which was held held in Karlsruhe, Karlsruhe, W. W. Germany Germany in August August 1983. Dr. Dr. Kulikowski Kulikowski was the organizer organizer for for an expert expert medical medical systems session session at at MEDINFO MEDINFO 83. B. National .11.\1 Projects at Rutgers Rutgers B. National AIM Projects The projects, approved· The national national AIM AI&l projects, approved. by by the AIM AIM Executive Executive Committee, Committee, that that are associated with with the RutgerS-AIM Rutgers-AIM node are the the following: following: 1. INTERNIST jCADUCEUS project, project, headed by INTERNISTjCADUCEUS by Dr. Dr. Myers Myers and Dr. Dr. Pople Pople from from the University University of of Pittsburgh, Pittsburgh, has been using using the Rutgers Rutgers Resource as a backup backup system system for for development development, and experimentation. experimentation. 2. Medical project, headed by Medica. Knowledge Knowledge Representation Representation project, by Dr. Dr. Chandrasekaran Chandrasekaran from from Ohio Ohio State State University, University, is doing doing most most of of its its research on the Rutgers Rutgers system. system. 3. PURSUIT project, directed PURSUIT project, directed by by Dr. Dr. Greenes from from Harvard Harvard University, University, is doing doing most most of of its its research on a Goal-Directed Goal-Directed Model Model of of Clinical Clinical Decision-Making Decision-Making at Rutgers. Rutgers. 4. Biomedical Biomedical Modeling, Modeling, by by Dr. Dr. Garfinkel Garfinkel from from the the University University of of Pennsylvania. Pennsylvania. 5. At.tending At,tending Project, Project, directed directed by by Dr. Dr. Perry Perry Miller Miller of of the Yale Yale Medical Medical Center, Center, is doing physician's plan of doing much much of of the the research on critiquing critiquing a physician’ s plan of management management at Rutgers. Rutgers. 6. MEDSIM project: MEDSIM project: This pilot project project designed to provide provide resource-sharing This is a pilot resource-sharing a.nd and community community building building facilities facilities for for about about 25 researchers researchers in bio-mathematical bio-mathematical modeling modeling and simulation. simulation. C. Critique Resource Management Management Critique of of SUMEX-AIAf SUMEX-AIM Resource Rutgers using DEC-20 primarily for SUMEX DEC-PO system system primarily for Rut,gers is currently currently using the SUMEX communication communication with with other other researchers in the AIM AIM community community and with with SUMEX SUMEX staff, staff, and E. A. Feigenbaum Feigenbaum 146 Rutgers Rutgers Research Research Resource Resource SP41 5P41 RR0078S-11 RROO785-11 also also for for backup backup computing computing in in demonstrations, demonstrations, conferences conferences and and site site visits. visits. Our Our usage usage isis currently currently running running at at less less than than 50 50 connect connect hours hours per per year year at at SUMEX, SUMEX, with with an an overall overall connect/CPU connect/CPU ratio ratio of of about about 30. 30. Rutgers Rutgers isis beginning beginning to to place place more more emphasis emphasis on on the the use use of personal personal computers, computers, and on on network network support support needed needed to to make make these these effective. effective. Sumex Sumex has has been been help in the following following wa.ys: ways : •l The The AIM AIM Executive Executive Committee Committee allocated allocated to to the the Rutgers-AIM Rutgers-AIM node node one one of the Xerox Xerox Dolphins Dolphins acquired acquired by by SUMEX, SUMEX, to to help help us us develop experience in supporting This machine was used used almost almost entirely entirely to help support,ing personal machines. This us us develop and test network network support(We support(We are using Ethernet Ethernet with with the Xerox Xerox PUP PUP networking networking protocols), protocols), and subsequently subsequently returned returned to SUMEX. SUMEX. •l Most Most of network network software software that that we use use Having this software available Having t,his software availa.ble has saved was originally originally developed at SUMEX. SUMEX. us an enormous enormous amount amount of time. time. •l Initia.lly Initinlly SUMEX SUMEX was very very helpful helpful in giving giving us advice about about setting setting up our our Ethernet Ethernet and the Dolphins. Dolphins. m. III. RESEARCH RESEARCH PLANS PLANS A. A. Project Pro;lect Goals Goals and a,nd Plans Plans We We are pla.nning planning to t,o continue continue along the main main lines of of research that that we have est.ablished est.n.blished in the t,he Resource to date. Our Our medical medical collaborations collaborations will will continue continue with with emphasis on development development of expert expert consultation consultation systems in rheumatology, rheumatology, ophthalmology ophthalmology and clinical inference and planning planning will will clinical pathology. pathology. The The basic AI AI issues issues of of representation, representation, inference continue Our with emphasis on further further continue to receive attention. attention. Our core work work will will continue continue with development representations and development of of the t,he EXPERT EXPERT framework framework and also on AI AI stUdies studies in representations problems of knowledge and expertise acquisition. We propose to work number of of to work on a number problems knowledge expertise acquisition. We terhnology will be affordable affordable by by our our technology transfer transfer experiments experiments to micro micro processing processing that that will biomedka.l research snd and clinical clinical collaborators. collaborators. We We also plan plan to to continue continue our our participation participation biomedica.1 AIM dissemination dissemination and training training activities activities as well well as our our contribution contribution -- via via the the in :\IM RUTGERS/LCSR computer computer -- to to the shared computing facilities facilities of of the national national AIM AIM RUTGERS/LCSR shared computing network. network. B. Justification Justification and and Requirements Requirements for for Continued SUMEX Use B. Continued SUMEX Continued access access to to SUMEX SUMEX is needed for: for: Continued Ba.ckllP for for demos, et,c. et.c. 1. Rack~lp Programs developed developed to to serve the National National AIM AIM Community Community should should be runnable runnable 2. Programs both facilit.ies. facilities. on both There should should be joint joint development development activities activities between between the the staffs staffs at at Rutgers Rutgers and and 3. There SUMEX in order order to to ensure portability, portability, share the the load, load, and and provide provide a wider wider SUMEX variety of of inputs inputs for for developments. developments. variety C. Needs Needs and and Plans Plans for for Other Other Computing Computing Resources Resources Beyond Beyond SUMEX-AIM SUAIEX-AL\1 C. Our comput,ing computing is going going to to move move in in the the direction direction of of personal personal computers. computers. We vVe will will Our continue to to use Sumes Sumex for for backup backup purposes, purposes, however. however. continue D. Recommendations Recommendations for for Future Future Community Community and and Resource Resource Development Development D. 147 E. A. A. Feigenbaum Feigenbaum E. Rutgers Research Resource Rutgers 5P41 RR00785-11 RR00785- 11 Use of of personal personal comput,ers compubers and minicomputers minicomputers is continuing continuing to to grow grow in the the AIM AIM community. We find find that that the the biggest biggest challenge challenge is supporting supporting these systems. Although Although community. vVe central computing computing will will continue continue to to be needed for for communication communication and coordination, coordination, some central that over over the next next few years years all AIM AIM research projects projects and even individual individual we believe that collaborators collaborators will will come to to have their their own own hardware. hardware. However However many many of of these community community members (particularly (particularly the the collaborators) collaborators) will will not not be in a position position to to support support hardware hardware or or members software software on their their own. own. We We would would certainly certainly expect expect SUMEX SUMEX to t,o continue continue to to provide provide expert expert. this area. However However we believe it it would would be helpful helpful for for SUMEX SUMEX to to have a formal formal advice in this program to support support smaller smaller computers computers in the field. field. We We envision envision this this as including including at least program following items: it,ems: the following A central central source of of information information on hardware hardware and software software that that is likely likely to to be •l A of interest interest to the AIM AIM community. community. SUMEX might might want want to become a of SUMEX distribution dist,ribution point point for for certain certain of of this this software, software, and even help coordinate coordinate quantity quantity purchase purchase of of hardware hardware if if this this proves proves useful. useful. •l Assistance Assistance in support support of of hardware hardware and software software in the the field. field. Depending Depending upon upon the hardware hardware involved, involved, this this might might involve involve advice over over the telephone telephone or or actual actual board-swapping board-swapping by by mail. mail. \Vith With our our Dolphins Dolphins we have found found that that there there are a number number of of problems problems that that can be resolved resolved over over the telephone telephone if if we can find find someone with with appropriate appropriate expertise. expertise. E. A. Feigenbaum Feigenbaum 148 Rutgers Rutgers Research Resource 5P41 5P4 1 RR0078S-11 RR00785- 11 ll.A.2.4. SECS: & of II.A.2.4. SECS: Simulation Simulation St Evaluation Evaluation of Chemical Chemical Synthesis Synthesis SECS SECS - Simulation Simulation and and Evaluation Evaluation or of Chemical Chemical Synthesis Synthesis Project Project Principal W. Principal Investigator: Investigator: W. Todd Todd Wipke Wipke Board Board or of Studies Studies in in Chemistry Chemistry University University or of California California Santa Santa Cruz, Cruz, CA. CA. 95064 95064 Coworkers: Coworkers: l. I. Kim Kim D. Rogers J. Chou Chou M. hl. Hahn Hahn M. hf. Yanaka >‘anska l. I. Iwataki Iwataki 1. I. (Grad (Grad student) student) (Grad (Grad Student) Student) (Postdoctoral) (Postdoctoral) (Grad (Grad Student) Student) (Postdoctoral) (Postdoctoral) (Postdoctoral) (Postdoctoral) SUMMARY OF SUMMARY OF RESEARCH RESEARCH PROGRAM PROGRAM A. Project Project Rationale Rationale -4. \Vith the SECS project project our principles of With our long range goal is to to develop the logical logical principles of molecula.r practical computer programs to molecu1a.r construction construction and to to use use these in developing developing practical computer programs to assist bio-organic molecules. a.ssist investigators investigators in designing designing stereospecific stereospecific syntheses of of complex complex bio-organic Our project, is aimed Our second area a.rea of of research, the XENO XENO project, aimed at at improving improving methods methods for for predicting potential potential biologica. biological activity plausibility of predictsing activity of of metabolites metabolites and plausibility of incorporation incorporation and excretion excretion of metabolites. metabolites. B. Medical Afedical Relevance Relevance and and Collaboration Collaboration The biological activity The development development of of new drugs drugs and the study study of of drug drug structure structure biological activit,y relationships depends upon the chemist's ability to synthesize new molecules relstionships chemist’s ability to synthesize molecules as as well well as as his to isotopic a.hility incorporating isotopic labels or or other ot,her a.bility to modify nlodify existing existing structures, structures, e.g., incorporating substituents bio-molecular substrates. The Simulation Simulation and Evaluation Evaluat,ion of of Chemical Chemical subst ituents into into bio-molecular substrates. The Synthesis project aims at assisting Synt,hesis (SEeS) (SECS) project assisting the synthetic synthetic chemist chemist in designing designing stereospecific stereospecific synt,heses biologically important The advantages advantages of of this this computer computer approach approach syntheses of of biologically important molecules. The over over normal normal manual manual approaches a.pproa.ches are many: many: 1) greater greater speed in designing designing a synthesis; synthesis; 2) freedom past experience and past solutions; freedom from from bias of of past solutions; 3) thorough thorough consideration consideration of all possibl(' syntheses llsing possible using a more extensive extensive library library of of chemical chemical reactions reactions than than any any individual individual person can remember; remember; 4) greater greater capability capability of of the computer computer to to deal with with the many many struct st.rucl ilHS ures which which result; result,: and 5) capability capability of of computer computer to to see see molecules molecules in a graph graph theoretical projection. theoretical S(,llSC. sense. free from from the bias of of 2-D projection. The using XENO predict the plausible plausible The objective objective of of using XENO in metabolism metabolism studies studies is to to predict metabolites of a given xenobiotic in order that they may be analyzed for metabolites of given xenobiotic order that they may analyzed for possible carcinogenicity. Metabolism carcinogenicity. Metabolism research may may also find find this this useful in the identification identification of of metabolites met,a.bolitcs in that t,lmt it it suggests what what to to look look for. for. Finally, Finally, one may may envision envision applications applications of of this problem domains t,his technology technology in problem domains where where one wishes to to alter alter molecules molecules in order order to to inhibit inhibit certain certain types types of of metabolism. metabolism. C. Highlights Highlights of Research Progress Progress o j Research 149 E. A. A. Feigenbaum Feigenbaum SECS: Simulation Simulation 8l 62 Evaluation Evaluation of of Chemical Chemical Synthesis Synthesis RR00785-11 5P41 RR00785-11 C.1 SECS SECS Project C.l Project Developmenta Developments The The majorit.y majority of of our our research has been aimed aimed at at strategic strategic planning planning in chemical chemical synthesis. Specific work work has included included the SST SST project project for for recognizing recognizing potential potential starting starting synthesis. Specific materials from from a target, target, the MCS MCS project project for for maximal maximal common common subgraph subgraph searching, searching, and a mat.erials project for for rapid rapid substructure substructure search using using parallelism. parallelism. project Starting Material Strategies. The C.l.a Material Strategies. C.1.a SST SST -- Starting The importance importance of of selecting selecting good starting materials materials for for a synthesis synthesis has been known known for for a long time, time, but but only only recently recently has starting work work st.arted started on applying applying computer computer techniques techniques to to the the selection selection process. The The selection selection of of st,arting material material for for a synthesis synthesis is frequently frequent,ly the major major discovery discovery in a synthesis synthesis and the st.arting of converting converting the starting starting material material to the target target is minor minor by by comparison. comparison. Last Last process of year year we reported reported development development of of the SST program program for for selecting selecting starting starting materials materials that that are appropriate appropriate for for a given given synthetic synthetic target target using a library library of of available available chemicals, chemicals, but but without without reference reference to to reactions. reactions. SST SST handles problems problems of of classes classes I-III I-III given given below: below: I) I) Target Target = = SM Identical Identical match match Superstructure match match Superstructure Substructure match match Substructure Similarity Similarity match match II) Target Target > > SM II) III) Target Target < < SM III) IV) IV) None of of these For a search over over our our abstracted abstracted file, file, the identical identical match match means that that the target target For starting materials materials are identical identical except except for for functionalization. functionalization. The The superstructure superstructure match match and sta.rt.ing t.he case case where where we must must make make carbon-carbon carbon-carbon bonds during during a synthesis. synthesis. The The is the substracture match match is the case case where where the the starting starting material material is larger larger than than the the target, target, so substructure carbon-ca,rbon bonds have to to be broken. broken. Finally, Finally, the the similarity similarity match match is where where carboncarboncarbon-carbon carbon carbon bonds have to be both both made made and and broken broken during during the synthesis. synthesis. Our Our research research in efficient efficient starting starting material material strategies strategies has continued continued this this past past year year in two two different different areas. In the first, first, we have explored explored the prospect prospect of of llsing using a.a parallel parallel computer in the graph graph matching matching process described described in the following following section section and in the computer solution to to the class IV IV problem problem (see (see above) which which is second we have developed a solution described in a subsequent subsequent section. section. described C.1.c Subgraph Subgraph Search Search Using Using Parallelism. C.Lc Parallelism. Subgraph Subgraph matching matching is an important important method method llsed used in many many different different computer computer applications in organic organic chemistry. chemistry, including including the the recognition recognition of of functional functional groups, groups, synthesis synthesis applications planning, planning, constraint const.raint testing testing in structure structure generation, generation, selection selection of of starting starting materials materials for for synt,hesis, and structure structure oriented oriented retrieval. retrieval. The synthesis. The fundamental fundamental problem problem is, given given a query query .d.~trutttwe (QS) and a5 candidate candidate superstructure (CS), determine determine if if there there exists a superstructure (CS), substructure (QS) mapping of of the atoms stems (nodes) of of the substructure substructure onto onto the candidate candidate superstructure superstructure such mapping that the connected connected atom atom pairs pairs in the query query substructure substructure are also connected connected in the that superst,ructure, and that that the atom atom and bond bond types types also correspond. correspond. superstructure. Although substructure substructure search is a non-numerical non-numerical problem. problem, it it is computationally computationally Although demanding demanding because because ultimately ult!imately it it involves involves establishing establishing an atom atom by by atom atom correspondence correspondence belween the QS and a.nd the CS, and this this problem problem is a member member of of the class of of NP-complete between NP-complete problems. In a worst worst case case for for N N atoms atoms in the QS and M M atoms atoms in the CS (M>N). (M>N), problems. one may have to t,o consider consider N!/(M-N)! N!/(M-N)! mappings for for each CS. The may mappings The objective objective of of our our research wa,s to t]o explore explore the feasibility feasibility of of applying applying parallel parallel processing processing to this this problem. problem. 'was Although the node ma.tching ma.tching process is an NP-complete problem, if if we eliminate eliminate all a.11 Although NP-complete problem. backtracking, the order order of of the algorithm algorithm reduces to t,o O(N), O(N), where where N number of of backtracking, N is the number Feigenba.um E. A. Feigenbaum 150 RR00785-11 5P41 RR00785-11 SECS: Simulation Simulation & Evaluation Evaluation of of Chemical Chemical Synthesis Synthesis atoms atoms in t.he the subgraph. suhgraph. This This would would represent represent a major major improvement. improvement. Unfortunately, Unfortunately, the algorithm NP-complete with a.lgorithm is now NP-complete with respect respect to to sequential sequential processors. configuration architecture; architecture; a central central processor with with a We We proposed proposed a "star" “star” configuration communication communication line to to a number number of of lower lower processors, with with no direct direct communication communication Each processor processor has a small small amount amount of of memory memory as as a allowed a,llowed between between the lower lower processors. Each which avoids avoids the problems problems inherent inherent in shared shared memory. memory. ",vorking space", The “working space”, which The communication communication packets packets are compact compact to reduce the storage storage and communication communication burden burden on the central cent’ra.1processor. The simulation simulation algorithm algorithm called called MOLSIM MOLSIM was implemented implemented using the SIMULA SIMULA The Isngua.ge. We We studied studied this this algorithm algorithm as a function function of of the number number of processors and and the language. nature nature of of the particular particular graphs graphs being being matched matched (6-31 non-H non-H atoms). atoms). We We found found an average utiliza.tion utiliza.tion of of 84% 84% on a ~5 25 cpu machine machine (figured (figured as as total total processor time/real time/real time), time), bnt but only only 50% SO% on a 50 cpu machine machine although although for for some structure structure matching matching questions, questions, the The average speed enhancement enhancement using this this size machine machine (50 efficiency efficiency rearhed reached 97%. 97%. The factor of of 30. A A real machine machine of of the architecture architecture needed to to run run this this processors) was a factor algorithm algorithm exists exists at Purdue Purdue University, University, and time time is being requested requested to test test the algorithm algorithm in real time. time. This This algorithm algorithm is a unique unique approach approach to to the problem problem of of graph graph matching matching and will will likely become practical pract.ical when when parallel pa.rallel processors are commonly commonly available available and inexpensive. inexpensive. likely (This .. ) C.1.b (This work work is submitted submitted for for publication publication in J. Chern. Chem. Inf. Inf. Comput. Comput. Sci Sci..) Alaximal Common AICS-- Maximal Search. AICS-Common Subgraph Subgraph Search. The second area of of starting starting material material strategy strategy work work this this year year has been in solving solving the The Our solution solution to to this this problem problem involves involves development development of of a class IV IV problem problem given given above. Our efficient maximal maximal common common subgraph subgraph matching matching algorithm. algorithm. chemists represent represent new efficient Since chemists organic organic molecules as as graphs, graphs, computational computational chemists chemists need graph graph theoretical t,heoretical techniques techniques such as as graph graph isomorphism, isomorphism, subgraph subgraph search, and maximal maximal common common subgraph subgraph search. Of Of these three three important import,a.nt procedures. procedures, maximal maximal common common subgraph subgraph search (MCSS) (MCSS) remains remains the rnost difficult difficult and least utilized. utilized. most The ext.ensive ext,ensive comput.a.tional computational demands demands of of MCSS MCSS has restricted restricted its its possible uses. uses. \Ve We The have previollsly previously not.ee! not,ed that. tha.t maximal maximal common common subgraph subgraph search sea.rch could could be useful in our our starting mat.erial ma,terial select.ioll selection program, program, SST, but but that that the computational computational demands were too too st.arting rigoro1ls: Cone et al. who interpretive rigorous.* who has used MCSS MCSS in their their "self-training “self-training interpretive and retrieval retrieval system" system” (STIRS). (STIRS), has noted noted that that other other potential potential uses uses of of for for MCSS MCSS include include comput.er-assisted computer-assisted organic organic synt.hesis synthesis and structure structure activity activity studies." studies.** Given two two gra.phs. gra.phs, finding finding a common common subgraph involves discovering discovering the assignment assignment Given subgraph involves of of some of of the nodes ane! and edges edges of of one graph graph onto onto the other other graph graph while while preserving preserving the adjacency adjacency relationships relat,ionships of of the nodes. The The size of of the common common subgraph subgraph is the number number of of graph of edges edges preserved preserved in the assignment. assignment. If If there t,here exists no common common sub subgraph of larger larger size, the common common subgraph subgrsph is called called maximal. maximal. Our Our approach approach to MCSS MCSS was to reduce redundant redundant searching searching and try try to to shrink shrink the size of of the t,he search space. We size We observe that that most most libraries libraries of of chemicals chemicals have compounds compounds which have similarities; simila.rities; by by capitalizing capitalizing on these similarities similarities we might might be able to t,o reduce which the t,he search space. The The essentia.l essentia>lprinciple principle is that that if if we know know a relationship relationship between between library library graph graph A A and B we can relate relate query query graph graph C to to A. A, then then we may may therefore therefore already already know know somet.hing about about) the t.he relationship relationship of of C to to B. We We establish establish the relationship relationship between between A A and a.nd B in a one-time-only one-time-only preprocessing preprocessing of of the library library. . •W.T. Inf. Comput. Sci., lC’.T. Wipke. Wipke, D. Rogers, Rogers, I.I. Chern. Chem. Inf. Corn&. Sci., 10S4, 1984, ,(in ,(in press) press) •• M.M MAI Cone. Cone, Venkataraghavan, F.W. Mclafferty, Melafferty, I.I. Am. Chem. Soc., Sot., 00, QQ, 766S, 7688, (1077) (1977) R. Venkataraghavan, F.W. Am. Chern. lSI 151 E. A. Feigenbaum Feigenbaum SEes: SECS: Simulation Simulation & & Evaluation Evaluation of of Chemical Chemical Synthesis Synthesis 5P41 5P41 RR00785-11 RR00785-11 The The objective objective of of our our study study was was to to demonstrate demonstrate the the feasibility feasibility and and study study properties properties of of such such an an algorithm. algorithm. The The following following design design features features were were deemed deemed important. important. •l The The processed processed library library should should be be storage storage efficient. efficient. •l It It should should be be possible possible to to add add compounds compounds incrementally increlnentally to the file, to avoid the cost cost of of reprocessing reprocessing the the file file whenever whenever changed. changed. •l The The time time required required to to store store aa new new compound compound in in the the processed processed library library should be be minimal; minimal: preferably, preferably, an an upper upper limit limit on on this this time time requirement requirement, should should be be known. known. •l It It should be be possible to create the compound compound from from its processed form, form, so so that that both both processed processed and non-processed libraries libraries do not have to be be kept. kept. \Ve We wished the system to be be useful for aa range of non-preselected non-preselected queries, therefore therefore we ruled out out simply simply "training" “training” the system system for aa small small class class of query query choices or depending depending on det.ailed detailed Knowledge knowledge of the allowed allowed queries. We had noticed noticed that that identical identical common common subgraph subgraph candidates candidates are often often generated generated during during the search of different different compounds compounds against ag?inst aa query. query. \Ve We envisioned envisioned collecting collecting these common allow the search to be conducted common segments into into an intermediate intermediate graph, graph, to. to.allow conducted once over over the common common featllre. feature. If If our our search procedure procedure can take take advantage advantage of of the abstraction abstraction of the common common graph graph from from two two or more test compounds, compounds, then then the number number of attempted attempted subgraph suhgraph matches mat)ches will will be be reduced. reduced. A A non-recursive non-recursive FORTRAN FORTRAN algorithm algorithm was implemented implemented for for the MCS-1 MCS-1 program program using aa, tree-st.ructured st.orage file. The tree storage search algorithm reductions tree-struct,ured storage The storage algorithm gives reductions from from 70 to to 90% 90% in the search relative relative to conventional conventional unstructured unstructured sequential sequential storage storage systems. Reductions are especially good for the case where library contains smaller Reductions especially for case where the library contains smaller syst,ems. graphs The of the search tree is gra.phs or or a series of of similar similar graphs. graphs. The general general structure structure of det.ermined tree are established, established, addition addition det.ernlined early en.rly in its creation; creation; once the major major nodes of the the tree of Sorting that the tree tree can of compounds compounds rarely rarely alters alters it it, significantly. significantly. Sorting experiments experiments showed showed that that improved improved results results can be obtained obtained when when searching searching over over the the be "seeded" “seeded” in such a wa.y that seeded library relative relative to to the unseeded library. library. seeded library \Ve have a.pplied applied this this MCS-1 MCS-1 algorithm algorithm to to the Class IV IV starting starting material material We recognition problem. problem. The The init.ial initial abstraction abstraction of of the starting material library library resulted resulted in an recognition starting mat,erial abstracted library library of of significantly significantly reduced reduced size. Organizing Organizing this this abstracted abstracted library library in in a abstract.ed tree-structured form form allows allows the the discovery discovery of of starting starting materials materials which which do not, not have a tree-structured subgraph or or super super graph graph relationship relationship with with the the target. target. A A trial trial run run with with morphine morphine was was subgraph successful in point,ing point.ing out out an interesting interesting starting starting material material candidate candidate not not found found by by the the SST SST progra.m. This This work work is being being submitted submitted to to the the J. Chem. Chem. Inf. Inr. Comput. Comput. Sci.. program. C.2 _YiTiVO TEND Program Program Developments Developments ~7.2 The metabolic metabolic fate fate of of various various compounds compounds in in the the human human body body is extremely extremely complex, complex, The extremely important important for for itit is known known that that through through metabolism metabolism certain certain otherwise otherwise yet. extremely harmless compounds compounds are converted converted into into toxic toxic and possibly possibly carcinogenic carcinogenic agents. agents. Because of of harmless this complexity complexity itit is difficult, difficult, looking looking at at a given given compound, compound, to to forecast forecast potential potential biological biological this activity of of t,hat that given given compound. compound. The The objective objective of of this this proposal proposal is to to develop develop a practical practical activit,y computer program program by by which which a biochemist biochemist or or metabolism metabolism expert expert can can explore explore the the computer metabolites of of a given given compound compound and and be alerted alerted to to the the plausible plausible biological biological activity activity of of each metabolites rnet,abolite. met,a.bolit,e. This research research aims aims to to explore explore the the degree to to which which current current knowledge knowledge of of metabolism metabolism This can be be used used by by a computer computer program program to to make make reasonable rea.sonable projections projections of of what what metabolites metabolites can E. A. A. Feigenbaum Feigenbaum E. 152 SPH SP41 RR00785-11 RR00785-11 SECS: SECS: Simulat.ion Simulation 8l & Evaluation Evaluation of of Chemical Chemical Synthesis Synthesis might might result result. from from exposure exposure of of aa compound compound to to aa biological biological system. system. The The project project involves involves representing in a computer the metabolic processes with all known specificities representing in a computer the metabolic processes with all known specificities and and applying applying these these processes processes to to aa xenobiotic xenobiotic compound compound to to generate generate aa set set of of plausible plausible metabolites metabolites which which may may t.hemselves themselves be be further further metabolized. metabolized. \Ve We also also plan plan an an evaluation evaluation module module to to appraise appraise the the plausible plausible biological biological activities activities of of each each metabolite metabolite using aa rule base base to Thus Thus the attention attention of the to rebte relate chemical chemical structure structure to to biological biological activity. activity. experimentalist experiment,alist will will be be attracted attracted to to those those metabolites metabolites that that are likely likely to be be biologically biologically active. active. C.2.a Atomic Atomic Charges. Chargeu. During During the past year, we implemented implemented rapid rapid atomic atomic charge calculation calculation algorithms algorithms to be be llsed used wit.h wit,h the XENO XENO program program in order order to better better predict predict the biological biological activity activity of metabolite~. metnbolites. The algorithms algorithms were based based on Gasteiger's Gasteiger’s PEOE PEOE (part,ial (partrial equalization equalization of orbital orbital electronegativity)· electronegativity)* and SO-POE SD-POE (sigma (sigma dependent dependent pi orbital orbital electronegativity)·· electronegativity)” mooels. models. The PEOE PEOE model has has been been used used for sigma charge calculations calculations for sigma sigma bonded bonded and non-conjugated non-conjugated pi systems. The The SO-POE SD-POE model has been used for for conjugated conjugated aliphatic aliphat,ic and single ring ring aromatic a.romatic molecules. For For polyaromatic polyaromatic systems, the pi charge cha.rge calculations calculations are being implemented implemented using the SO-POE SD-POE model model and the Longuet-Higgins Longuet-Higgins approximat.ion. In the PEOE PEOE model, the SO-POE SD-POE model, model, and polyaromatic polyaromatic approximation.***••• hydrocarbon pi charge calculation, atoms by hydrocarbon calculation, atoms are characterized characterized by their their orbital orbital e1ectronega,ti vi ties. electronegativities. \Ve We have shown that that the charges so calculated calculated are reasonable when when compared compared to the work Our work of others others in the literature. literature. Our purpose is then then to to correlate correlate the biological biological centers in the the metabolites. metabolites. a,ctivit,y activit,y of metabolites meta.bolites with wit,h the atomic atomic charge on electrophilic electrophilic centers \Ve extent by by atomic atomic charges We also think think that that metabolism metabolism itself itself can be controlled controlled to to some extent so the metabolic metabolic transforms transforms may may make make use use of this this data data eventually. eventually. C.2.b ph’ pKa a Calculations. Calculations. \Ve have continued continued work work on the problem problem of of the the estimation estimation of of the the dissociation dissociation \Ve constant.s for for organic organic acid and bases bases to be used to to increase the expertise expertise of of the the XENO XENO constant,s program. \Ve have been investigating investigating two two approaches approaches to to do this this type type of of estimation: estimation: pr0gra.m. LFER (linear (linear free energy energy relations), relations), and theoretical theoretical or or quantum quantum chemical chemical approaches.**** approaches .. • .. LFER The LFER LFER computation computation is performed performed automatically automatically by by first first selecting selecting appropriate appropriate The skeleton st#ructures structures from from a library, library, then then recognizing recognizing attached attached groups groups and and finally finally skelet.on calculating the the pKa pKa from from the relevant relevant equations equations and group group substituent substituent constants. constants. This This is ca.lculat,ing first automatic automatic pKa pKa estimation estimation algorithm algorithm ever developed developed and and promises promises wide wide utility utility on on the first its own own outside outside of of the XENO XENO program. program. its To determine determine the the most most representative representative pKa pKa ifif more more than than one acid acid or or base center center is To present, several several empirica. empirical rules rules are followed: followed: present, • acids acids -- use the the ionized ionized form form of of the the acid acid center center with with the the lowest lowest pKa pKa as a substituent for for the the pKa pKa calculation calculation of of the the next next lowest lowest acid, acid, and and so on. on . substituent l GasteigE'r and and M. M. Marsili, Marsili, Tetrahedron Tetrahedron Iett., lett., 34, 84,3181, (1981) l•J.J. Gasteiger 3181, (1081) Gasteiger and and hl. 11. Ma&i, Marsili, Tetrahedron, Tetrahedron, 38, 88, 3219, 3210, (1080) (1980) l••*J.J. Gasteiger •••H,H, C.C. Longuet-Higgins, Longuet-Higgins, 1. I. Chem. Chern. Phys., Phy,., f3, 18, 275, 275, (1050) (1950) l *. •••• D.O. Perrin, B. B. Dempsey, Dempsey, and and E.P. E.P. Serjeant, Serjeant, ‘·pKa Prediction for for Organic Organic Acids Acids and and Bases., Bases·, Chapman Chapman and and D.D. Perrin, pKa Prediction Hall, New New York, York, 1081. }081. Hall, 153 E. A. A. Feigenbaum Feigenbaum E. SECS: Simulation Simulation 8[ & Evaluation Evaluation of of Chemical Chemical Synthesis Synthesis 5P41 5P41 RR00785-11 RR00785- 11 •l bases bases - use use the protonated form pKa as a the protonated form of of the the base center center with with the highest highest pKa substituent for the pKa calculation of the next highest' base, and so on. substituent for pKa calculation of the next highest’base, •l strong acids bases - compute pKa's strong acids and and bases compute pKa’ s for for the acids first, first, then then use use ionized ionized form pKa's bases. form of of the acids as substituents subst,ituents to to compute compute pKa’ s of of the the bases. •l strong acids basis - compute pKa's bases first, compute pKa’ s for for bases first, then then use use strong acids and and weak weak basis protonated form bases as pKa's protonsted form of of the bases as substituents substituents to to compute compute pKa’ s of of the acids. •l weak basis - compute pKa's bases first, compute pKa’ s for for weak weak bases first, then then use use weak acids acids and and weak weak basis deprot.onated bases as substituents pKa's deprot,onated form form of of bases substituents to compute compute pKa’ s for for the acids. C.2.g Collaborative Efforts. In past year program Collaborative Efforts. In the past year most most work work was aimed aimed at program development problems, however development rather rather than than application application to to laboratory laboratory problems, however in the next next year year we do expect program modifications perform several expect after aft*er completion completion of of the current current program modifications to to perform practical analyses in conjunction NIH. and other practical conjunction with with our our collaborators collaborators at lCl ICI of of UK, UK, NIH, other parties that parties that have indicated indicated interest. interest. The project continues pharmaceutical The SECS project continues to to have collaborations collaborations with with the pharmaceutical indust,ry joint program program development. indust,ry who who are adding adding chemical chemical transforms transforms and doing doing some joint development, for for example. example, Dr. Dr. Yanaka Yanaka continued continued work work started started at Santa Santa Cruz Cruz after after he returned returned to to Kureha paper has been prepared prepared on that Kureha. Chemical Chemical in Japan Japan and a paper that work. work. D. List List of Project Publications Publications D. of Current Current Project 1. Rapid Subgraph Parallelism Using Parallelism Subgraph Search Search Using 1. Wipke. Wipke, W.T W.T.,.. and Rogers, D.: D.: Rapid J. Chen. Chen. Inr. Inf. Comput. Comput. Sci (submitted (submitted 24 April April 84) 2. Wipke. \-V.T.: "An A-Z: A Wipke, W.T.: “An Integrated Integrated System System for for Drug Drug Design" Design” in Computers Computers A-Z: AluTwfacturer's Hardward and for the Pharmaceutical Pharmaceutical ~l~untrfacturer ‘s Guide Guide to Hardward and Software Software for Industry Aster Industry Aster Publishing Publishing Co., Sprinfield. Sprinfield, Oregon. Oregon, (in (in press) 3. Wipke. synthetic design. Symmetry and and organic organic synthetic design. Wipke, W.T. W.T. and Huber. Huber, M.: M.: Symmetry Accepted Accepted in Tetrahedron. Tetrahedron. 4. Wipke, Prediction of Wipke, W.T., W.T., Ouchi, Ouchi, G.I. G.I. and Chou, Chou, J.T.: J.T.: Computer-Assisted Computer-Assisted Prediction of Afetabolism. IN L. Goldberg (Ed.). STRUCTURE-ACTIVITY Goldberg STRUCTURE-ACTIVITY &fetnbolism. IN (Ed.), CORRELATIONS AS A TOOL IN CORRELATIONS A PREDICTIVE PREDICTIVE TOOL IN TOXICOLOGY. TOXICOLOGY. Hemisphere New York, Hemisphere Publishing Publishing Corp Corp.,.. New York, 1983. 1983. 5. Johnson. A., Johnson, C.K.. C.K., Thiessen. Thiessen, W.E., W.E., Burnett, Burnett, M.N., M.N., Condran. Condran, P. Ronlan, Ronlan, A., Yanaka. \Vipke. W.T.: procedure.'} Ysnaka. M. and Wipke, W.T.: Systematic Systematic derivation derivation of of chemical chemical procedures products, J. of for tra7I·'Jforming .surplus for transforming surplus hazardous hazardous chemicals chemical8 to useful useful products, of Hazardous Materials. Alaterials. (In Hazardous (In press) o. Automated Inference Inference in Planning Organic in Planning Organic Synthesis Synthesis 6. Dolata. Do1a.t.a. D.P.: D.P.: QED: QED: Automated (Ph.D. (Ph.D. dissertation). dissertation). University University of of California, California, Santa Santa Cruz, Cruz, 1984. 1984. 7. Rogers. Artificial Intelligence Intelligence in SST: SST: Starting Starting Rogers, D.: Artificial in Organic Organic Chemistry. Chemistry. Alaterial Selection (Ph. Material Selection Strategies Strategies (Ph. D. dissertation). dissertation). University University of of California. California, Santa Sant,a Cruz. Cruz. 1984. Wipke. Artificial Intelligence Intelligence in SST: in Organic Organic Synthesis. Synthesis. SST: Wipke, W.T W.T.,.. and Rogers. Rogers, D.: Artificial Starting Afaterial Selection An Application Application of Search. Starting Material Selection Strategies. Strategies. An of Superstructure Superstructure Search. J. Chern. Chem. lnf. Inf. Cornput. Comput. Sci.. Sci., 24:0000. 24:0000, 1984. 1984. E. Funding Funding Status Status E. A. A. F'eigenbaurn F‘eigenbaum 154 154 5P41 5Pfl RR00785-11 RR00785-11 SECS: SECS: Simulation Simulation & & Evaluation Evaluation of of Chemical Chemical Synthesis Synthesis 1. 1. Computer-Assisted Computer-Assisted Prediction Prediction of of Xenobiotic Xenobiotic Metabolism Metabolism Principal Principal Investigator: Investigator: \V. W. Todd Todd Wipke, Wipke, Professor, Professor, UCSC UCSC Agency: NIH, Environmental Health Sciences; Agency: NIH, Environmental Health Sciences; No: No: ES02845-02 ES02845-02 4/1/82-3/31/85 $257,801 4/1,‘82-3/31/85 $257,801 TDC TDC 4/1/84-3/31/85 $$ 89,140 4/l/84-3/31/85 89,140 TDC TDC 2. 2. Graphical Graphical Display Display of of Chemical Chemical Inferences Inferences and and Molecular Molecular Relationships Relationships Principal Principal Invest,igator: Investigator: W. Todd Todd Wipke, Wipke, Professor, Professor, UCSC Agency: Evans and a,nd Sutherland Sutherland Corporation Corporation Gift Gift of PS300 PS300 B/W B/W High Performance Performance Graphical Graphical Display Display System Permanent, Permanent, value $95,000 TDC TDC 3. 3. Compnter Computer Synthesis Principal Principal Investigator: Investigator: W. W. Todd Todd Wipke, Wipke, Professor, UCSC Agency: Stauffer St,auffer Chemical Chemical Company Company Permanent, Permanent, $5.000 $6,000 TDC TDC F. I;. Research Research Env£ronment Environment At At the University University of California, California, Santa Cruz, Cruz, we are connected to the SUMEX-AIM SUMEX-AIM resource by a 4800 baud multiplexed leased line. Our by 4800 multiplexed leased Our video terminals terminals consist of a Z-29, ‘Z-29. DM-3025, 1620, We have a PS300 graphic DM-3025, TIi45. TI745, CDI-I030. CDI-1030, DIABLO DIABLO 1620, and an ADM-3A. ADM-3A. graphic display display which which is driven driven by SUMEX. SUMEX. UCSC UCSC has only only a small IBM IBM 370/145, 370/145, a PDP-ll/45, PDP-11/45, 11/70 jobs for student 11/70 and a V1\..'( VAX 11/780, llj780, (the l1's 11’s are restricted restricted to running running small jobs student timetimesharing) sharing) all of which which are unsuitable unsuitable for for our our current current research. The The SECS laboratory laboratory is located locst#ed in 125 125 Thimann Thimann Laboratories, Laboratories, adjacent adjacent to to the synthetic synthetic organic organic laboratories laboratories at at Santa Sa.nta Cruz. Cruz. n. II. INTERACTIONS WITH WITH THE THE SUhdEX-AIM SUMEX-AIM RESOURCE RESOURCE INTERACTIONS A.. Medical AJedicaJ Collaborations Collaborations and and Program Program Dissemination Dissemination via via SUMEX SUAJEX A. available in the GUEST GUEST area of of SUMEX SUMEX for for casual users, and in the SECS SECS is available DE~'fO area for for serious collaborators collaborators who who plan plan to to use use a significant significant amount amount of of time time and DEMO need to to save the synthesis synthesis tree generated. Much Much of of the the access access by by others others has been through through graphic terminal terminal equipment equipment at at Santa Santa Cruz, Cruz, so much much more more convenient convenient for for structure structure the gra.phic input and output. output. Demonstrations Demonstrations and sample synthetic synthetic analyses were generated for for input numerOllS visitors visitors from from t,he the US and abroad. abroad. Demonstrations Demonstrations of of SECS in Sweden were numerous performed by by Dr. Dr. R. E. Carter, Carter, University University of of Lund, Lund, Sweden, at at many many universities universities and performed companies. Professor Wipke vVipke has also used several SUMEX SUMEX programs programs such as CONGEN CONGEN in in his Professor Computers and Information Information Processing in Chemistry. Chemistry. Communication Communication between on Computers course on collaborators is facilitated facilitated by by using SUMEX SUMEX message message drops, especially especially when when time time SECS collaborators differences between between the U.S. and Europe Europe and Australia Australia makes normal normal telephone telephone differences communication difficult. difficult. Testing Testing and and collaboration collaboration on on the the XENO XENO and and FSECS FSECS project project communication with researchers rese8-rchers at at the the NC1 NCI depend on on having having access access through through SUMEX SUMEX and and TYMNET. TYMNET. with Collaboration with with Lund Lund University. University. The The introduction introduction of of SECS to to organic organic chemists chemists Collaboration in Sweden was one of the seeds that led to the establishment of a computer graphics in of the seeds that to the establishment of computer graphics laboratory for for organic organic chemistry chemistry at at the the University University of of Lund, Lund, with with strong strong support support from from aa laboratory government agency, agency, the the National National Swedish Swedish Board Board for for Technical Technical Development Development (STU). (STU). government 155 155 E. A. A. Feigenbaum Feigenbaum E. SEes: SECS: Simulation Simulation & & Evaluation Evaluation of of Chemical Chemical Synthesis Synthesis 5P41 5P41 RR00785-11 RR00785- 11 Interest Interest in the the applications applications of of computers computers and computer computer graphics graphics in organic organic chemistry chemistry has spread very rapidly throughout the country, and chemist.s at all of the major spread very rapidly throughout the country, chemists at all of major Swedish Swedish universities, universit)ies, as as well well as in the pharmaceutical pharmaceutical industry, industry, have taken taken steps to participate participate in the exciting exciting developments developments in this this field. field. Interest Interest in the pedagogical pedagogical value value of of SECS at the graduate graduate level has led to to its its use to to illustrate illustrate the concepts concepts of of retrosynthetic retrosynthetic planning planning and analysis analysis in conjunction conjunction with with a course "Synthetic given “Synthetic Organic Organic Chemistry--Chemistry--given by by Prof. Prof. Paul Paul Helquist Helquist (SUNY, (SUNY, Stony Stony Brook): Brook): Modern The same course was given given at the Royal Royal Institute Institute of of Modern Methods h4ethods and Strategy Strategy”.II . The Technology Technology in Stockholm, Stockholm, and at the the Chemical Chemical Center Center in Lund, Lund, using using SECS as an integral integral part part of of the course. A A Workshop Workshop on Computers Computers in Organic Organic Chemistry Chemistry was sponsored sponsored by by STU STU on May May 17-18,1983, in Gothenburg Gothenburg to to help organic organic chemists chemists in Sweden enter enter this this area of of work. work. Daniel Daniel P. Dolata Dolata from from Prof Prof Wipke's Wipke’s group group in SC was an invited invited speaker. speaker. A A chemist chemist from from Lund, Lund, Dr. Dr. Alvin Alvin Ronlan, Roman, spent spent a sabbatical sabbatical leave with with the Wipke Wipke group. group, and a graduate graduate of of the Wipke Wipke group, group, Dr. Dr. Dolata, Dolata, is spending spending a postdoctoral postdoctoral stay stay in Lund. Lund. In In collaboration collaboration with with the SECS group group at UCSC, UCSC, Dolata Dolata will will inst,all install the SECS program 11/780 for for the use of of the chemists chemists in program and the QED QED program program on a new VAX VAX 11/780 Lund. Lund, and will will continue continue research with with QED. QED. For For example, example, it it would would be of of interest interest to develop rule rule bases bases to to assist the the chemist chemist in structure structure elucidation, elucidation, and structure-activity snructure-activity relationships. re1a.tionships. Another Anot,her area of of collaboration collaboration involves involves compilation compilation of of chemical chemical transforms transforms by by the chemists chemists in Lund. Lund. Some of of the chemists chemists in Lund Lund work work with with natural natural products products (isolation (isolation and synthesis), synthesis), with with a view view toward toward the discovery discovery and characterization characterization of of physiologically physiologically active active substances. For For example example a strongly strongly mutagenic mutagenic compound compound has been isolated isolated from from a Swedish mushroom (Lactarius vellereus), its structure determined, and a total synthesis mushroom (Lact,arius vellereus), its structure determined, total synthesis elaborated. other work, work, a traditional traditional abortifacient abortifacient from from Bangladesh Bangladesh is being being isolated isolated elaborat.ed. In other from from plant plant material, material, and a psychoactive psychoactive substance substance is being being extracted extracted from from the leaves of of a Nigerian plant. A collaboration collaboration with with a university university in Holland Holland is now now developing developing along Nigerian p!ant. A similar similar lines, and Cornell Cornell University University is planning planning a similar similar center center for for computer computer applications applications in chemistry. chemist,ry. B. Examples Examples of with Projects R. o j Cross-fertilization Cross- fertilization with other other SUAfEX-A1Af SUhfEX-,4Ihf Projects The The AILIST AILIST bulletin bulletin board board has been used extensively extensively for for interacting interacting with with many many projects projects and locating locating references references for for further further information information related related to to program program design and AI AI technology. t.echnology. There There are no longer longer any any other other chemical chemical or or biochemical biochemical projects projects on SUMEX SUkIEX so our our interaction int.eraction with with the community community is limited limited to AI AI technology technology interchange, interchange, attending attending seminars at Stanford, etc. seminars Stanford, C. Critique Resource Services Clritique of o j Resource Services SUMEX-AIM SUMES-AIM gives us at UCSC, UCSC, a small sma.11university, university, the advantages advantages of of a larger larger group group of of colleagues, and interaction interaction with with scientists scientist,s all ail over over the the country. country. Previollsly Previously we were provided provided very very good service service by by SUMEX-AIM, SUMEX-AIM, but but since 1 April April 1984, the the computer computer service has been very National AIM very poor. poor. Although Although the the National AIM usage usage of of SUMEX SUMEX has been small, small, our This is a very very our project. project, has been put put in a separate separate class with with a 3% 3% cpu limitation. limit,ation. This severe restriction restriction which which prevents prevents short short usage usage peaks from from being being averaged averaged with with other other users. Our Our project project is the only only project project subjected subjected to to such limitations. limitations. The The poor poor response time time we are observing observing (load (load averages of of 25-50!) is significantly significantly hindering hindering our our ability abilit,y to to perform perform the research NIH NIH funded funded to be done on SUMEX. SUMEX. This This is worsened worsened by by the fact fact we are in ect. th thee Ia.st la,st year year of of the proj project. E. A. Feigenbaum Feigcnbaum 156 156 RR00785-11 5P41 RR00785-11 SECS: Simulation Simulation 8l 62 Evaluation Evaluation of of Chemical Chemical Synthesis Synthesis D. Collaborations Atfedical Use of Programs D. Collaborations and and Medical o j Fkograms via via Computers Computer8 other other than than SUMEX SUMEX SECS 2.9 has been installed installed on the CompuServe CompuServe computer computer networks networks for for the past past three three years years so anyone can access access it it without without having having to convert convert code for for their their machine. machine. This This has proved proved very very useful useful as as a method method of of getting getting people to to experiment experiment with with this this new t.echnology. t,echnology. Dr. Dr. George Purvis Purvis of of Battelle Battelle has accessed accessed SECS via via CompuServe, CompuServe, as as has SECS also resides on the Gene Dougherty Dougherty of of Rohm Rohm and Haas and many many others. others. Medicindat Medicindat machine machine at the University University of of Gothenborg, Gothenborg, Sweden, and is available available all over over Sweden by Similarly in Australia, Australia, SECS resides at the University University of of Western Western by phone. Similarly Australia A Australia and is available available throughout throughout Australia Australia over over CSIRONET. CSIRONET. A lecture lecture series was given given on SECS in Tokyo Tokyo and a.nd SECS has been installed installed at at two two locations locations in Japan. FSECS FSECS has been installed National Laboratory installed on a DEC-IO DEC-10 at at Oak Oak Ridge Ridge National Laboratory and serves for for collaborative collaborat.ive development development of of that that approach approach with with Carroll Carroll Johnson. PRXBLD PRXBLD has been disseminated disseminated to over over 30 sites on various various types types of of computers computers including including DEC-lO, DEC-10, DEC-20, DEC-20, IBf'.I. IBRI, VAX, VAX, PRIt-.m PRIME and Honeywell. Honeywell. III. RESEARCH PLANS (4/84-4/85) ill. RESEARCH PLANS (4/84-4/85) A. Near-Term Near-Term ProJkct Project Goals Plans Our A. Goals and and Plans Our research projects projects will will move off off SUMEX-AIM SUMEX-AIM by by 31 31 March March 1985 1985 to some other other as as yet yet unspecified unspecified computer computer system. system. Therefore Therefore our our research object,iYes objectives on SUMEX-AIM SUMEX-AIM are to complete complete research in progress, consolidate consolidate programs programs and files for for moving moving to to another another system. system. The The QED QED and SST SST projects projects have been completed completed and the first first phase phme of of the RXAN R,YAN project project outlined outlined last year year has been completed. On On the SECS SEC’S project, project, the reaction reaction library library is being extended extended by by Dr. Dr. Iwataki. Iwataki. completed. \Ve \Ve will will continue continue to collaborate collaborate with with coworkers coworkers in SECS research on other other machines machines but but on SUMEX We SLIMES will will primarily primarily be preparing preparing SECS for for removal removal from from SUMEX. SUMEX. We are exploring exploring PROLOG PROLOG as as a replacement replacement for for the QED QED system system and plan plan some preliminary preliminary sample PROLOG PROLOG programs programs to compare compare the the capabilities capabilities of of PROLOG PROLOG and QED. QED. But But the majority majority of of our our activities activities will will be aimed aimed at completion completion of of the XENO XENO program. program. XENO Goals Our dY&\rO God3 Our objectives objectives for for this this year year follow follow our our pla.n plan in the original original proposal. proposal. Basically in this this next next year year we plan plan to complete complete the implementation implementation of of algorithms algorithms not not yet yet Basically completed completed and focus on testing testing with with applications applications to demonstrate demonstrate the current current power power of XENO XENO on typical typical laboratory Iaborat.ory metabolism metabolism problems. problems. In In this this last last year year of of this this project project our our goal is to bring bring XENO XENO to a relatively relatively stable stable finished finished point point which which will will be useful to to other other will complete complete the the algorithms algorithms in progress, document document them them and researchers. \Ve believe we will submit submit, publications publications on all of of the work work within within the year. year. The The major major areas of of focus are listed below. below. listed A.1 Charge Calculations. Calculations. A.1 ,&ornic Atomic Charge We We also plan plan to complete complete our our correlations correlations between between atomic atomic charges and sites of of metabolism, metabolism, as as we have already already done with with bond bond reactivity. react,ivity. \Vhen CZ’hen such correlations correla.tions are established, established, then then they they can be used to to guide XENO XENO to to apply apply metabolic metabolic transforms transforms more more selectively selectively to to the most most active active parts parts of the molecules. A.2 pKa pJ(a Calculations. A.2 Calculatione. We We plan plan to complete complete testing testing of of the pKa pKa algorithm algorithm on different different. groups groups on metabolites metsbolites so that that information information may may be used by by XENO XENO for for activity activity evaluation, evaluation, select,ion selection of of further further possible metabolism, metabolism, and estimation estimat#ion of of excretion excretion and t.ra.nsport. transport. A.S Three-Dimensional Criteria. We -4.3 Three-Dimensional Criteria. We plan plan to to complete complete our our work work on threethreedimensional dimensional constraints constraints that tha.t apply apply to metabolism metabolism which which have been obtained obtained through through study This study of many many metabolic metabolic studies studies in the literature. literature. This will will require require extensions extensions of of the 157 Feigenbaum E. A. Feigenbaum SECS: Simulation Simulation 86 & Evaluation Evaluation of of Chemical Chemical Synthesis Synthesis 5P41 RR00785-11 RR00785-11 ALCHEM .4LCHEM language t,o to accommodate accommodate new types types of of three-dimensional three-dimensional relationships relationships such as as overall overall molecular molecular size, width, width, thickness, thickness, ratio ratio of of length length to to width, width, etc etc.. .1.4 -4.4 Log P Calculation. Calculation. In In order order to to more more accurately accurately estimate estimate the possibility possibility of of excretion excret,ion and a.nd binding binding for for met.abolism, met,abolism, we plan plan to to incorporate incorporate a log P calculation calculation module. module. This This will will provide provide the partition partition coefficient coefficient between between octanol octanol and water. water. There There are already already programs Our programs to to calculate calculate log P, and these have been shown shown to to be very very accurate. accurate. our objective, objective, time time permitting, permitting, will will be to to include include such a module module in the XENO XENO program program and correlate correlate log P with with metabolism metabolism transform transform application application and with with excretion. excretion. B. Justification Justification and Requirements for for Continued B. and Requirements Continued Use of of SUAIEX SUMEX The The XENO XENO project project which which resides on SUMEX SUMEX is in its its last year year of of support, support, consequently consequently we need to complete complete that that research on the SUMEX SUhlEX machine. machine. By By 31 March March 1985, plan to move XENO 1985, we we plan XENO and all our our research off off SUMEX SUMEX onto onto some other other computer. computer. \Ve are currently currently exploring exploring what what machine machine may may be suitable suitable and available. available. After After 1 April April 198,5 1085 we will will not not need SUMEX SUMEX for for computational computational support, support, but but will will need access a.ccessto be able to respond to to electronic electronic mail, mail, and continue continue to to t,o retrieve retrieve certain cert.ain files from from archive, archive, respond participate participate in the AIM AIM scientific scient,ific interchange interchange through through electronic electronic mail mail and bulletin bulletin boards. boards. It. practical to It, is not not, pract,ical to retrieve retrieve every every file file we have ever archived, archived, it it would would use use too too much much SUMEX SUMEX operator operator time, time, and it it is unnecessary unnecessary as as long as as we can access access them them if if we need them them in the future. future. That That access access would would not not require require significant significant resources. prior to 31 March However 1985 we have obligations obligations to to complete complete the research on However prior March 1985 t,he NIH and need sufficient the XENO SENO project project supported supported by by NIH sufficient SUMEX SUMEX cpu CPU time time to to accomplish accomplish t.his this goal. This This means normal normal editing, editing, compile, compile, load, and test test executions executions plus plus some runs to some metabolic problems. It appears the current removal applicat.ion application runs to metabolic problems. It of our our current removal of project from National AIM project, from the t,he National AIM portion portion of of the the SUMEX-AIM SUMEX-AIM resource and placement pla.cement in a of class restricted restricted to 3% 3% peak utilization utilization is hindering hindering the research productivity productivity of this this project.. \Ve are experiencing project,. experiencing load averages of of 25-50 a high high percentage percentage of of the the time. time. We We request National AIM request to to have om our project project placed back back in the National AIM portion portion of of the SUMEX-AIM SUMEX-AIM resource as as we were allocated, allocated, and we will will carefully carefully monitor monitor to to see see that that our our resource utilization utQlizat.ion does does not not exceed our our quota quota of of time. time. We We feel this this is a reasonable request request in light light of National community of the mission mission of of SUMEX-AIM SUMEX-AIM to the National community of of which which this this project project is a part. part. C. Needs Needs Beyond Beyond SUA-lEX-AIM SUMEX--,;1IM As mentioned mentioned above our our project project needs additional additional computing computing resources and we are exploring exploring acquiring acquiring a computer computer for for installation installation at at UCSC UCSC and obt.aining obt.aining the necessary resources to to support support it. it. \Ve We are seeking information information about about comparisons comparisons between between machines machines and cost effectiveness effectiveness of of different different hardware hardware combinations. combinations. D. Recommendations for for Community and Resource Development Development L?. Reconmendations Community and Resource It It, appears the the’ SUMEX-AIM SUMEX-AIM resource is increasingly increasingly becoming becoming basically basically a Stanford Stanford reSOllrce resource and a.nd that t,hst. there there is a difference difference between between the the portion portion of of the resource allocated allocated to the National community National community. Our Nationa. communit,y and the portion portion actually actually used by by the National community. Our project National community project, is part pa.rt of of the National community and in need of of better better service, service, we hope that that can be improved. improved. An An important important part part of of medicine medicine is treatment treatment of of disea:-es diseases with with drugs, drugs, chemicals, chemicals, chemicals chemicals that t,hat were designed and synthesized synthesized by by chemists. chemists. Since the termination termination of of the DENDRAL project, there seems to be declining declining support support for for artificial artificial intelligence intelligence DENDRAL project. there seems applications We applications in chemistry. chemistry. We feel that that support support of of this this area is essential to the advancement in this The NIH Research advancement of of medicine medicine this country. country. The lack lack of of chemists chemists on NIH Resources computing computing peer review review 'is .is contributing contributing to the problem. problem. In general general the AIM AIM communit.y community would would benefit benefit by by involving involving disciplines disciplines other other than than computer computer science. E. A. Feigenbaum Feigenbaum 158 SOLVER SOLVER Project Project 5P41 5P41 RR00785-11 II.A.2.5. SOLVER Project II.A.2.5. SOLVER Project SOLVER: Problem Solving Expertise SOLVER: Problem Solving Expertise Dr. Dr. P. P. E. E. Johnson Johnson Center for in Learning Center for Research Research in Human Human Learning University of University of Minnesota Minnesota Dr. Dr. W. W. B. B. Thompson Thompson Department of Science Department of Computer Computer Science University of University of Minnesota Minnesota I. I. SUMMARY OF PROGRAM SUMMARY OF RESEARCH RESEARCH PROGRAM A. A. Project Project Rationale Rationale This This project project focuses focuses upon the development development of strategies strategies for for discovering discovering and documenting documenting the t,he knowledge knowledge and skill skill of of expert expert problem problem solvers. In the last last fifteen fifteen years, considerable considerable progress has been made in synthesizing synthesizing the expertise expertise required required for for solving solving extremely Computer programs programs exist exist with with competency competency comparable comparable to ext’remely complex complex problems. problems. Computer human human experts exprrt,s in diverse areas ranging ranging from from the analysis analysis of mass spectrograms spectrograms and nuclear nuc!ear magnetic rnagnet,ic resonance (Dendral) (Dendral) to the diagnosis of of certain certain infectious infectious diseases diseases (Mycin). (Mycin). Design of particular task of an expert expert system system for for a particular task domain domain usually usually involves involves the interaction of two distinct groups of individuals, "knowledge engineers," interaction of two distinct groups of individuals, “knowledge engineers,” who who are primarily concerned problem solving primarily concerned with with the specification specification and implementation implementation of of formal formal problem solving t,echniques, problem area) who provide factual (in the relevant relevant problem who provide factual and techniques, and "experts" “experts” (in heuristic problem solving hcurist,ic information information of of use use for for the problem solving task t,ask under under consideration. consideration. Typically Typically the knowledge particular knowledge engineer engineer consults consults with with one or or more more experts experts and decides on a particular represent3tional Next, "units" “units” of of factual factual information information representational structure structure and inference inference strategy. strategy. Next, are specified. properties of problem domain specified. That That is, properties of the problem domain are decomposed into into a set of of manageable processing by by the inference Once this this msnagesble elements elements suitable suitable for for processing inference operations. operations. organization ha.s has been established, established, major major efforts efforts are required required to to refine refine representations representations and organization acquire fact.ual factual knowledge knowledge organized organized in an appropriate appropriate form. form. Substantial research research a,cquire Substantial prohlems exist exist in developing developing more more effective effective representations, representations, improving improving the the inference inference problems process, and in finding finding better better means of of acquiring acquiring information information from from either either experts experts or or the the problem area itself. itself. problem Programs currently currently exist exist for for empirical empirical investigation investigation of of some of of these questions questions for for a Programs particular problem problem domain domain (e.g. AGE, AGE, UNITS, UNITS, RLL). RLL). These tools tools allow allow the investigation investigation particular of a.lt,ernnt,e alternate organizations, organizations, inference inference strategies, strategies, and rule rule bases bases in an efficient efficient manner. manner. of \Vhat is still still lacking, lacking, however, however, is a theoretical theoretical framework framework capable of of reducing reducing dependence \Vhst the expert’ expert's intuition or or on near near exhaustive exhaustive testing testing of of possible organizations. organizations. Despite Despite on the s intuition their successes, successes, there there seems to to be a consensus that that expert expert systems systems could could be better better than than their they are. Most expert expert systems systems embody embody only only the the limited limited amount amount of of expertise expertise that that Most they individuals are able to to report report in in a particular, particular, constrained constrained language (e.g. production production rules). rules). individuals current systems systems are approximately approximately as good good as human human experts, experts, given given that that they they represent represent IfIf current only a port,ion of what individual human experts know, then improvement in the the only portsion of what individual human experts know, then improvement in "knowledge capturing” capturing" process should should lead to to systems systems with with considerably considerably better better “knowledge performance. performance. 159 A. Feigenbaum Feigenbaum E. A. RR00785-11 5P41 RR00785-11 SOLVER SOLVER Project Project and Collaboration Collaboration B. Medical AIedical Relevance Relevance and Collaboration Collaboration with with Dr. Dr. James Moller Moller MD MD in the Department Department of of Pediatrics, Pediatrics, Dr. Dr. Donald Donald Connelly Connelly MD MD in the Department Department of of Laboratory Laboratory Medicine, Medicine, at the University University of of r-.finnesota. hfinnesota. Collaboration Collaborat,ion with with Dr. Dr. Eugene Rich Rich MD MD and Dr. Dr. Terry Terry Crowson Crowson MD MD at St. Paul Paul Ramsey Ramsey Medical Medical Center. Center. of Research C. Highlights Highlights of Research Progress Progress Prior research at Minnesota Minnesota on expertise expertise in Accomplishments Accomplishments of of This This Past Past Year Year -- Prior diagnosis diagnosis of of congenital congenital heart heart disease disease has resulted resulted in a theory theory of of diagnosis diagnosis and an embodiment embodiment of of that that theory theory in the form form of of a computer computer simulation simulation model, model, Galen, Galen, which which diagnoses cases cases of of congenital congenital heart heart disease disease [Thompson, [Thompson, Johnson & Moen, Moen, 1983]. 19831. Galen Galen is descended from from two two earlier earlier programs programs written written here at at Minnesota: Minnesota: and Deducer [Swanson, a.nd Dedrrcer (Swanson, 1977]. 19771. Deducer Deducer is a program program that that builds builds hemodynamic hemodynamic models of of the circulatory circulatory system system that that describe specific specific diseases. diseases. The The models are built built, by by using usirig knowledge knowledge about about how how idealized idealized parts parts of of the the circulatory circulatory system system are causally causally related. related. Diagnoser program Dia.gnoser is a recognition-driven recognition-driven program that that performs performs diagnoses by by successively successively hypothesizing hypothesizing one or or more more of of these models and matching matching them them against against patient patient data. data. The The models that tha,t match match best are used as the final final diagnosis. diagnosis. A A series of of experiments experiments carried carried out at Minnesota Minnesota have shown shown that that Diagnoser/Deducer Diagnoser/Deducer performs performs as well well (and (and sometimes sometimes out better) better) than than expert expert human human cardiologists cardiologists [Johnson [Johnson et aI., al., 1981]. 19811. Dial]lloser Diagnouer Despite their their early early successes, successes, Diagnoser Diagnoser and Deducer Deducer did did not not have a clear, clear, Despite comprehensible comprehensible structure structure that that is required required for for the the kind kind of of experiments experiments we wish wish to t,o perform. perform. Galen Galen was built built to remedy remedy this this problem, problem, taking taking advantage advantage of of the experience experience gained gained in the of Diagnoser Disgnoser and Deducer. Deducer. design of Galen Galen consists of of four four major major components: components: a working working memory memory called called the scratchpad, scratchpad, a knowledge proposer and a knowledge base base of of rules rules and hypotheses, hypotheses, a procedure procedure called called the proposer procedure called called the reviewer. reviewer. procedure The The scratchpad scratchpad contains contains data data about about the problem problem that that Galen Galen is trying trying to to solve and the hypotheses hypotheses that that are being being investigated investigated to to explain explain that that data. data. In In effect, effect, the scratchpad scratchpad represents represents Galen's Galen’s current current execution execution state. state. Rules are pattern-action pattern-action pairs. pairs. The The pattern pattern part part of of a rule rule describes a possible state state of the t,he scratchpad. scratchpad. Patterns Patterns can contain contain imbedded imbedded logical logical connectives connectives (e.g. ANDs, ANDs, ORs, of NOTs) and can be constructed NOTsj constructed to match match at at varying varying levels of of detail. detail. The The action action part part is a procedure procedure that that is executed executed if if the pattern pattern part part matches matches the scratchpad's scratchpad’s contents. contents. Each Each action nct,ion part part writes writ,es an assertion assertion on the the scratchpad scratchpad about about a hypothesis, hypothesis, together together with with the evidence for for making making that that assertion. assertion. These assertions assertions can express that that a new hypothesis hypothesis is being considered, considered, or or that that an old old hypothesis hypothesis has been accepted, rejected, rejected, confirmed confirmed or or disconfirmed. Action disconfirmed. Action parts parts can also assert that that a hypothesis hypothesis is sufficient sufficient to to solve the the current. problem, problem, or or that that the problem problem is not not solvable. solvable. current. Because the pattern pattern parts parts of 01 rules rules can examine examine anything anything on the scratchpad, scratchpad, it it is possible to express rules rules about about hypotheses hypotheses as well well as as rules rules about about problem problem data. data. In pa.rticular, this this makes it it possible to to directly directly examine examine the accumulated accumulated evidence for for and particular, against against each currently currently contending contending hypothesis, hypothesis, making making numerical numerical measures of of certainty certainty unnecessary. unnecessary. A hypothesis hypothesis is simply simply a named named collection collection of of rules. The The hypotheses hypotheses in Galen's Galen’s A knowledge knowledge base base can be thought thought of of as a directed directed graph, graph, in which which vertices vertices are hypotheses hypotheses and snd edges edges are rules. One hypothesis hypothesis IIpoints “points toll to” another another if if the first first hypothesis hypothesis contains contains a rule rule whose action action part part can assert something something about about the second. E. A. Feigenbaum Feigenbaum 160 SOLVER SOLVER Project Project 5P41 5P41 RR00785-11 The The level of detail detail of such aa knowledge knowledge base base leads leads to serious problems problems with with the computational focuses its computational computational resources compllt,st.iona! complexit.y comp1exit.y of search processes. processes. Galen focuses so so that that the knowledge knowledge embodied embodied in the graph graph of hypotheses can be be used used in an efficient efficient manner. manner. Successful diagnoses result result from from good first first hypotheses about about possible defects and efficient efficient mechanisms mechanisms for refining refining these hypotheses. Galen Gslen works works by using the proposer proposer and the reviewer reviewer to investigate investigate hypotheses hypotheses (Le. (i.e. search the graph) graph) by applying applying rules (Le. (i.e. following following the edges edges from from one vertex vertex to another). another). vVhenever Whenever a new piece of problem problem data data is is written written on the scratchpad, scratchpad, the proposer proposer applies all a!! relevant relevant rules specific specific to the type type of that that piece of data. data. These rules write write assertions assertions on the scratchpad scra.tchpa,d about about new hypotheses, effectively effectively identifying identifying vertices vertices in the graph graph that that are worthy Next, the reviewer reviewer applies all relevant relevant rules worthy starting starting points point,s for further further search. Next, contained contained in the hypotheses hypotheses that that are named in assertions on the scratchpad. scratchpad. Successfully Successfully applying applying one of these rules corresponds corresponds to propagating propagating the search along a specific specific edge edge of the graph. graph. The The search is is constrained constrained because because (1) only only the most promising promising vertices vertices in the graph graph are ever used used to initiate initiate search; (2) only only a small small number number of of edges edges are ever followed; followed; a.nd and (:3) (3) most most rules in a hypothesis hypothesis deal with with evidence for for and against against the hypothesiS hypothesis itself, itself, giving giving a graph gra.ph where where the number number of of effectively effectively outward-pointing outward-pointing edges edges at each vertex vertex is small. small. The The read-propose-review rea,d-propose-review cycle repeats repeats in this this way way until until some hypothesis hypothesis has been shown shown correct, correct, until unt,il the t,he problem problem has been shown shown unsolvable, unsolvable, or or until until all the data data has been examined. examined. Currently, Currently, data data given given to Galen Galen is taken taken from from a (possibly (possibly imaginary) imaginary) patient's patient’s medical Hypotheses medical chart. chart. Hypotheses in the knowledge knowledge base base represent represent the ten most most commonly commonly occurring physiological occurring congenital congenital heart heart diseases diseases and their their variants, variants, useful intermediate intermediate physiological findings, findings, and classes classes of of hypotheses. hypotheses. Since hypotheses hypotheses are implemented implemented as named named teams teams of of production rules, it product,ion it is also possible to to represent represent other other kinds kinds of of hypotheses hypotheses should should the need arise. Moreover, arise. IMoreover, Ga.len Ga,!en has been constructed constructed so that that its its inference inference engine does not not contain contain any procedures procedures specialized pediatric cardiology. specialized for for pediatric cardiology. It It is therefore therefore conceivable conceivable to extend extend Galen Gnlen to other ot,her domains domains if if effective effective knowledge knowledge bases bases for for those domains domains can be constructed. constructed. To To determine determine the generality generality of of our our model model of of expertise expertise in diagnostic diagnostic reasoning, reasoning, we are part of a.re also investigating investigating domains domains outside outside medicine. medicine. As part of this t,his effort, effort, we have developed developed a computational program debugging computational model of of the fault fault localization localization process in program debugging [Sedlmeyer, [Sedlmeyer, 198:3] with our our work work in congenital congenital heart heart disease, disease, 19831 that that is not not, based directly directly on Galen. Galen. As with concentrated on the design of of mechanisms mechanisms for for structuring structuring problem problem specific specific we have concentrated knowlt'dge a,nd and for for focusing focusing limited limited computational computational resources. knowledge Progress -- Since human human experts experts are notoriously notoriously poor poor at at describing describing Resra.rch in Progress Rrsca.rch their own own knowledge, knowledge, our our work work requires requires the the creation creation of of problem problem solving solving tasks tasks through through t,heir which expert.s experts can reveal reveal crit.eria criteria for for initiating initiating specific specific hypotheses hypotheses and methods methods for for which those hypot.hcses. hypotheses. invest iga ting t,hose investigating Current techniques t.echniques of of representing representing hypotheses hypotheses and their their expectations expectations for for diagnosis diagnosis Current do not, not, however, however, provide provide much much detailed detailed information information about about the the control control processes expert,s experts do use to to guide guide their t.heir reasoning. reasoning. Such control control processes typically typically incorporate incorporate highly highly refined refined use heuristics about about. which which the the experts experts are almost almost wholly wholly unaware. unaware. To To discover discover the the needed heuristics cont.rol knowledge, knowledge, we ask experts experts to to complete complete tasks tasks in which which we have systematically systematically cont,rol perturbt~d aspects of of the the problem problem data. data. The The data data in in these tasks tasks are chosen so that that perturbed members of an overlapping set of hypotheses will be suggested during while solving the members of overlapping of hypotheses will during while solving the problem. Success in solving such problems depends on the ability to overcome an initially problem. Success in solving problems on the ability to overcome initially plallsihle incorrect incorrect hypot,hesis hypothesis in favor favor of of a later, later, more more correct correct alternative. alternative. plausible 161 161 A. Feigenbaum Feigenbaum E. A. 5P41 RR00785-11 RR00785-11 SOLVER SOLVER Project Project Several Several examples examples illustrate illustrate our our approach. approach. \Ve We are studying studying performance performance of of Galen Galen cases [Johnson [Johnson & & Thompson, Thompson, 1981] 19811 that that were initially initially misdiagnosed misdiagnosed in on "garden “garden path" path” cases hospit.al hospit,al files. Analysis Analysis of of such cases cases suggest that that errors errors are made because because experts experts rely rely on very efficient efficient heuristics heuristics that that are not not universally universally correct. correct. In In one such example, example, a seemingly seemingly very plausible plausible hypothesis hypothesis is suggested early early in the case. case. Although Although the hypothesis hypothesis superficially superficially seems seems to t,o explain explain what what is observed observed about about the patient, patient, the hypothesis hypothesis is incorrect. incorrect. Because the incorrect incorrect hypothesis hypothesis seems seems marginally marginally adequate, it it acts to prevent prevent a more correct correct Success in such a case case hinges on the ability ability hypothesis hypothesis from from being being suggested in in its its place. place. Success to use use the proper proper set of of competing competing hypotheses hypotheses in order order to provide provide more than than one to Investigation of of this this phenomenon phenomenon in human human experts experts has explanation explanation of of the case case data. data. Investigation linking disease disease hypotheses hypotheses in Galen. Galen. It It has suggest.ed suggested implementation implementation of of "transition “transition rules" rules“ linking that watch watch for for potential potential garden garden also suggested implementation implementation of of "monitor “monitor hypotheses" hypotheses” that pat.h path errors errors and avoid avoid them them before before they they become serious. We are also investigating investigating several research questions questions relevant relevant to to t.he the architecture architecture of of We Galen. Galen. \Ve We have designed an interface interface to to Galen Galen so that that users who who are unfamiliar unfamiliar with with the inner workings workings of of the program program can interactively interactively enter enter case case data. data. Designing Designing the interface interface inner questions about about what what forms forms of of data data are necessary to adequately adequately and completely completely raised questions We are also studying studying ways in which which a causal reasoning reasoning represent represent all possible cases. cases. We component. Proposer component, can be integrated integrated with with the prototypical prototypical reasoning reasoning components components (the (the Proposer and Reviewer) Reviewer) that that are already already present present in Galen. Galen. In In particular, particular, we are interested interested in st.udying studying ways in which which causal reasoning reasoning can aid or or replace prototypic prototypic reasoning reasoning when when it it becomes inadequate inadequate to reach a diagnosis. diagnosis. become~ In another another project, project, we are investigating investigating methods methods of of probabilistic probabilistic reasoning. reasoning. Most Most sy:::tems systems rely rely on numerical numerical schemes for for weighting weighting evidence or or ranking ranking observed observed data. data. weights are often often probabilistic probabilistic in nature, nature, but but other other schemes have also been used. These weights Myrin. Mycin, for for example. example, uses uses certainty certainty factors factors and PIP PIP uses uses likelihoods likelihoods composed of of matching matching scores and binding binding scores. In In contrast, contrast, humans humans do not not seem to rely rely upon upon such numerical numerical techniques. t,echniques. Research Resea.rch has shown shown that that people are often often quite quite poor poor at probabilistic probabilistic reasoning. However, reasoning. However, experts experts make make decisions which which involve involve weighting weighting evidence and selecting selecting from from competing compet.ing alternatives. alternatives. They They must must utilize utilize a reasoning reasoning process which which serves as as an alternative alternative to a numerical numerical weighting weighting technique. technique. We believe the process of of weighting weighting alternatives alternatives along various various criterial criteria] dimensions dimensions \Ve is not not a domain domain specific specific technique, technique, but but rather rather a general process which which is applied applied in specific specific t,he coming coming year, year, we wiJI will examine examine this t,his process in various various domains domains and instances. In the instsnccs. a.ttempt a.tt,empt to ut.ilize the result·s result,s in designing designing more more powerful powerful reasoning reasoning techniques. techniques. In In the area a.rea of of law, law, our our work work has focused focused on the area of of corporate corporate law law (the (the problem problem of a proposed We of structuring structuring proposed corporate corporate acquisition). acquisition). We have collected collected data data from from ~4 24 pra.ct.icing law.\"ers lawyers and in the coming coming year year a PhD PhD thesis will will be completed completed describing describing this this practicing work. work. In the coming coming year year we will will be also completing completing a study study of of the corporate corporate acquisition acquisition problem problem in man~1!;ement man:tgement in order order to to further further refine refine our our knowledge knowledge capturing capturing tools. tools. D. List Li.'3t of Relevant Publications Publications D. of Relevant 1. Connelly, Connelly, D. and Johnson, P.E.: P.E.: Medical solving. Human Human Pathology, Pathology, 1. fl.;fedical problem problem solving. 11(5):41?-419, 11(5):41~-419. 1980. 1980. 2. Elstein, Proposed Reaeatch Research Elstein, A., A., Gorry, Gorry, A., A., Johnson, P. and Kassirer, Kassirer, J.: Proposed Efforts. IN Efforts. IN D.C. D.C. Connelly, Connelly, E. Benson and D. Burke Burke (Eds.), (Eds.), CLINICAL CLINICAL DECISION MAKING MAKING AND AND LABORATORY LABORATORY University of of Minnesota Minnesota DECISION USE. University Press. Press, 1982, 1982, pp. 327-334. E. A. Feigenbaum Feigcnbaum 162 152 SOLVER SOLVER Project Project 5P41 RR00785-11 RR00785-11 P.J.: Knowledge based components P.J.: Knowledge components diagnosis. Learning diagnosis. Learning Research and Development Development PDS-:!, PDS-2, University University of of Pittsburgh, Pittsburgh, September, September, 1981. 1981. 3. Feltovich, Feltovich, of in of expertise expertise in Center Technical Center Technical medical medical Report Report Role and 4. Feltovich, Feltovich, P.,I., P.J., Johnson, Johnson, P.E., P.E., Moller, Moller, J.H. J.H. and Swanson, D.B.: D.B.: The Role and Development of J\Jedical Knowledge Knowledge in Diagnostic Ezpertise. Expertise. IN IN Development of &fedical in Diagnostic W. W’. Clancey Clancey and E.H. E.H. Shortliffe Shortliffe (Eds.), (Eds.), READINGS READINGS IN IN MEDICAL MEDICAL AI. AI. (In (In press) 5. Johnson, Models of Medical Problem Problem Solvers. Cognitive Models of Medical Solvers. IN IN D.C. D.C. Johnson, P.E.: P.E.: Cognitive Connelly, Connelly, E. Benson, D. Burke Burke (Eds.), (Eds.), CLINICAL CLINICAL DECISION DECISION MAKING MAKING AND AND LABORATORY USE. University LABORATORY University of of Minnesota Minnesota Press, 1982, 1982, pp. 39-51. 6. Johnson, Johnson, P.E.: P.E.: }Vhat kind should a system system be? J. Medicine What kind of of expert ezpert should Medicine and Philosophy, Philosophy, 8:77-97, 8:i7-97, 1983. Expert Mind: Mind: A new for the Information Information 7. Johnson, P.E., P.E., The Expert new Challenge Challenge for SciWtl:st IN SYSTElvl Scientist IN Th. Th. M. M. A. A. Bemelmans Bemelmans (Ed.), (Ed.), INFORMATION INFORMATION SYSTEM DEVELOPMENT FOR EFFECTIVENESS, Elsevier DEVELOPMENT FOR ORGANIZATIONAL ORGANIZATIONAL EFFECTIVENESS, Elsevier Science Publishers Publishers B. V. V. (North-Holland), (North-Holland), 1984. 1984. Design of 8. Johnson, D.G. and Feltovich, Feltovich, P.J.: P.J.: Design of decision decision Johnson, P.E., P.E., Severance, D.G. support agatema systems in Rationale and principles from from the analysis support in medicine: medicine: Rationale and principles analysis of of ph!Jsician expertise. physician expertise. Proc. Proc. Twelfth Twelfth Hawaii Hawaii International International Conference Conference on System System Science, 'Western Western Periodicals Periodicals Co. Co. 3:105-118, 3:105-118, 1979. 1979. 9. Johnson, Johnson, P.E., P.E., Duran, Duran, A., A., Hassebrock, Hassebrock, F., F., Moller, Moller, J., Prietula, Prietula, M., M., Feltovich, Feltovich, P. and Swanson, D.: Expertise Expertise and in diagnostic diagnostic reasoning. reasoning. Cognitive Cognitive and error error in Science 5::!35-283, 5:235-283, 1981. 1981. 10. Johnson. Jvfodels Johnson, P.E. P.E. and Hassebrock, Hassebrock, F.: F.: Validating Validating Computer Computer Simulation Simulation Models of Expert Reasoning. IN R. Trappl (Ed.), CYBERNETICS AND SYSTEMS of Expert Reasoning. IN Trappl (Ed.), CYBERNETICS AND SYSTEMS RESEARCH. North-Holland Publishing RESEARCH. North-Holland Publishing Co., 19.82. 1982. and Thompson, down garden path: path: Thompson, \V.B.: W.B.: Strolling Strolling down the garden and from error problem solving. solving. Proc. and recovery recovery from error in in expert expert problem Proc. Seventh Seventh IJCAI, IJCAI, Vancouver, Vancouver, B.C., B.C., August, August, 1981, 1981, pp. 214-217. 11 11.. .lohnson, Johnson, P.E. P.E. Detection Detecfion P.E., Jvfultimethod study study of Moller. J.H.: J.H.: bfultimethod of P.E., Hassebrock, Hassebrock, F. and Moller, clinical judgement. Organizational Behavior clinical judgement. Organizational Behavior and Human Human Performance Performa.nce 30:201-230, 1982. 12. 12. Johnson, Johnson, 13. l\1011er, New techniques hfo!ler, J.H J.H.,.. Bass, G.M., G.M., Jr. and Johnson, P.E.: P.E.: New techniques in in the of patient management problems. Medical Education ron.'ltrurtion construction of patient management problems. Medical Education 15:1.')0-153, 1981. 15:l.W153, 1981. 14. simulation of problem solving solving in 14. Swanson, D.B.: D.B.: Computer Computer simulation of expert expert problem in medical medical diagnosis. diclgnosis. Unpublished Unpublished Ph.D. Ph.D. dissertation, dissertation, University University of of Minnesota, Minnesota, 1978. 1978. Psychological Analysis Analysis of 15. Swanson, D.B., D.B., Feltovich, Feltovich, P.J. P.J. and Johnson, P.E.: P.E.: Psychological of Physician Expertise: Implications for The Design of Decision Support Physician Ecpertiae: Implications for Design of Decision Support Systems. In D.B. North-Holland Systems. D.B. Shires and H. \Vold \Vold (Eds.), (Eds.), MEDINF077, MEDINF077, North-Holland Publishing Publishing Co., Amsterdam, Amsterdam, 1977, 1977, pp. 161-164. 161-164. 15. Recognition-based diagnostic 16. Thompson, Thompson, W.B., W.B., .Johnson, Johnson, P.E. P.E. and Moen, Moen, J.B.: Recognition-based diagnostic reasoning. reasoning. Proc. Proc. Eighth Eight,h IJCAI, IJCAI, Karlsruhe, Karlsruhe, West West Germany, Germany, August, August, 1983. 1983. 163 163 E. A. A. Feigenbaum Feigenbsum 5P41 RR00785-11 RR00785-11 SOLVER SOLVER Project Project 17. 17. Sedlmeyer, Sedlmeyer, R.L., R.L., Thompson, Thompson, W.B. W.B. and Johnson, P.E.: P.E.: localization in debugging. The localization in debugging. The Journal Journal of of Systems Systems and Knowledge-based fault fault Knowledge-based Software Software (in (in press). 18. Diagnostic reasoning 18. Sedlmeyer, Sedlmeyer, R.L., R.L., Thompson, Thompson, W.B. W.B. and Johnson, P.E.: P.E.: Diagnostic reasoning in software fault fault localization. Proc. in software localization. Proc. Eighth Eighth IJCAI, IJCAI, Karlsruhe, Karlsruhe, West West Germany, Germany, August, August, 1983. 19. Sedlmeyer, Knowledge-Based Fault Fault Sedlmeyer, R.L., R.L., Thompson, Thompson, W.B., W.B., 8l & Johnson, Johnson, P.E.: P.E.: Kno,wledge-Baaed Debugging. The Localization in Localization in Debugging. The Journal Journal of of Systems Systems and Software Software (in (in press). E. Funding Funding and E. and Support Support \Vork \Vork on the SOLVER SOLVER project project is currently currently supported supported by by a grant grant from from the Control Control Data to Dat,a. Corporation Corporation to Paul Paul Johnson Johnson ($90,000; ($90,000; 1983-85) and by by a grant grant from from the Microelectronics Microelectronics and Information Information Sciences Center Center at the University University of of Minnesota Minnesota to to Paul Paul Johnson, William William Thompson Thompson and two two colleagues ($800,000; ($800,000; 1984-87). II. II. INTERACTIONS vVITH THE RESOURCE INTERACTIONS WITH THE SUMEX-AIM: SUMEX-AJM RESOURCE A. Medical Afedical Collaborations Program Dissemination Dissemination via A. Collaborations and and fiogram via SUAIEX SUMEX \Vork \Vork in medical medical diagnosis dia.gnosis is carried carried out out with with the the cooperation cooperation of of faculty faculty and students Paul Ramsey Ramsey Medical Medical students in tht' the University University of of Minnesota Minnesota Medical Medical School and St. Paul Center. Cent,er. B. Sharing Interactions with ProJects B. Sharing and and Interactions with Other Other SUA-lEX-AIM SUfifEX-AIM Projects A A year year ago, conversations conversations were begun with with \Villiam William Clancey Clancey at Stanford Stanford University University regarding regarding collaboration collaboration on the study study of of current current knowledge knowledge capturing capturing methods. methods. We We plan plan to develop develop this this collaboration collaboration in the coming coming year. year. C. Critique Resource Management Management Critique of o j Resource (None) (None) ill. RESEARCH PLANS III. RESEARCH PLANS A. Project Project Goals Plans A. Goals and and Plans Near term Near term -- Our Our research objectives objectives in the near term term can be divided divided in three three parts. First, parts. First., we are committed committed to to the design, implementation, implementation, and evaluation evaluation of of Galen, Ga.len, as as described We have completed completed an interactive interactive front front end so that that physicians physicians can described above. \Ve directly direct.& enter enter patient ont.ient data, data, and Galen's Galen’s knowledge knowledge base b%e is currently currently being "tuned" “tuned” with with tile the help of Dr. Dr. James Moller Moller MD, MD, an expert expert physician physician collaborator collaborator from from the University University of of f\Iinnesota r\linnesot,n Pediatric Pediatric Cardiology Cardiology Clinic. Clinic. During During the coming coming year, year, Galen's Galen’s performance performance will will be compareo compwred with wit,11that thn.t of of the Diagnoser Dingnoser program program and with with expert expert phYSicians. physicians. Our Our second objective objective consists of of making making extensions extensions to the knowledge knowledge capturing capturing strategies loped in our strategies deve developed our original original work work in medical medical diagnosis. In In the near term term this this work work will will examine examine descriptive descriptive strategies strategies in which which experts experts attempt attempt to use use a formalized formalized language to express what production rules), rules), observational observational strategies strategies in which which what they they know know (e.g. production experts perform t;1sks designed to reveal information from which a theory expert,s perform t.asks reveal information from which theory of of task task specific specific expl'rtise expcrtize can be bnilt, built, and intuitive intuitive strategies strategies in which which either either experts experts behave as as knowledge knowledge engineers or knowledge knowledge engineers attempt attempt to to perform perform as as pseudo experts. experts. In the coming coming year year we ,viII will also be attempting attempting to develop a program program to to automate automate the early early stages of of E. A. Feigenbanm Feigenbaum 154 164 SOLVER SOLVER Project Project 5P-11 RROO785-11 knowledge knowledge capturing, ca.pturing, analogous to the "prototype “prototype stage" stage” of design referred referred to in software software engineering. engineering. Our Our third t)hird near term term objective objective will will be be to investigate investigate one one of the central central problems problems of recognit.ion classify problems problems when solving solving them. them. recognition ba.<;ed based problem problem solving, solving, how to classify Questions Questions related relat.ed to problem problem classification classification which which we will will be be examining examining include: include: What What patterns patterns do experts experts and novices detect detect in aa problem problem that that allows them them to classify classify it it as as an instance inst,ance of aa problem problem type type that that is already already known? known? How does does an expert expert make an initial initial choice of the level of abstraction abstraction to be be used in solving solving aa problem? problem? How How can an expert expert recover recover from from an a.n initial initial incorrect incorrect choice of levels? levels? How How can the difference difference between between causal and a.nd prot,otypic prot,ot,ypic modes of reasoning reasoning be modeled as as differences differences in levels of abstraction, abstraction, and how can aa common common model for for these two two types of reasoning reasoning be be constructed? constructed? \Ve We will will be be pllrsuing pllrsuing these questions in the area of physics problem problem solving, solving, as as well well as as in medicine. medicine. Long range -- Our objective is is to improve improve the methodology methodology of the Our long range objective "knowledge “knowledge capturing" capturing” process that that occurs in the early early stages of the development development of expert expert systems systems when when problem problem decomposition decomposition and solution solution strategies stra.tegies are being specified. specified. Several Severs1 related related questions questions of interest interest include: include: What What are the performance performance consequences consequences of of different different approaches, spproaches, how can these consequences consequences be evaluated, evaluated, and what what tools tools can assist in making making the best choice? choice? How How can organizations organizations be determined determined which which not not only only perform perform well, well, but, but are structured structured so as as to facilitate facilitate knowledge knowledge acquisition acquisition from from human human experts? In t.he t,he coming coming year year we will will be exploring exploring these questions in areas of of design and management management a.' as3 ','iell well as as in law, physics physics and a.nd medicine. medicine. B. Justification for Continued Jelati jication and and Requirements Requirements for Continued SUMEX SUMEX Use Use Our Our current current model model development development takes takes advantage advantage of of the the sophisticated sophisticated Lisp Lisp programming environment programming environment, on SUMEX. SUMEX. Although Although much much current current work work with with Galen Galen is done A.,'( 11/780, benefit from using aa. version version running running on a local V VAX 11/780, we continue continue to benefit from the int.eraction by the SUMEX use int,era.ct,ion with with other other researchers facilitated facilitated by SUMEX system. system. We We expect expect to use SU~IEX to allow program. We plan to continue use SU:,lEX allow other ot,her groups groups access access to t.he the Galen Galen program. We also plan continue use of of the knowledge knowledge engineering engineering tools tools available available on SUMEX. SUMEX. C. Needs Needs and Plans for for Other Resources Beyond Beyond SUAfEX-AIA1 and Plans Other Computing Computing Resources SUibfEX-Al&f Paul Paul .Johnson Johnson is a member member of of the the group group of investigators investigators who who have recently recently submitt.ed proposal to submitt,ed a proposal to establish establish a national national computer computer network network for for cognitive cognitive scientists scientists (COGNET). In addition, addition. our our current current grant grant will will permit permit the the purchase purchase of of some single-user single-user (COGNET). computers (we are currently currently comparing comparing several several alternative alternative machines). machines). SUMEX will will computers SUMEX continue to to be used for for collaborative collaborative activities activities and for for program program development development requiring requiring cont.inue tools not not available available locally. locally. tools D. Recommendations Recommendations for for Future Future Community Community and and Resource Resource Development Development D. (None) (None) 165 A. Feigenbaum Feigenbaum E. A. Pilot Pilot Stanford Stanford Projects Projects 5P41 RR00785-11 RR00785- 11 II.A.3. Pilot Projects II.A.3. Pilot Stanford Stanford Projects Following pilot projects projects currently Following are descriptions descriptions of of the informal informal pilot currently using using the Stanford portion of resource, pending funding, St,anford portion of the SUMEX-AIM SUMEX-AIM resource, pending funding, full full review, review, and authorization. authorization. In addition progress reports presented here, abstracts project are addition to the the progress reports presented abstracts for for each project submitted submitted on a separate separate Scientific Scientific Subproject Subproject Form. Form. E. A. A. Feigenbaum Feigenbaum 156 166 CAMDA CAMDA Project Project 5P41 RfW0785-11 51’ 41 RFt00585-11 IT.A.3.1. CAMDA CAMDA Project Project II.A.3.1. CAMDA Project Project CAMDA CAMDA Research Research Staff: Staff: CAMDA Samuel Holtzman, Holtzman, Co-PI Co-PI Samuel Prof. Ronald Ronald A. A. Howard, Howard, Co-PI Co-PI Prof. Jack Breese Breese Jack Dr. Emmet Emmet Lamb Lamb Dr. Dr. Robert Robert Kessler Kessler Dr. Dr. Frank Frank Polansky Polansky Dr. Engineering-Economic Engineering-Economic Systems Systems Engineering-Economic Systems Systems Engineering-Economic Engineering-Economic Engineering-Economic Systems Systems School School of of Medicine Medicine School School of of Medicine Medicine School School of of Medicine Medicine Associated faculty: faculty: Associated Engineering-Economic Engineering-Economic Systems Systems Engineering-Economic Engineering-Econom ic Systems Systems Prof. Edison Edison Tse Tse Prof. Prof. Ross Ross Shachter Shachter Prof. I. SUMMARY OF OF RESEARCH RESEARCH PROGRAM PROGRAM I. SUMMARY A. Project Project Rationale Rationale A. The project is an attempt The Computer-Aided Comput,er-Aided Medical hiedical Decision Decision Analysis Analysis (CAMDA) (CAMDA) project attempt to t,o develop by taking develop intelligent int,elligeut medical medical decision decision systems by taking advantage advantage of of the complementary complementary methodologies methodologjes of of decision decision analysis a.nalysis and artificial artificial intelligence. intelligence. Medical Relevance Relet'ance and B. Medical und Collaboration Collaboration The primary effort project during The primary effort of of the CAMDA CA>lDA project during 1983 was focused on the design and implementation implement,st.ion of of RACHEL, RACHEL, an intelligent intelligent decision system system for for infertile infertile couples. This This effort physicians and patients patients deal with effort is aimed a.imed at at helping helping physicians with difficult difficult choices regarding regarding pertinent. RACHEL is being developed in close close cooperation cooperation between between pertinent, medical medical procedures. procedures. RACHEL t.he depart,ment of Obstetrics Obstetrics and the department depart*ment of Engineering-Economic Engineering-Economic Systems, the department Gynecology, Gynecology, and the t.he department department of of Surgery Surgery (Urology (Urology Division), Division), all at Stanford. Stanford. C. Highlights Highlights of of Research Research Progres8 Progress C.l past year year C.l Accomplishments Accomplishments this this pad The The CAMDA CAbIDA project project began in the summer summer concentrated concentrated our our efforts efforts on three three specific specific tasks: the development development of a formal formal representation representation for for uncertain uncertain decisions, the design and implementation implementat.ion of solution solution algorithms algorithms for for formal formal decision problems, problems, and the construction constructZion of an inferential inferential processor specifically specifically tailored tailored to the process of of formalizing formalizing decision problems. problems. Most Most of our our research has been based on the concept concept of an influence influence diagram diagram (Howard (Howard and Matheson, Mstheson, 1984) which which is is generalization generalization of decision trees as as aa representation representation for decision problems. Influence diagrams diagrams (IDs) (IDS) have several major major features fentures that that make problems. Influence them them attractive attractive for use use in intelligent intelligent decision systems. Technically, Technically, IDs IDS prevent prevent the loss loss of information information often often incurred incurred in constructing constructing asymmetric asymmetric decision trees (Olmsted, (Olmsted, 1984), 157 167 Feigenbaum E. A. Feigenbaum P41 P41 RR00785-11 RR00785-11 CAMDA CAMDA Project Project wit,hout. wit,hout suffering suffering from from the the explicit explicit exponential exponential growth growth of of symmetric symmetric decision decision trees. trees. Furthermore, Furt,hermore, unlike unlike decision decision trees, trees, ID ID decision decision models models ta.ke take full full advantage advantage of of probabilistic probabilistic inclependence independence relat.ions, relations, which which can can have have aa significant significant impact impact on on the the simplicity simplicity of of the the decision decision model, model, and and on on the the efficiency efficiency oC of its its formal formal solution. solution. A A particularly particularly useful useful feature feature of of influence influence diagrams diagrams which which we have recently recently begun investigating investigating isis that that fact fact that that they they can can be be used used to to represent represent deterministic determinist.ic as as well as as probabilistic probabilistic relations relations between between model model elements. elements. In In Cact, fact, deterministic deterministic relations relations can be be exploited exploited to to describe describe complex complex probabilistic probabilistic behavior. behavior. This This Ceature feature allows the construction construction of simple knowledge bases (composed primarily of deterministic knowledge bases primarily deterministic statements) statements) which which can be be llsecl used to create crea.te problem-specific problem-specific probabilistic probabilistic decision models. In addition addition to their their technical technical advantages advantages IDs IDS have been been empirically empirically shown to be intuitively int.uit,ively appealing a.ppealing to decision makers makers (Owen, (Owen, 1978), 1978), and to provide provide an excellent excellent means of communication communication between experts experts in different different fields. In the context context of our our own efforts, effort,s, we have found found that that the physicians physicians who who are participating participating in the development development of RACHEL RACHEL have ha.d had liWe little difficulty difficulty using IDs IDS as as aa simple simple representation representation for for expressing the t,he decisions they they and their their patients pat!ients face. face. Another Another important important feature feature of IDs IDS is the Cact fact that that they they are naturally naturally constructed constructed in a backwards, bsckwards. goal-directed goal-directed fashion fashion (decision (decision trees usually usually lead to to a forward-reasoning forward-reasoning approach). approach). Backward Backward development development of decision models has two two important importa.nt advantages advantages for for our our purposes. First, First, it it has a strong strong attention-focusing attention-focusing effect effect since it it encourages the dt~ci"ion decision maker maker to first first think think of what what he or she she wants, wants, and then then about about what what can be done to change the t.he world world according according to the expressed preferences. preferences. Decision Decision trees usually usually have t,he opposite effect. Thus, they often lead the decision process along paths that although t,he opposite effect. Thus, they often decision along paths that although releva.nt attention-focusing effect effect of of releva.nt to the decision decision at at hand, hand, have little little effect effect on it. it. The The attention-focusing IDs its efficiency. efficiency. The The second IDS on the t,he decision making making process tends to to contribute contribute to to its advantage of intelligent intelligent decision decision advantage of of the goal-directed goal-directed nature nature of of IDs IDS for for the the construction construction of systems is that amenable to to computer-based computer-based that it it makes the formulation formulation of of decision problems problems amenable automation a,utomation as as a rule-based rule-based system. system. Having decided decided on IDS IDs as a means to to represent represent decision decision problems, problems, we have designed Having implemented several algorithms to to solve well-formed well-formed influence influence diagrams. diagrams. This This effort effort and implemented several algorithms resulted in the development development of of a powerful powerful software package which which can generate generate has resulted software package optimal strategies their certain certain equivalent equivalent directly directly from from an ID. ID. This This package is optima.1 st.ra.tegics and their beginning to to be tested tested and a.nd augmented augmented to to make make itit easier to to use by by researchers researchers other other tha.n than beginning its developers. developers. In part, part, this this package package is based on the the work work of of Olmsted Olmsted (1983), (1983), and and on a its constructive proof proof by by Shacht.er Shachter (1984) (1984) that, that, given given certain certain technical technical features, features, shows shows that that an construct,ive influence diagram can always be solved in finite time. inflllence diagram a.lways solved in finite time. An importSant important feature feature of of RACHEL RACHEL is that that itit attempts attempts to to help help its its users in in the the An development of of models models for for their their decisions. decisions. Thus, Thus, unlike unlike most most other other decision decision analysis analysis tools, tools, development RACHEL is designed to to use domain domain knowledge. knowledge. Therefore, Therefore, a central central element element in in the the RACHEL archit,ecture of of the the RACHEL RACHEL system system is an an algorithm algorithm which which performs performs symbolic symbolic inference. inference. archit.ecture Although several several general-purpose general-purpose inference-engines inference-engines exist exist within within our our research research environment, environment, hlt.holugh found itit advantageous advantageolls to to implement implement our our own own for for reasons of of efficiency efficiency and and we have found compatibility. Furthermore, Furthermore, our our inference inference algorithms algorithms are particularly particularly well well suited suited for for the the compatibility. construction of of decision-analytic decision-analytic models. models. construcbion Finally, from from the the standpoint standpoint of of computer computer implementation, implementation, we we have have developed developed a Finally, d~tta structure structure which which allows allows us to t,o represent represent a wide wide class of of multiple-entry multiple-entry disconnected disconnected data cyclical directed directed graphs, graphs, where where both both vertices vertices and and edges can can be be associated associated with with arbitrary arbitrary cyclical data structures (such as frames). For short, we refer to these graphs as WEBs (as in in aa data structures (such frames). For short, we refer to these graphs as WEBS (as spider's), and we we have have used used them them to to represent represent a multitude multitude of of small small and and medium-sized medium-sized spider’ s), and E. A. A. Feigenbaum Feigenbaum E. 158 168 5P4 1 RR00785-11 RROOi85- 11 5P41 objects such as as objects tables, help text tables, text CAMDA CAMDA Project Project influence diagrams, diagrams, medical medical influence databases, and mathematical mathematical decision knowledge knowledge bases, bases, command command parse data (e.g., vectors and matrices). data vectors matrices). 169 159 A. Feigenbaum Feigenbaum E. A. CAr-.mA CAhlD~4 Project Project P41 P41 RR00785-11 RR00785-11 C.2 C.2 Research Research in in progress progreso The immediate project immediate goal of the the CAMDA CAMDA project is is to complete complet,e aa pilot-level pilot-level implementation implementation of the RACHEL RACHEL system system within within the next few months. months. As we define it, aa pilot pilot system system isis aIle one where the essential algorithms algorithms work work beth bcth individually individually and interactively interactively with with one another, another, operating operating with with knowledge knowledge that that is is representative representative of the system's system’s domain. domain. Such a system system lacks two two important important elements elements that that must must exist exist within within a prototype-level prototype-level implementation: implement,ation: an extensive extensive knowledge knowledge base, base, and aa front front end usable by trained trained users who who may the system. system. may not not be be familiar fa.miliar with with the details details of the To To complete complete aa pilot pilot implementation implementation of RACHEL, RACHEL, we intend intend to direct direct our our efforts efforts towards towards the following following four four tasks: incorporating incorporating aa medical medical value model elicitation elicitation facility. facility, strengthening strengthening our our influence-diagram influence-diagram solution solution procedure, procedure, improving improving the performance performance of RACHEL's implementing an explanation explanation module module to justify justify the R.4C’HEL’s inference inference engine, and implementing deci~ion decision model being developed. Once this this implementation implementation is completed, completed, RACHEL RACHEL will will be brollght broiight. to the t,he participating participating physicians physicians to begin to develop its knowledge knowledge base. base. D. D. Publication.,] Publications "IDEAS: “IDEAS: Influence Influence Diagram System", Department of of Engineering-Economic Engineering-Economic Systems, Diagram Elicitation Elicitation System “, Department Stanford Stanford University, University, Stanford, Stanford, California, California, 1983. 1. 1. Breese, J.S., Davis, Davis, D., Parnell, Parnell, G.S., and Taneja, Taneja, R.: 2. Holtzman, Ana/ysi., Holtzman, S.:"A S.: “A A/ode! Model of of the Decision Decision Analysis Process", Process II, Department Department of of Engineering-Economic Systems, Stanford Engineering-Economic Stanford University, University, Stanford, Stanford, California, California, 1981. 1981. 3. Holtzman, Decision Aid Aid for for Patients Patients with End-Stage Renal Renal Disease Disease",“, Holtzman, S.:"A S.: aA Decision with End-Stage Department Department of of Engineering-Economic Engineering-Economic Systems, Stanford Stanford University, University, Stanford, Stanford, California, 1983. California, 4. Holtzman, IrOn Holtzman, S.: “On the Use of Formal Models Models in Decision Making Making",“, Proc. o j Formal in Decision Proc. TIMS/ORSA Nat. Mtg., TIhLIS/ORSA Joint Joint Nat. Mtg., San Francisco, Francisco, May, May, 1984. 5. ((*) .. ) Holtzman, Holt,zman, Department Depa.rtment forthcoming. fort,licoming. S.: of of "Intelligent Decision Systems", Dissertation, “Intelligent Decision Systems “, Ph.D. Ph.D. Dissertation, Engineering-Economic Systems, Stanford University, Stanford Engineering-Economic University, Howard, R..4., R.A., and Matheson, Matheson, J.E.: J.E.: "Influence Diagrams", in Howard, Howard, R.A., R.A., 6. Howard, “Influence Diagram”“, Matheson, J.E. J.E. (Eds.): (Eds.): “The "The Principles Principles and and Applications Applications of of Decision Decision and Matheson, Analysis, ”II Vol. Vol. II, II, Strategic Strategic Decisions Decisions Group, Group, Menlo Menlo Pa,rk, Park, California, California, 1984. Analysis, Olmsted, S.M.: S.M.: “On IrOn Representing Representing and and Solving Solving Decision Decision Problems”, Problems", Ph.D. Ph.D. 7. Olmsted, Dissertation. Department Department of of Engineering-Economic Engineering-Economic Systems, Systems, Stanford Stanford Dissertation. University, St,anford, Stanford, California, California, 1983. University, Owen, D.L.: D.L.: "The Influence Diagrams in in Structuring Structuring Complex Complex 8. Owen, The Use oof j In jluence Diagrams Deri.'3ion Problems Problems",“, in Howard, Howard, R.A., R.A., and Matheson. Matheson, J.E. (Eds.): (Eds.): “The "The Derision Principles a.nd and Applications Applications of of Decision Decision Analysis,” Analysis," Vol. Vol. II, Strategic Decisions Decisions Principles II, Strategic Group, Menlo Menlo Park, Park, California, California, 1984. Group, Shachter, R.‘ R.:: “Evaluating "Evaluating In Influence Diagrams", working paper, paper, Department Department 9. Shacht.er. jluence Diagrams “, working of Engineering-Economic Engineering-Economic Systems, Systems, Stanford Stanford University, University, Stanford, Stanford, California, California, of 1984. A. Feigenba,um Feigenbaum E. A. 170 170 CAMDA CAMDA Project Project 5P41 5Pal RROOi85-11 RR00785-11 E. E. Funding Funding Support Support The However, The CA~mA CAniDA project project does does not not yet yet have have direct direct funding funding support. support. However, in in addition addition to to SUMEX SUAIEX computer computer usage, usage, the the project project has has benefited benefited from from aa number number of of hardware hardware gifts gifts and and research research support support for for individuals. individuals. E.1 E.l Stanford Stan ford .Medical Medical School School The The dt'partment department of of Obstetrics Obstetrics and and Gynecology Gynecology and and the department department of Surgery Surgery (Urology Samuel (Urology Division) Division) have have provided provided various various types types of of support support to the project. project. Samuel Holtzman Holtzmsn has has received research assistantship assistantship awards for several quarters. quarters. In addition, addition, the Inft'rtility Infert.ility Clinic Clinic at Stanford Stanford has has purchased purchased several terminals terminals for for the specific specific purpose of developing RACHEL and other CAMDA decision systems. developing RACHEL other CAMDA E.2 E.2 Decision Decision Systems Systems Laboratory Laboratory The CA~mA CAhlDA project project has access access to the facilities facilities of the Decision Decision Systems Laboratory Laboratory (DSL) Systems, and constitutes (DSI,) in the Department Department of Engineering-Economic Engineering-Economic constitutes the laboratory's laborat.ory3 most most active act,ive research project. project. The DSL maintains maintains several terminals, terminals, printers printers and aa personal computer computer for research on the development development of computer-based computer-based decision systems. The majority majority of the terminals terminals and printers printers were recently recently donated donated to the DSL DSL by by Qume MAD Qume Corporation. Corporation. MAD Computer Computer of Santa Santa Clara Clara has also contributed contributed to to the support support of the CAMDA CAMDA project project through through the consignment consignment of a MAD-l MAD-l persona.l personal computer, computer, and provision provision of a.a research assistantship assistantship for for Samuel Holtzman. Holtzman. IT. WITH THE RESOURCE II. INTERACTIONS INTERACTIONS WITH THE SlThtIEX-AIM SUMEX-AIM RESOURCE II.A Medical Afedical Collaborations Dissemination Via SUMEX II.A Collaborations and and Program Program Dissemination Via SUMEX Since its inception, from an active active relationship relationship inception, the CAMDA CAMDA project project has benefited benefited from between decision analysts, computer scientists, and members of the Stanford medical between analysts, computer scientists, members of the Stanford medical community. In particular, RACHEL is being developed in close cooperation with particular, RACHEL being developed cooperation with community. physicians in the Infertility Infertility Clinic Clinic at at Stanford. Most of of this this cooperation cooperation has, up up to to this this physicia,ns Stanford. Most point, consisted consisted of of an intense intense mutual mutual learning learning experience for for all all project project participants. participants. The The point, primary purpose pmpose of of this this initial initial effort effort has been to to develop develop an effective effective means to to represent represent primary medical decision decision knowledge. knowledge. As we have described described above, this this work work has culminated culminated in in the medical definition of of a representation representation language based on influence influence diagrams. diagrams. definition \Vithin the next next few months, months, RACHEL RACHEL is expected expected to to attain attain pilot-level pilot-level \\‘ ithin the performance, and its its knowledge knowledge base will will begin begin to to be deveioped. developed. At At this this point, point, most most of of performa.nce, interaction involving involving participating participating physicians physicians will will shift shift to to the the design and and the intera.ction implementation of of an infertility infertility decision decision knowledge knowledge base. This task task will will involve involve implementation This considerable direct use of the SUMEX facility by medical personnel. considerable direct, use of the SUMEX facility by medical personnel. As an added added benefit benefit of of the the development development of of Rr\CHEL, RACHEL, itit often often occurs occurs that that specific specific ,4s subsystems become useful useful in in their their own own right. right. For For instance, instance, a simple simple program program to to aid aid subsystems physicians in in determining det.ermining a course of of action action in in cases cases of of idiopathic idiopathic infertility infertility has been physicians implemented and made available available on SUMEX SUMEX to to the the staff staff of of the the Stanford Stanford infertility infertility clinic, clinic, implemented who have used itit on an experimental experimental basis. who IT.B. Sharing Sharing and and Interactions Interactions with with other other SUMEX-AIM SlThtIEX-AIM Projects Projects ILB. II.B.l SUMEX-AIM SUAIEX-AlN! 1988 1983 Workshop: Workshop: II.B.l Samuel Holtzman Holtzman chaired chaired the the working working group group on on decision decision analysis analysis and and artificial artificial Samuel intelligence in in medicine. medicine. This This group group considered considered the the current current status status and and future future of of medical medical intelligence 171 E. A. A. Feigenbaum Feigenbaum E. RR00785-11 P41 RR00785-11 CAMDA CAh.IDA Project Project A full full report report of of the the working working group's group’s deliberations deliberations and conclusions conclusions is decision decision systems. A online SUMEX file < HOLTZMAN.CAMDA>AIM-DA-FINALHOLTZMAN.CAMDA >AIM-DA-FINALavailable at SUMEX in tile < ova.ilable online REPORT.TXT. and should REPORT.TXT, should appear appear in the the forthcoming forthcoming workshop workshop report. report. lI.B.2 Participation in Knowledge Representation Representation seminar II.B.2 Participation in the Knowledge seminar at Stanford Stanford As part part of of the CAMDA CAMDA project, project, we have made several presentations presentations to to the general general Stanford medical medical and and’'computer computer science community. community. These presentations presentations have been made Stanford within jointly by within the context context of of the Knowledge Knowledge Representation Representation seminar, seminar, held jointly by the computer computer science department department and the medical medical school, and well well attended attended by by other other SUMEX SUMEX researchers researchers at Stanford. Stanford. The The speakers, and titles titles of of the the most most recent recent presentations presentations follow: follow: Samuel Holtzman: HOI tzman: Samuel On the the Design Design and Implementation Implementation of of On Computer-Based Computer-Based Decision Decision Systems. Systems. Samuel Samuel Holtzman: Holtzman: A Simple Simple Representation Representation for Uncertain Uncertain Knowledge A for Prof. Prof. Ross Shachter: Shachter: Influence Diagrams Diagrams and their their Use Use in Influence Representing and Solving Solving Complex DeciSion Decision Representing Problems. Problems. II.C. Critique Resource Management Management II.C. Critique of of Resource The The CAMDA CAMDA project project has been immeasurably immeasurably aided by by the availability availability of of the SUt-.IEX SUhlEX computing computing resources. In general general we find find the overall overall physical physical facilities facilities to to be of of excellent In excellent quality. quality. In addition, addition, we have been quite quite impressed with with the quality quality of of the particular, we have found found it it to to be a pleasure to deal with with Ed Ed SUMEX In particular, SUMEX staff. staff. Pa.t.termann, Pa.t,terma.nn, who who has ha.s been invariably invariably courteous, courteous, responsive responsive to our our needs, needs, and effective effective in his actions. act,ions. Pam Pa.m Ryalls Ryalls has also provided provided much much needed help in managing managing the CAMDA CAMDA project. project. in a manner manner that that is friendly friendly and efficient. efficient. There are, a.re, however, however, two two areas where where we feel service and performance performance could could be There improved to the t,he benefit benefit of of the entire entire SUMEX SUMEX community. community. The The first first concerns concerns the SUMEX SUMEX improved facilities facilities themselves. themselves. the other other refers refers to to our our means of of communicating communicating with with these facilities. facilities. lI.(U II.C.l SU.~IEX XrAIEX load load In the pt'riod period that that the CAMDA CAMDA project project has been active, active, we have noticed noticed a significant during significa.nt increase increa.se in the maximum maximum machine machine loading, loading, particularly particularly during weekday weekday afternoons. afternoons. Although Although this this is a normal normal feature feature of of time-shared time-shared computer computer systems, the load has become sufficiently high become.sufficiently high in in recent recent months, months, that that it it is beginning beginning to to be difficult difficult to work work on SUMEX SUMEX d1uing during business hours. hours. In addition, addition, reliability reliability has ha5 been adversely adversely affected affected in inst,a.nces. An An increase in SUMEX SUMEX computing computing capacity. capacity, or or a means of of preventing preventing some inst.ances. overloading overloading of of the machine machine should should be considered. considered. We We believe that that an emphasis on distributing some of of SUl'vlEX's SUMEX’s computing computing power power away away from from a centralized centralized mainframe mainframe could could distributing have a significant significant effect effect on reducing reducing the system system load. II.C. II.C.22 Ethernet Ethernet The The CA~lDA CALIDA project project uses uses SUMEX SUMEX almost almost exclusively exclusively through through Ethernet Et.hernet software software hardware located located in Terman Terman Engineering Engineering Center, Center, where and hardware where the department department of of Engineering-Economic Systems Systems is located. located. This Engineering-Economic This software software has on occasion been extremely extremely unreliable for for extended extended periods periods of of time, time, resulting resulting in substantially substantially reduced reduced productivity productivity for for unreliable project. Adequate Adequate communicat.ion communication facilities facilities at Stanford Stanford are of of critical critical importance importance t.o to the project. E. A. Feigenbaum Feigenbsum 172 CAMDA Project Project CAMDA 5P411 RR00785RR00785-11 5P4 11 the successful successful conduct conduct of of our our research. research. Although Although Stanford Stanford Ethernet Ethernet management management is not not the directly under under the the jurisdiction jurisdiction of of SUMEX, SUMEX, in in order order for for the the SUMEX SUMEX resource resource to to be utilized utilized directly to the the fullest, fullest, the the planning planning and and administration administration of of networking networking at at Stanford Stanford needs to to be to better coordinated. coordinated. We We have have begun begun to to explore explore several several means to to improve improve the the current current better situation. and and we believe believe that that explicit explicit SUMEX SUMEX support support of of our our efforts efforts would would be quite quite situation, beneficial. beneficial. m. III. RESEARCH PLANS PLANS RESEARCH III.A Project Project Goals Goals and and Plans Plans III.A For the the near near term t.erm future future the the primary primary goals of of the the CAMDA CAMDA project project are to to develop develop a For "pilot" and and then then “prototype” "prototype" version version 01 of the the RACHEL RACHEL system. system. Over Over an extended extended period, period, “pilot” our objective object.ive is to to arrive arrive at at useable, fully-validated fully-vaiidated and and documented documented systems systems for for support support our of medical medical decision decision making making in in infertility infertility and and other other domains. domains. of Implementat.ion of of a pilot pilot system system is primarily primarily an int.egrative integrative task task at at this this point, point, Implementation bringing together together the the medical medical knowledge knowledge base, symbolic symbolic inference inference procedure, procedure, decision decision bringing problem solution solution procedure, procedure, and and influence influence diagram diagram data data structures. structures. All of of these All problem components exist exist independently. independently. The The pilot pilot system system will will consist consist 01 of these systems systems interacting interacting components to provide provide a simplified simplified version version of of infertility infertility decision decision counseling. counseling. to The prototype prototype implementation implementation of of RACHEL RACHEL will will include include substantially substantially greater greater The amounts of of medical medical knowledge knowledge than than the the pilot. pilot. The The major major task task at at this this stage will will be the the amounts in~orporation of of expert expert knowledge knowledge regarding regarding functional functional relations, relations, probability probability distributions, distributions, incorporation a TILl decision decision alt,ernat,ives alt.ernatives in the infertility infertility domain. domain. At At this this point point in its its development, development, the the and system system will will be available available for for use by by participating participating physicians physicians at at the the infertility infertility clinic clinic on a "test" justification of “test” basis, beginning beginning the the critical critical phase of of validation validation and justification of the the system. system. A A major major goal of of the the project project is to to bring bring RACHEL RACHEL to to a IIdefensible “defensible” ll level of of performance domain. A A working working system system with wit.h full full documentation, documentation, performance in the the infertility infertility domain. expla.nation Over Over the the long long term, term, cxpla.nation of of it,s it,s conclusions, conclusions, and and user interface interface is envisioned. envisioned. infcrtilit.y infertilit,y is but, but a single single example example of of the the range of of medical medical decisions decisions amenable amenable to to decision decision analytic After RACHEL RACHEL has been fully fully implemented implemented analytic treatment t.reatment in an automated automated system. system. After and tested. tested, other other systems systems focusing focusing on cardiology cardiology or or oncology, oncology, for for example, example, might might be These systems would consist of a common core of procedural knowledge developed. developed. systems would consist of common of procedural knowledge based on decision analysis, and be instantiated with the medical decision analysis, instantiated with medical knowledge knowledge of of the particular particular domain. domain. IlI.B and for Continued SUlHEX 1II.B Justification Justification and Requirements Requirements for Continued SiYMEX Use The It It draws draws on elements elements of of decision decision The CAMDA CAMDA project project is truly truly interdisciplinary. interdisciplinary. analYSis. The project project has the potential potential to to analysis, artificial artificial intelligence, intelligence, and medical medical science. The contribute SUMEX-AIM provides provides the contribute to each of of these disciplines disciplines in important important ways. ways. SUMEX-AIM resonrces resources to continue continue this this research research with with the necessary access access to to members members of of the Stanford Stanford research community. communit,y. The The development development of of automated automated decision decision systems systems has the the potential potential to greatly greatly increase the use In the past, past, although although use and acceptance accept,ance of decision analysis analysis methods. methods. decision decision analysis analysis has be~n been shown shown to be an extremely extremely effective effective means of asSisting assisting in dccision effort involved involved in sad uncertain uncertain domains, domains, the cost and effort decision m'aking making in complex complex and producing producing an analysis analysis was prohibitive prohibitive for for most most individuals. individuals. Automated Automated decision decision analysis analysis can result result in a much much lower lower cost per user, allowing allowing decision decision theoretic theoretic techniques techniques to achieve much much wider wider application. application. The The development development of decision decision systems systems owes much much to advancements advancements in the fields of 173 173 E. E. A. Feigenbaum Feigenbaum CA~1DA Project Project CA>lDA P41 RR0078511 RR0078S-11 P41 artificial intelligence, intelligence, expert expert systems, systems, and and knowledge knowledge engineering. engineering. One continuing continuing One artificial challenge in in these these fields fields has has been been representation representation and and reasoning reasoning with with probabilistic probabilistic challenge knowledge. The The representation representation of of knowledge knowledge in in influence influence diagrams, diagrams, and and the the use use of of knowledge. decision analysis analysis in in probabilistic probabilistic reasoning reasoning are are both both significant significant topics topics of of research research being being decision pursued within within the the CAMDA CAMDA project. project. pursued For the the medical medical community community the the CAMDA CAMDA project project has has the the potential potential for for providing providing For tools and techniques that greatly improve the quality of decision making in medicine. t.ools and techniques that greatly improve the quality of decision making in medicine. RACHEL explicit.ly considers uncertainty, decision alternatives, and patient preferences in RACHEL explicitly considers uncertainty, decision alternatives, and patient preferences in developing recommendations. The objective is to develop insight and understanding The objective is to develop insight and understanding developing recommendations. regarding t,radeoffs tradeoffs and and alternatives, alternatives, both both for for the the patient patient and and the the attending attending physician. physician. rega.rding SUMEX-AIM provides provides aa unique unique resource resource for for the the continuation continuation of of the the CAMDA CAMDA SUMEX-AIM project. The available computing resources, plus access to the Stanford AI and medical project. The available computing resources, plus to the Stanford AI and medical communities are of critical importance for the successful completion of the research. communities are of critical importance for the successful completion of the research. . III.C Needs and Plans for other Computing Resources beuond SUAIEX-AIM III.C Needs and Plans for other Computing Resources beyond SUMEX-AIM \Ve are are pursuing pursuing the the purchase purchase or or donation donation of of several several computing computing resources resources for for We the Decisions Decisions System System Laboratory. Laboratory. Our Our primary primary need at at present present is for for a inst,allation in in the installation LISP machine machine (e.g., (e.g., Symbolics Symbolics 3600), 3600), enabling enabling us to to perform perform local local processing processing and and increase increase LISP our graphics graphics capabilities. capabilities. our At present present the the project access to to one MAD-l MAD-l personal personal computer computer (IBM-PC (IBM-PC type). type). At project has access \Ve are considering considering various various other other PC/workstation PC/workstation facilities facilities to to use as front front ends for for We CAMDA products. products. CAA4DA IlI.D Recommendations Recommendations for for Future Future Community and Resource Resource Development Development III.D Community and Increases in distributed distributed computing computing capabilities capabilities on the the SUMEX-AIM SUMEX-AIM system system is a Increases primary II.C.l, distributed distributed file file editing editing and and primary need at this this point. point. As we mentioned mentioned in Section Section II.C.l, graphics the mainframe. mainframe. At At this this time, time, gra.phics capabilities capabilities would would simultaneously simultaneously reduce load on the we are pn.rtiClllarly environment where where a particularly interested interested in the possibility possibility of designing designing an environment cen~ralized interact at at a high high level level with with a centralized processor processor (such (such as the the SUMEX SUMEX 20/60) 20/60) would would interact mllch much less less powerful powerful dedicated dedicated processor processor (such as as a SUN SUN workstation, workstation, or or an Apple Apple Lisa Lisa or or Macintosh) A?acint.osh) wit,h with specific specific capabilities capabilities such as as bit-mapped bit-mapped graphics graphics and special special purpose purpose hardware. hardware. E. E. A. A. Feigenbaum Frigenbsum 174 174 5P41 RR00785-11 RR00785-11 5P41 MENTOR Project Project MENTOR ll.A.3.2. MENTOR MENTOR Project Project II.A.3.2. MENTOR Project Project MENTOR Stuart M. M. Speedie, Speedie, Ph.D. Ph.D. Stuart Terrence F. F. Blaschke, Blaschke, M.D. M.D. Terrence Department of or Medicine Medicine Department Division of of Clinical Clinical Pharmacology Pharmacology Division Stanford University University Stanford SUM1viARY OF OF RESEARCH RESEARCH PROGRAM PROGRAM I.I. SUMMARY .'L Project Projat Rationale Rationale A. The goal of of the the MENTOR MENTOR (Medical (Medical EvaluatioN EvaluatioN of of Therapeutic Therapeutic ORders) ORders) project project is The to design design and and develop develop an expert expert system system for for monitoring monitoring drug drug therapy therapy for for hospitalized hospitalized to patients t,hat that will will provide provide appropriate appropriate advice advice to to physicians physicians concerning concerning the the existence existence and and patients management of of adverse adverse drug drug reactions. reactions. The The computer computer as a recording-keeping recording-keeping device device is management becoming increasingly increasingly common common in hospital-based hospital-based health health care, but but much much of of its its potential potential becoming remains unrealized. unrealized. Furthermore, Furthermore, this this information information is provided provided to to the the physician physician in in the the remains form of raw data which is often difficult to interpret. The wealth of raw data may The wealth of raw data may form of raw data which often difficult to interpret. effectively important information the patient patient from from the the physician. physician. This This is effectively hide important information about a.bout the part.icularly true by particularly true with with respect respect to to adverse reactions reactions to to drugs drugs which which can only only be detected detected by simultaneous simultaneous examinations examinations of of several several different different types types of of data data including including drug drug data, data, laboratory laboratory tests and clinical clinical signs. In In order order to to detect detect and appropriately appropriately manage manage adverse drug drug reactions, reactions, sophisticated sophisticated medical medical knowledge knowledge and and problem problem solving solving is required. required. Expert Expert systems systems offer offer the possibility possibility of of embedding embedding this this expertise expertise in a computer computer system. system. Such a system system could could automatically automatically gather gather the approprhte appropriate information information from from existing existing record-keeping record-keeping systems systems and continually continually monitor monitor for for t.lle t,he occurrence occurrence of adverse drug drug reactions. reactions. Based on a knowledge knowledge base of of relevant relevant data, data, it it, could could analyze a.nalyze incoming incoming data data and inform inform physicians physicians when when adverse reactions reactions are likely likely to occur The MENTOR MENTOR project project is an attempt attempt to explore explore the occur or when they they have occurred. occurred. The problems problems associated with with the development development and implementation implementation of such a system system and to implement implement a prototype protot,ype of a drug drug monitoring monitoring system system in a hospital hospital setting. setting. B. R. Medical Medicnl Relevance Relevance and and Collaboration Collaboration A number number of independent independent studies studies have confirmed confirmed that that the incidence incidence of adverse reactions reactions t,o to drugs drugs in hospitalized hospitalized patients patients is is significant significant and that that they they are for the most most part pa.rt preventable. preventable. Moreover, Moreover, such statistics statistics do not not include include instances instances of suboptimal suboptima,l drug drug t.herapy extended length-of-stay, length-of-stay, or ineffective ineffective therapy which which may may result result in increased costs, extended therapy. thera.py. Data Data in these areas are sparse, though though medical medical care evaluations evaluations carried carried out out as as part of hospital quality assurance programs suggest that suboptimal therapy is common. part of hospital quality assurance programs suggest that suboptimal therapy is common. Other. Other computer computer systems systems have been been developed developed to influence influence physician physician decision decision making by monitoring patient data and providing feedback. However, most of of these these making monitoring patient data and providing feedback. However, most ~ystems This shortcoming shortcoming involves involves the systems suffer suffer from from aa significant significant structural structural shortcoming. shortcoming. This evaluation In all all cases, cases, these criteria criteria consist consist of of evaluation rules that that are are used to generate generate feedback. feedback. In discrete, discrete, independent independent rules. rules. Yet, Yet, medical medical decision decision making making isis aa complex complex process process in in which which many many factors factors are are interrelated. interrelated. Thus Thus attempting attempting to to represent represent medical medical decision-making decision-making as as aa 175 175 E. E. A. A. Feigenbaum Feigenbaum 5P41 11 5P41 RR00785RR00785-11 MENTOR Project Project MENTOR discrete set set of of independent independent rules, rules, no no matter matter how how complex, complex, is a task task that that can, can, at at best, best, discrete result in in a first first order order approximation approximation of of the the process. This This places an inherent inherent limitation limitation on on result the quality quality of of feedback feedback that that can can be provided. provided. As As a consequence itit is extremely extremely difficult difficult to to the develop feedback feedback that that explicitly explicitly takes takes into into account account all all information information available available on on the the develop patient. One One might might speculate speculate that that the the lack lack of of widespread widespread acceptance acceptance of of such such systems systems may may patient. be due to to the the fact fact that that their their recommendations recommendations are often often rejected rejected by by physicians. physicians. These These systems must must be made made more more valid valid ifif they they are to to enjoy enjoy widespread widespread acceptance acceptance among among systems physicians. physicians. The proposed proposed MENTOR MENTOR system system is designed designed to to address the the significant significant problem problem of of The adverse drug drug reactions reactions by by means of of a computer-based computer-based monitoring monitoring and and feedback feedback system system to to adverse inflnence physician physician decision-making. decision-making. ItIt will will employ employ principles principles of of artificial artificial intelligence intelligence to to influence create a more more valid valid system system for for evaluating evaluating therapeutic therapeutic decision-making. decision-making. create The work work in the the MENTOR MENTOR project project is intended intended to to be a collaboration collaboration between between Dr. Dr. The at Stanford Stanford and and Dr. Dr. Speedie at at the the University University of of Maryland. Maryland. Dr. Dr. Speedie is Blaschke 3.t Blaschke spending t,he the 1983-84 academic year year on sabbatical sabbatical with with Dr. Dr. Blaschke Blaschke in the the Division Division of of spending 1983-8-I academic Pharmacology at at Stanford University. While While at at Stanford, Stanford, Dr. Dr. Speedie has been Clinical Pharmacology Clinical Stanford University. strengthening his expertise expert.ise in the the area area of of artificial artificial intelligence intelligence and establishing establishing links links in the the strengthening AI community. community. Dr. Dr. Speedie has begun begun work work on the the development development of of the the MENTOR MENTOR system system AI pilot project project on the t.he SUMEX-AIM SUMEX-AIM facility. facility. Over Over the the past past nine months, months, Drs. Drs. Blaschke Blaschke and and pilot have worked worked closely closely together together to to design the the MENTOR MENTOR project. project. The The blend blend of of Speedie ha.ve previous experience, experience. medical medical knowledge, knowledge, computer computer science knowledge knowledge and evaluation evaluation previous expertise they they represent represent is vital vital to to the the successful successful completion completion of of the the activities activities in the the design expert,ise MENTOR project. project. h!ENTOR C. Highlights Highlights of Research fiogrcsa Progress of Research The The MENTOR MENTOR project project was initiated initiated in December December 1983. The The work work to date date has consisted National Center consisted of of preparation preparation of of a grant grant proposal proposal for for the National Center for for Health Health Services Research and initial initia.1 exploration exploration of of the problem problem of of designing designing the MENTOR MENTOR system. system. Work Work has begun begun on constructing constructing a system system for for monitoring monitoring potassium potassium in patients patients with with drug drug therapy therapy that thjt can adversely adversely affect affect potassium potassium levels. E. Funding Funding Support E. Support Application Applica.tion for for grant grant support support is pending. pending. ll. WITH RESOURCE II. INTERACTIONS INTERACTIONS WITH THE THE SUMEX-AIM SUMEX-AIM RESOURCE A i\ledical Collabor.ations via .-I...•Medical Collaborations and and Program Rogram Dissemination Dissemination via SUMEX SUMEX This This project project represents represents a collaboration collaboration between between faculty faculty at Stanford Stanford University University Medical hizdical Cent.er Center and t.he the University University of of Maryland Maryland School of of Pharmacy Pharmacy in exploring exploring SUMEX, through through its communications communications computer-based SUMEX, computer-based monitoring monitoring of of drug drug therapy. therapy. returns to the University University of of capabilities. capabilities, will will facilitate facilitate this this collaboration collaboration when when Dr. Dr. Speedie returns Maryland Maryland in August August of of 1984. B. Sharing Projects Sharing and and Interactions Inteiactions with with Other Other SUAIEX-AIM SUMEX-MM Projects Interactions projects Interactions with with other other SUMEX-AIM SUMEX-AIM projects has been on an informal informal basis. bcasis. Personal Personal contacts contacts have been made with with individuals individuals working working on the ONCOCIN ONCOCIN project project concerning concerning issues issues related related to the formulation formulation of the previously previously mentioned mentioned proposal. proposal. We expect expect interactions interactions with with other other projects projects to increase significantly significantly once the groundwork groundwork has E. A. ‘4. Feigenbaum Feigenbaum 176 176 MENTOR hlENTOR Project Project SPH SF’41 RR00785-11 RRO0785-11 been been laid laid and and issues issues directly directly related related to to AI AI are are being being addressed. addressed. Given Given the the geographic geographic separation separation of of the the investigators, investigators, the the ability ability to to exchange exchange mail mail and and programs programs via via the the SUt\lEX SUMEX system system as as well we!! as as communicate communicate with with other other SUMEX-AIM SUMEX-AIM projects projects isis vital vita! to to the the success success of of the ect. t’he proj project. C. C. Critique Critique of of Resource Resource AJanagement A4anagement To To date, dn.te, the resources of of SUMEX SUMEX have been been fully fully adequate adequate for for the needs needs project. The staff staff have been been most helpful helpful with with any any problems problems we have had and and project. The fully fully sat.isfied sat,isfied with with the current current resource management. management. The The only only concern concern we have to the t.he stat·e st.8t.eof the documentation documentation on the system. of of this we are are relates ill. PLANS III. RESEARCH RESEARCH PLANS A. Project Project Goals Gouls and and Plans Plans To To accomplish accomplish the goa.ls goa.!s described in the Project Project Rationale, Rationale, aa number number of tasks will will be be underta.ken. underta.ken. The short-term short-term task task is to develop an initial initial prototype prototype of the medical medical knowledge knowledge base base and inference mechanisms mechanisms for arriving arriving at appropriate appropriate therapy therapy monitoring monitoring decisions. This This initial init,ial work work focuses focuses on monitoring monitoring for for hyperkalemia hyperkalemia and the decisionmaking making process wit.h with respect to ordering ordering potassium potassium levels. We will will then then attempt attempt to construct const,ruct a system system combining combining frames and rules that that will will model mode! this this process. The The purpose purpose of this this initial initial exercise is to explore explore the problems problems involved involved in constructing constructing an AI AI system system that that me('ts meet)s the needs needs of drug drug therapy therapy monitoring monitoring and to establish establish development development guidelines guidelines for ect. for the larger larger proj project. The project depend on the outcome The long-range long-range plans for for the the MENTOR MENTOR project outcome of of the funding project has the funding decision. However, However, assuming assuming a favorable favorable decision, decision, the full full project following following goals: 1. prototype computer patient drug 1. Implement Implement a prototype computer system system to to continuously continuously monitor monitor patient drug thernpy t,hera.py in a hospital hospit,al setting. setting. This This will will be an expert expert system system that that will will use use a modular, frame-oriented form of medical knowledge, separate inference frame-orient.ed form of medical knowledge, a separate inference modular, engin(' for for applying applying the th(' knowledge knowledge to to specific specific situations situations and automated automated engine collection of of data data from from hospital hospital information information systems to to produce produce therapeutic therapeutic collectlion ad visories. advisories. 2. Select a small small number number of of important important and frequently frequently occurring occurring medical medical settings settings !?. combination therapy therapy with with cardiac cardiac glycosides glycosides and di:retics) diuretics) that that can lead (e.g., combination to therapeutic therapeut.ic misa,dventures, misadventures, construct construct a comprehensive comprehensive medical medical knowledge knowledge to base necessary to to detect detect these situations situations using using the the information information typically typically found found b:lse comput.erized hospital hospital information information system system and generate generate timely timely advisories advisories in a comput,erized intended to to a!t,er alter behavior behavior and and avoid avoid preventable preventable drug drug reactions. reactions. intended 3. Seiect Seiect and test test several several methods methods of of formulating formulating and and providing providing physicians in order to find an optimal method of feedback that physicians in order to find optimal method of feedback that and useful useful to to physicians physicians and and is feasible to to implement. implement. and advisories to to advisories is acceptable acceptable Design and and begin begin to to implement implement an evaluation evaluation of of the the impact impact of of the the prototype prototype 4. Design MENTOR system system on on physicians’ physicians' therapeutic therapeutic decision-making decision-making as well well as on on MENTOR outcome measures measures related related to to patient patient health health and and costs of of care. outcome B. .Justification Justification and and Requirements Requirements for for Continued Continued StJMEX SUMEX Use B. This project project needs needs continued continued use use of of the the SUMEX SUMEX facilities facilities for for two two reasons. reasons. First’ First is is This that itit provides provides access access to to an an environment environment specifically specifically designed designed for for the the development development of of AI AI that 177 E. A. A. Feigenbaum Feigenbaum E. rvtENTOR MENTOR Project Project RR00785-11 5P41 RR00785-11 systems. The The MENTOR MENTOR project project focuses focuses on the development development of of such as as system system for for drug drug monitoring that t,hat will will explore explore some neglected neglected aspects of of AI AI in medicine. medicine. Access to to SUMEX SUMEX monitoring is necessary for for t.imely t,ime!y development development of of the MENTOR MENTOR system, system, as as well well as advice and a£sistance assistance in the design and development development of of a well-designed well-designed and efficient efficient system. system. Access to SUMEX SUMEX is also necessary to to support support the collaborative collaborative effort effort in this this project project as described described to previously. previously. and Pfans Other Computing Computing Resources SUMEX-AIM C. Needs Needs and Plans for for Other Resources Beyond Beyond SUAIEX-AIM A A major major long-range long-range goal of of the MENTOR MENTOR project project is to to implement implement this this system system on a independent It independent hardware hardware system system of of suitable suitable architecture. architecture. It is recognized recognized that that the full full monitoring monit.oring system system will will require require a large patient patient data data base base as well well as as a sizeable medical medical knowledge base base and must must operate operate on a close to to real-time real-time basis. Ultimately, Ultimately, the SUMEX SUMEX knowledge facilil·ies Thus facilities will will not not be s11itable suitable for for these applications. applications. Thus we intend intend to transport transport the the prototype prot.otype system system to to a dedicated dedicated hardware hardware system system that that can fully fully support support the the planned planned system system and which which can be integrated integrated into into the SUMe SUMC Hospital Hospital Information Information System. System. However, no firm firm decisions have been made about about the requirements requirements for for this this system system since However, many many specification specification and design decisions remain remain to to be made. D. Recommendations Recommendations for for Future Future Community Resource Development Development D. Community and and Resource In the brief brief time time we have been associated with with SUMEX, SUMEX, we have been generally generally In p!ca,sed with with the facilities facilities and services. However, However, it it is evident evident that that disk disk space is a critical critical pleased factor in the functioning functioning of of the facility. facility. It factor It would would seem wise to to increase disk disk storage storage in order to to meet meet the needs of of the users. Our Our experience also indicates indicates that that an attempt attempt, needs order to with to be made to organize organize and update update the the documentation documentation associated associated with the the various various SUMEX systems. Being Being new users, we found found that that paths paths to to useful useful software software was somewhat somewhat SUMEX longer longer t.han than one might might, expect. expect. An An expanded expanded introduction introduction to the system system that, that, at at least, briefly briefly described described the software software available available on SUMEX SUMEX would would be useful. E. A. Feigenbaum Feigenbaum 178 Protein Protein Secondary Secondary Structure Structure Project Project, 5P41 5P41 RH00785-11 RR00785-11 ll.A.3.3. Protein Secondary Structure Project lI.A.3.3. Protein Secondary Structure Project Protein Project Protein Secondary Secondary Structure Structure Project Robert M.D. Robert M. M. Abarbanel, Abarbanel, M.D. Section Science Section on on Medical Medical Information Information Science University of Medical University of California California Medical Center Center University of at University of California California at San San Francisco Francieco I. I. SUM1vL<\.RY PROGRAM SXJMMARY OF OF RESEARCH RESEARCH PROGRAM A. A. Projcct Project Rationalc Rationale Development Development of a protein protein structure structure knowledge knowledge base base and tools for for manipulation manipulation of that knowledge to aid in the investigation of new structures. System that knowledge investigation structures. System to include include cooperating cooperat,ing knowledge knowledge sources that that work work under under the guidance guidance of other other system system drivers drivers to find Evaluations of structure structure predictions predictions using find solutions solutions to protein protein structure st,ructure problems. problems. Evaluations known known proteins prot,eins and other other user feedbacks available available to aid user in developing developing new methods methods of prediction. prediction. B. Afcdical Medical Relcvancc Relevance and wad Collaboration Collaboration !\!any Many important important proteins proteins have been sequenced sequenced but but have not, not, as as yet, yet, had their their secondary The systems systems developed here would would aid medical medical secondary or or tertiary tertiary structures structures revealed. revealed. The scientists particular configurations, scientists in the search for for particular configurations, for for example, example, around around the active active sites in enzymes. Predictions Predictions of of secondary secondary structure structure will will aid in the determination determination of of the the full full "natural" configuration of important biological materials. Development of systems “natural” configuration of important biological materials. Development of systems such 3S as these will will contribute cont,ribute to to our our knowledge knowledge of of medical medical scientific scientific data data representation representation and retrieval. retrieval. Highlights of of Research Rcsearch Progress Progrcss C. Nighlights The prediction prediction of of beta-alpha beta-alpha protein protein structures structures is complete. complete. The The system system was The developed on a. a VAX VAX 11 /750 at at the University University of of California, California, San Francisco, Francisco, to to allow allow developed 111750 rese3Tchers to to describe describe patterns patterns of of amino amino acid acid residues that that will will be sought sought in the the sequences sequences researchers under st*udy. study. The The presence or or absence of of these “primary” "primary" patterns patterns are then then combined combined under with ot’ otht'r measures of of structure, structure, like like hydrophobicity, hydrophobicity, to to suggest possible alpha alpha helix helix or or wit,11 her measnres beta sheet sheel or or turn turn configurations. configurations. beta The segments segments of of a sequence sequence between between turns turns are then then analyzed analyzed to to determine determine the the The allowable extent extent of of the the possible possible secondary secondary structure structure assignments. assignments. Any Any segments segments remaining remaining allowable to generate generate al! all possible possible complete complete structures. structures. Only Only two two beta beta strands strands with with the the are t.hen used to character of of sheet edges are allowed allowed in in any any prediction. prediction. This This hierarchical hierarchical generation generation and and character pruning result,s r,~sults in nearly nearly 95% 95% t,urn turn prediction prediction accuracy, accuracy, and and excellent excellent delimiting delimiting of of helices pruning and sheets. In In some cases, cases, one and and only only one secondary secondary structure structure is predicted. predicted. and Research in in Progress Progress --- At At this this time, time, work work is under under way way to to extend extend this this a/P a/ f3 Research assignment work work to to a set set of of cancer cancer causing causing viral viral proteases. proteases. These These proteins proteins are believed believed t,o to .lulsant, B. and and Servan-Schreiber, Servan-Schreiber, D.:Knowledge D.:Knowledge engineering: engineering: daily 14. AA daily acit'vity on a hospital hospital ward, wurd, Computers Computers in Biomedical Biomedical Research, 1984. activity Spitzer, R. L., L., Endicot.t, Endicott, J. and and Robins, Robins, E.: Research Research diagnostic diagnostic criteria, criteria, (2d (2d 15. Spitzer, cd.) New New York York State Sta.te Department Department of of Mental Mental Hygiene, Hygiene, New New York York Psychiatric Psychiatric ed.) Institute, Biometrics Biometrics Research Research Division, Division, 1975. Institute, Spitzer, R. L.: L.: (Ed.).Diagnostic (Ed.).Diugnostic and and statistical stutistical manual manuul of of mental mental disorders, disorders, 16. Spit,zer, (3d ed.). Washingt,on, ·Washington, D. D. C.: C.: American American Psychiatric Psychiatric Associa.tion, Association, 1980. (3d 17. Vsn Van Melle, Melle, W.:The W.:The EIZffCIN EJvfYCIN Manual, Manual, Computer Computer Science Department, Department, Stanford University, Report HPP-81-16, 1981. Stanford University, Report HPP-81-16, E. Funding Funding Support Support E. \Ne have have We at the Univ of at t,he Univ of submitted submitted Michigan, hfichiga.n, an application application for for support support to to the the Vice-President Vice-President for for Research Research who has funds for "seed money" for faculty research (Total who has funds for “seed money” for faculty research (Total 199 E. A. A. Feigenbaum Feigenbaum E. 5P41 RR00785-11 RR00785-11 RXDX Project Project RXDX Direct Direct Costs Costs = = $3215). \Ve We have prepared prepared a grant grant application, application, to to be sent to to the NIH NIH "Small Program for for the May Ma.y I, 1, 1984 deadline deadline (Total (Total Direct Direct Costs = = $13,850). $13,850). “Small Grants" Grants” Program These funds funds should should enable us to to gather gather the pilot pilot data data we will will need as as part part of of a major major gran Iication. grsnt t app application. ll. WITH RESOURCE II. INTERACTIONS INTERACTIONS WITH THE THE SUMEX-.AIM: SUMEX-AIM RESOURCE A. Collaboration and and Program via SUMEX SUMEX A. Medicai AJediwl Collaboration Program Dissemination Dissemination via \Ve We are collaborating collaborating via via SUMEX SUMEX with with Dr. Dr. Jon Heiser, who who worked worked with with Ruven Ruven Brooks Brooks on HEADMED HEADMED in the late 1970's. 1970’s. We We are sharing sharing a common common SUMEX SUMEX account, account, and Dr. Heiser Heiser will will write write the section section of of the expert expert communicating communicating using computer computer mail. mail. Dr. system system dealing dealing with with the treatment treatment of of depression (and (and eventually eventually of of other other psychiatric psychiatric disorders) disorders) while while Drs. Feinberg Feinberg and Lindsay Lindsay work work on the diagnostic diagnostic parts parts of of the system. system. B. Sharing with Projects B. Sharing and and Collaboration Collaloration with other other SUAIEX-AIAf SUMEX-AIM Projects \Ve \%‘e are also collaborating, collaborating, although although more more loosely, with with Mess~s. Ivfessrs. Benoit Benoit Mulsant Mulsant and David David Servan-Schreiber. Servsn-Schreiber. They They wrote wrote an expert expert system system (BLUEBOX) (BLUEBOX) for for the the diagnosis diagnosis and treatment treatment of of depression depression which which was a first first step in the direction direction we are going. going. We We have access to BLUEBOX BLUEBOX through through SUMEX, SUMEX, and have been able to learn learn from from its its successes successes and access failures. David will, will, we expect, expect, be able to offer offer us many many helpful helpful suggestions suggestions on failures. Ben and David our our expert expert system system (RXDX) (RXDX) as as they they pursue pursue their their training training in Psychiatry Psychiatry and continue continue their their work in AI AI in medicine. medicine. work C. Critique Resource Management AJanagement Critique of of Resource We We have been using using EMYCIN EMYCIN to to set up our our knowledge knowledge base, and have found found this this program pr0gra.m invaluable, invaluable, since it it has saved us many many hours hours of of programming programming in LISP. LISP. There There are problems with with EMYCIN, EMYCIN, many many of of which which center center around around discrepancies discrepancies between between the some problems the version version of of EMYCIN EMYCIN described described in the manual manual and the version version actually actually running running on SUMEX. vVe We would would suggest that that EMYCIN EMYCIN be more more strongly strongly supported supported than than is now now the SUMEX. case. if if it it and SUMEX SUMEX are to to be even more more useful useful to to beginners beginners in AI AI in Medicine. Medicine. This This cast'. may involve involve added expense, expense, such as would would be involved involved in the purchase purchase of of an updated updated may version version of of EMYCIN, EMYCIN, but but we would would certainly certainly be able to to make make use use of of the updated updated version. version. We SUivlEX SUMEX itself itself has been invaluable. invaluable. We don't don’t have easy access access to any any other other machine ma.chine of of equal qua.1 computing computing power power which which also has a strongly strongly supported supported LISP LISP available. available. Specifically, Specifically, the Dandelion Dandelion LISP LISP machine machine at Michigan Michigan is not not easily easily accessible, while while t.he t.he LISP compiler compiler available available on the Amdahl Amdahl 5860 here differs differs from from those used at major major AI AI LISP centers such as as Stanford Stanford and MIT. MIT. We We have also made good use use of of the ARPANET ARPANET centers connections connections that that SU~IEX SUMEX offers. offers. Feinberg Feinberg will will spend a month month of of his sabbatical sabbatical working working with Prof. Prof. Peter Peter Szolovits Szolovits at at MIT, MIT, learning learning about about AI AI in Medicine. Medicine. (This (This is an a;n obvious obvious and with necessary step for for any any physician physician wanting wanting to to begin work work in the the field.) field.) This This visit visit was Lindsay and Feinberg Feinberg will will be able to arranged arranged using computer computer mail mail through through SUMEX. SUMEX. Lindsay continue continue t,heir their collaborative collaborative work work while while the latter latter is in Cambridge, Cambridge, using using the same medium. medium. The The alternative alternative would would be days lost lost in the mails mails and many many dollars dollars spent spent on phone calls. vVe We have also been able to to get rapid rapid help with with problems problems that that arise with with Er-.-IYCIN EMYCIN using computer computer mail, mail, saving saving days and/or and/or dollars. dollars. ill. III. RESEARCH RESEARCH PLAN PLAN A. Goals and and Plans A.. Project Project Goals Plans Our immediate immediate objective objective is to to develop an expert expert system system which which can cau differentiate differentiate Our patients pat.ient,s with wit,11 the various various subtypes subtypes of of depressive disorder, disorder, and prescribe prescribe appropriate appropriate Feigenbaum E. A. Feigenbaum 200 RXDX RXDX Project Project 5P41 SP41 RR0078.')-1l RR0078.5-11 treatment. This This system system should should perform perform at at about about the the level level of of aa board-certified board-certified treatment. psychiatrist, psychia.t.rist, i.e. i.e. better better than than an an average average resident resident but but not not as as well well as as aa human human expert expert in in depression. depression. Eventually, Eventually, we we plan plan to to enlarge enlarge the the knowledge knowledge base base so so that that the the expert expert system system can can diagnose diagnose and and prescribe prescribe for for aa wider wider range range of of psychiatric psychiat)ric patients, patients, particularly particularly those those with with illnesses illnesses which which are are likely likely to to respond respond to to psychopharmacological psychopharmacological agents. We will will design design the the system syst,em so so that that itit could could be be used used by by non-medical non-medical clinicians clinicians or by non-psychiatrist non-psychiat.rist MD's MD’s as as an an adjunct adjunct to to consultation consultation with with aa human human expert. expert. B. . ~tification and for continued continued SUMEX StJMEX use R. Ju Justification and Requirements Requirements for This This project project is is entirely entirely dependent dependent on access access to SUMEX. SUMEX. We are using the EMYCIN EMYCIN syst.em on SUMEX. That software is not available to us anywhere system SUMEX. That software is available anywhere else. else. We also make extensive use EX as use of SUM SUMEX as aa means of communication communication and file-sharing file-sharing with with our our consnltant" consulta.nt3, Jon Heiser, and with with David David Mulsant Mulsant and Benoit Benoit Servan-Schreiber. Servan-Schreiber. The The access access to SUMEX S’CJMEX resources isis essentially essentially our our sale sole means of maintaining maintaining contact contact with with the community community of researchers working working on applications applications of AI AI in medicine. medicine. \Ve We ant.icipate ant,icipate that that our our requirements requirements for computing computing time time and file space space will will continue continue to t,o grow grow as as the system system evolves. C. Needs and and Plans Plans for for Other Other Computing Computing Resources Resources As our our project project evolves and we run run into into the limitations limitations of of EMYCIN EMYCIN and the timetimeshared SUMEX SUMEX facility, facility, we anticipate anticipate employing employing different different expert expert systems software. software. At At this this time, time, we are not not at a stage to say exactly exactly what what that that will will be, but but our our project project is not not sufficiently sufficient,ly large that that we will will be able to mount mount such a software software development development project project ourselves, so we will depend on development and support elsewhere. Ultimately, when our our will development support Ultimately, when consultant is made available for field trials and clinical use, it will need to be transported consulta.nt available for field trials clinical use, it will to transported to a personal yet inexpensive inexpensive enough enough to to personal computer computer that that is large enough support support the system system yet be widely \Vhile current current prices prices of of the widely available. ava.ilable. A A LISP LISP machine machine is an obvious obvious candidate. candidate. While necessa.ry to drop. drop. Our Our design necessa.ry hardware hardware are too too high, high, computer computer prices are continuing continuing to st.rategy to that that which which is affordable affordable st,rategy is to t,o avoid avoid limiting limiting ourselves and our our aspirations aspirations to today; instead inst.ead we will will attempt attempt to to project project the growth growth of of our our project project and the the pricepricetoday: performance curve curve of of computing computing such that that they they meet at at some reasonable point point in the performance fut.nre. future. D. Recommendations Recommendations for for Future Future Community and Resource Resource Development Development D. Community and Valuable as a.s the the present present SUMEX SUMEX facilities facilities are to to us, they they are in in many many ways ways limited limited Valuable and awkward awkward to to use. The The need for for more more and more more computer computer cycles and memory memory cont.inues to to grow, grow, of of course. However, However, the the major major limitation limitation we feel is the the difficulty difficulty and and continues sometimes t,hr the impossibility impossibility of of making making contact contact with with everyone everyone who who could could be of of value value to to us. sometimes greater emphasis will will be put put on internetwork internetwork gateways. gateways. ItIt is important important not not \Vee hope that that, greater W’ only t,o t.o establish est.ablish more more of of these, but but to to develop develop consistent consist.ent and and convenient convenient standards standards for for only elert.ronic mail, mail, electronic electronic file file transfers, transfers, graphic graphic information information transfer, transfer, national national archives archives elect.ronic and dsta data ba.ses, ha:oes, and personal personal filing filing and and retrieval retrieval (categorization) (categorization) systems. systems. The The present present and state of of the the art art is quite quite limiting, limiting, now now that that the the basic concepts concepts of of comput.er computer networking networking have st.ste available and and have proved proved their their potential. potential. become available 201 E. A. A. Feigenbsum Feigenbaum E. Books, Papers, and Abstracts Abstra.cts 5P41 5P41 RR00785-11 RR00785-11 II.B. and ILB. Books, Books, Papers, Papers, and Abstracts Abstracts Publications Publications for the various various collaborative collaborative projects projects are summarized summarized in their their respective They also have been submitted submitted separately separately on the Scientific Scientific respective progress reports. reports. They Subproject Subproject Form Form lIB. IIB. They They are not not reproduced reproduced here to avoid avoid redundancy. redundancy. IT.C. Resource Summary Table KC. Resource Summary Table Detailed is Detailed resource usage usage information information is summarized summarized starting starting on page 30. Tabulations of this information also have been submitted separately Tabulations this information submitted separately on the requested request.ed Scientific Subproject Form. These are not reproduced Scientific Subproject Form. not reproduced here to avoid avoid redundancy. redundancy. A. Fcigenbaum Feigenbanm E. A. 202 202 Appendix Appendix A A SP41 5P41 RR0078S-11 RR00785-11 Appendix A Appendix A AIM Committee Membership AIM Management Management Committee Membership Following are the Following management mana.gement committees: committees: current current membership membership lists lists of of ALr..J Erecutiue Executive Committee: AIM Committee: FEIGENBAUM, FEIGENBAUM, Edward E.dward A., A., Ph.D. Ph.D. (Chairman) (Chairman) Principal Principal Investigator Investigator - SUMEX SUMEX Heuristic Heuristic Programming Programming Project Project Department Department of of Computer Computer Science Margaret Margaret Jacks Hall Hall Stanford Stanford University University Stanford. Stanford, California California 94305 (415) 497-4879 LEDERBERG, LEDERBERG, Joshua, Ph.D. Ph.D. President President The The Rockefeller Rockefeller University University 1230 York York Avenue Avenue New York, New York New York, New York 10021 10021 (212) 570-8080, 570-8000 KULIKOWSKI, Casimir, KULIKOWSKI, Casimir, Ph.D. Ph.D. Department Department of of Computer Computer Science Rutgers Rutgers University University New Brunswick, New Jersey 08903 New Brunswick, New (201) 932-2005 932-2006 203 the various various SUMEX-AIM SUMEX-AIM Appendix AA Appendix 5P41 RR00785RR00785-11 5P41 1I LINDB8RG, Donald Donald A.B., A.B., M.D. M.D. LINDBERG, 605 Lewis Lewis Hall Hall 605 University of of Missouri Missouri University Columbia, Missouri Missouri 65201 65201 Columbia, (314) 882-6966 882-6966 (314) (Adv Grp Grp Chrmn) Chrmn) (Adv MYERS, Jack Jack D., D., M.D. M.D. MYERS, School of Medicine School of Medicine Scaife Hall, Hall, 1291 1291 Scaife University of Pittsburgh University of Pittsburgh Pittsburgh, Pennsylvania 15261 Pittsburgh, Pennsylvania 15261 (412) 624-2649 624-2649 (412) SHORTLIFFE, Edward Edward H., H., M.D., M.D., Ph.D. Ph.D. SHORTLIFFE, Co-Principal Investigator Investigator -- SUMEX SUMEX Co-Principal Division of of General General Internal Internal Medicine, Medicine, TC117 TC 117 Division Stanford University University Medical Medical Center Center Stanford Stanford, California California 94305 Stanford, (415) 497-6970 (415) 204 204 5P41 5P41 RR00785-11 RR00785-11 Appendix Appendix A A AIM Advisory Advisory Group: AIM Group: LINDBERG, LINDBERG, Donald Donald A.B., A.B., M.D. M.D. 505 Lewis Hall 605 Lewis Hall University University of Missouri Missouri Columbia, Columbia, Missouri Missouri 55201 65201 (314) (314) 882-5955 882-6966 (Chairman) (Chairman) AMABEL, AhIAREL, Saul, Ph.D. Ph.D. Depart.ment Depart,ment of Computer Computer Science Science Rutgers Rutgers Universit.y University New Jersey 08903 New Brunswick, Brunswick, New (201) (20 1) 932-3546 932-3546 BAKER, (Exec. Secretary) Secretary) BAKER, \Villiam William R., Jr., Jr., Ph.D. Ph.D. Biotechnology Resources Program Biotechnology Resources Program National Institutes National Institutes of Health Health Building 31, Room Building Room 5B43 9000 Rockville Rockville Pike Pike Bethesda, Maryland Maryland 20205 20205 (301) (301) 496-5411 496-5411 FEIGENBAUM, FEIGENBAUM, Edward Edward A., A., Ph.D. Ph.D. (Ex-officio) (Ex-officio) Principal Principal Investigator Investigator - SUMEX SUMEX Heuristic Heuristic Programming Programming Project Project Department Department of Computer Computer Science Science Margaret Margaret Jacks Hall Hall Stanford Stanford University University Stanford, Stanford, California California 94305 94305 (415) (415) 497-4879 497-4879 KULIKOWSKI, KULIKOWSKI, Casimir, Casimir, Ph.D. Ph.D. Department Department of Computer Computer Science Science Rutgers Rut.gers University University New Jersey 08903 New Brunswick, Brunswick, New (201) 932-2006 932-2006 LEDERBERG, LEDERBERG, Joshua, Ph.D. Ph.D. President President The The Rockefeller Rockefeller University University 1230 1230 York York Avenue Avenue New York, York, New York York 10021 (212) (2 12) 570-8080, 570-8080, 5iO-8000 570-8000 MIi\SKY. MINSKY, Marvin. Marvin, Ph.D. Ph.D. Artificial Artificial Intelligence Intelligence Laboratory Laboratory Massachusetts Massachusetts Institute Institute of of Technology Technology 545 Technology Technology Square Cambridge. Cambridge, Massachusett.s Massachusetts 02139 (517) (617) 253-5854 253-5864 205 Appendix Appendix A A 5P41 RR00785-11 RR00785- 11 f"IOHLER, MOHLER, William William C., M.D. M.D. Associate Associate Director Director Division Division of of Computer Computer Research and Technology Technology National Institutes National Institutes of of Health Health Building Building 12A, Room Room 3033 9000 Rockville Rockville Pike Pike Bethesda, Bethesda, Maryland Maryland 20205 (30l) (301) 496-1168 MYERS, MYERS, Jack Jack D., M.D. M.D. School of of Medicine Medicine Scaife Hall, Hall, 1291 1291 University University of of Pittsburgh Pittsburgh Pittsburgh, Pittsburgh, Pennsylvania Pennsylvania 15261 15261 (412) 624-2549 624-2649 PAUKER, PAUPER, Stephen Stephen G., M.D. M.D. Department Department of of Medicine -Medicine - Cardiology Cardiology Tufts New England Tufts New England Medical Medical Center Center Hospital Hospital 171 171 Harrison Harrison Avenue Avenue Boston, Boston, Massachusetts Massachusetts 02111 (517) (617) 956-591O 956-5910 SHORTLIFFE, (Ex-officio) SHORTLIFFE, Edward Edward H., M.D., M.D., Ph.D. Ph.D. (Ex-officio) Co-Principal Co-Principal Investigator Investigator - SUMEX SUMEX Division Division of of General General Internal Internal Medicine, Medicine, TC117 TC117 Stanford University Medical Stanford University Medical Center Center Stanford, Stanford, California California 94305 (415) (415) 497-0970 497-6970 SIMON, SIMON, Herbert Herbert A., A., Ph.D. Ph.D. Department Depa.rt,ment of of Psychology Psychology Baker Baker Hall, Hall, 339 Carnegie-Mellon Carnegie-hlellon University University Schepley Schenley Park Park Pit.tsburgh. Pittsburgh. Pennsylvania Pennsylvania 15213 (412) (4 12) 578-2787, 578-2787, 578-2000 578-2000 205 206 Appendix Appendix A A RR00785-11 5P41 RRO0785-11 Stanford Community Advisory Advisory Committee: Committee: Stan ford Community FEIGENBAUM, Edward Ed'ward A., A., Ph.D. Ph.D. (Chairman) (Chairman) FEIGENBAUM, Heuristic Programming Programming Project Project Heuristic Department of of Computer Computer Science Department Margaret Jacks Hail Hall Margaret Stanford University University Stanford Stanford, California California 94305 Stanford, (415) 497-4879 (415) DJERASSI, Carl, Carl, Ph.D. Ph.D. DJERASSI, Department of of Chemistry, Chemistry, Stauffer Stauffer I-106 1-106 Department Stanford University University Stanford Stanford, California California 94305 Stanford, (415) 497-2783 (415) ~'IAFFLY, Roy Roy H., M.D. M.D. MAFFLY, of Nephrology Nephrology Division of Division Veterans Administration Administration Hospital Hospital Veterans Miranda. Avenue Avenue 3801 hfiranda Palo Alto, Alto, California California 94304 Palo 8.58-3971 (415) 858-3971 SHORTLIFFE, Edward Edward H., M.D., M.D., Ph.D. Ph.D. SHORTLIFFE, Co-Principal Investigator Investigator - SUMEX SUMEX Co-Principal Division of of General General Internal Internal Medicine, Medicine, TC117 TC117 Division Stanford Stanford University University Medical hjedical Center Center Stanford, Stanford, California California 94305 (415) 497-5970 497-6970 207 207 5P‘1I RR00785-11 RR00785- 11 5P41 Appendix Appendix B Appendix B Appendix B Community Growth and Project Project Abstracts Community Growth and Abstracts This appendix appendix contains contains a graphical graphical display display of the development development of the SUMEX-AIM SUMEX-AIM This community over over the years and abstracts abstracts of of currently currently active active projects. projects. Figure community Figure 15 15 below illustrat,es the substantial substantial growth growth in the cumulative cumulative number number of projects projects in the Stanford, Stanford, illustrates Nat,ional AIM, AIhf, and Rutgers-AIM Rutgers-AIM communities communities since the resource began operation operation in 1974 1974 National until this this past year. year. The The recent recent decrease decrease in the total total number number of projects projects is due to the up until severa,l long time Lime SUMEX-AIM SUMEX-AIM projects, projects, namely namely Dendral, Dendral, Puff/Vm, Puff/Vm. Act, Act, and closure of several Protein. Activity Activity in the community community however however remains remains high, as as evidenced by the number number of Prot,ein. pilot projects projects (5 (5 Stanford SLa.nford pilots, pilots, 22 Aim Aim pilots, pilots, and 11 Rutgers Rutgers pilot) pilot) currently currently active active in the pilot SUMES-AIM community. community. SUMCX-AIM 25 25 Numbzr of Projects Projects Number of Rutgers Projects Projects Rutgers 20 p=-j 4J+------L IS National AIM AIM Projects Projects National 10 ~--------~'---- r--------------------~ 5 ~/ P974 1975 1976 1977 Stanford Projects 1978 Figure Figure 15: 15: 1:)79 1930 "l!JS1 19B;? 1D83 198'~ 1985 SUh?EX-AIM SUMEX-AIM Growth Growth by by Community Community 209 :::!09 A. Feigenbaum Feigenbaum E. A. 5P41 RR00785-11 RR00785-11 Appendix Appendix B National AIM National AIM Project: Project: CADUCEUS CADUCEUS (formerly (formerly INTERNIST) INTERNIST) Principal Principal Investigators: Investigators: Jack D. Myers, Myers, M.D. M.D. (MYERS@SUMEX-AIM) (MYERSBSUMEX-AIM) Harry Harry E. Pople, Pople, Ph.D. Ph.D. (POPLE@SUMEX-AIM) (POPLEQSUMEX-AIM) University University of of Pittsburgh Pittsburgh Pittsburgh, 15261 Pittsburgh, Pennsylvania Pennsylva.nia 15261 Dr. Dr. Pople: Pople: (412) 624-3490 6243490 The major major goal of of the CADUCEUS CADUCEUS Project Project is to to produce produce a reliable reliable and adequately adequately The complete complete diagnostic diagnostic consultative consultative program program in the field field of of internal internal medicine. medicine. Although Although this this program is intended intended primarily primarily to to aid skilled skilled internists internists in complicated complicated medical medical problems, problems, program the program program may may have spin-off spin-off as a diagnostic diagnostic and triage triage aid to to physicians' physicians’ assistants, assistants, rural health health clinics, clinics, military military medicine medicine and space travel. travel. In In the the design of of CADUCEUS CADUCEUS and rural its I, we have attempted its predecessor INTERNIST INTERNIST attempted to to model model the the creative, creative, problemproblemformulation formulation aspect of of the clinical clinical reasoning reasoning process. The The program program employs employs a novel novel heuristic procedure procedure that that composes differential differential diagnoses, dynamically, dynamically, on the basis of of heuristic clinical clinical evidence. During During the course of of a CADUCEUS CADUCEUS or or INTERNIST INTERNIST I consultation, consultation, it it is not not uncommon uncommon for for a number number of of such conjectured conjectured problem problem foci foci to to be proposed proposed and investigated, with with occasional occasional major major shifts shifts taking taking place in the program's program’s conceptualization conceptualization investigated, of of the task task at hand. hand. SOFTWARE SOFTWARE AVAILABLE AVAILABLE ON ON SUMEX SUMEX Versions of of INTERNIST INTERNIST available for for experimental experimental use, use, but but the project project Versions are available continues continues to to be oriented oriented primarily primarily towards towards research and development; development; hence, a stable stable production version version of of the system system is not not yet yet available available for for general general use. use. production REFERENCES REFERENCES Pople, DIALOG model The DIALOG model of of diagnostic diagnostic Pople, H.E., H.E., Myers, Myers, J.D. and Miller, Miller, R.A.: R.A.: The logic logic and and its its use in in internal internal medicine. medicine. Proc. Proc. Fourth Fourth IJCAI, IJCAI, Tbilisi, Tbilisi, USSR, September, September, 1975. in diagnostic diagnostic problem Pople, H.E.: H.E.: The formation of composite composite hypotheses hypotheses in Pople, formation of problem solving: exercise in in synthetic synthetic reasoning. reasoning. Proc. Proc. Fifth Fifth IJCAI, IJCAI, Boston, Boston, An exercise solving: An August, August, 1977. Feigenbaum E. A. Feigenbaum 210 210 5P41 5P41 RR00785-11 RR00785-11 Appendix Appendix BB National National AIM AIM Project: Project: SECS SECS --- SIMULATION SIMULATION AND AND EVALUATION EVALUATION OF OF CHEMICAL CHEMICAL SYNTHESIS SYNTHESIS Principal Principal Investigator: Investigator: W. W. Todd Todd Wipke, Wipke, Ph.D. Ph.D. Department Department ot of Chemistry Chemistry University University ot of California California at at Santa Santa Cruz Cruz Santa Santa Cruz, Cruz, California California 95064 95064 (408) (408) 429-2397 429-2397 (WIPKE~SUMEX-AIM) (WIPKE@SUMEX-AIM) The The SECS SECS Project Project aims at developing developing practical practical computer computer programs programs to assist investigators in designing syntheses of complex organic molecules of biological investigators complex organic biological interest. interest. Key Key features features of this research include include the use use of computer computer graphics graphics to allow allow chemist chemist and computer computer to work work efficiently efficiently as as aa team, the development development of knowledge knowledge bases bases of chemical chemica.1 reactions, reactions, and the formation formation ot of plans to reduce the search for solutions. solutions. SECS is is being used used by the pharmaceutical pharmaceutical industry industry for designing syntheses of drugs. A spin-off spin-off project, project, XENO, XENO, is aimed at predicting predicting the plausible plausible metabolites metabolites of foreign foreign compounds compounds for carcinogenicity carcinogenicity studies. studies. First, First, the metabolism metabolism is simulated; simulated; then then the metabolites met,abolites are evaluated evaluated for for possible carcinogenicity. carcinogenicity. SOFTWARE SOFTWARE AVAILABLE AVAILABLE ON ON SUMEX SUMEX SECS-- An An organic organic synthesis synthesis design prograr.1 progra,m available available with with a reaction reaction library library over 500 reactions. The program is accessible to users via over reactions. The program to via a teletype teletype DEC DEC GT40 GT40 type type graphics graphics terminal. terminal. of of or or XENO-XENO-- A of xenobiotic xenobiotic compounds. compounds. A program program for for prediction prediction of metabolites metabolites of Although early development development stages, this this Although the project project is still still in the early program testing. program is available available for for preliminary preliminary exploration exploration and testing. PRXBLD-PRXBLD-- A molecular models models A facility facility for for building building approximate approximate 3-dimensional 3-dimensional molecular from The program program employs employs an from their their 2-dimensional 2-dimensional representations. representations. The energy minimization minimization approach approach and is available available both both stand-alone stand-alone and as energy part of of SECS. part QED-QED-- A domain-independent domain-independent inference inference engine which which represents represents knowledge knowledge in A first order order predicate predicate calculus. calculus. first FSECS-FSECS-- forward-working synthesis synthesis prototype prototype program program for for finding finding starting starting A forward-working A material oriented oriented syntheses. material SST-SST-- A program program for for searching searching through through a library library of of possible possible starting starting materials materials A to suggest potential starting materials for a given target molecule. to potentia.1 starting materials for given target molecule. REFERENCES REFERENCES \Vipke, W.T., W.T., Rogers, Rogers, D.: D.: Rapid Rapid Subgraph Subgraph Search Search Using Using Parallelism. Parallelism. \Vipke, Chern. If. If. Comput. Comput. Sci. (Submitted (Submitted April April 24, 1984). J. Chem. W.T.: An An Integrated Integrated System System for for Drug Drug Design, Design, in in COMPUTERS COMPUTERS \Vipke, W.T.: Wipke, A-Z: AA Manufacturer’ Manufacturer's Guide to to Hardware Hardware and and Software Software for for the the A-Z: s Guide Pharmaceutical Industry, Industry, Aster Aster Publishing Publishing C., C., Springfield, Springfield. Oregon. Oregon. Pharmaceutical (In press) press) (In Wipke, W.T., \V.T., and and Rogers, Rogers, D.: D.: Artificial Artificial Intelligence Intelligence in in Organic Organic Wipke, 211 211 E. A. A. Feigenbaum Feigenbaum E. 5P41 RR00785-11 RR00785-11 Appendix Appendix B Synthesis. An Synthesis. SST: SST: Starting Starting !vfaterial Material Selection Selection Strategies. Strategies. An Application of Application of Superstructure Superstructure Search. Search. J. Chern. Chem. Inf. Inf. Comput. Comput. Sci., 24:0000, 1984. Wipke, prediction Wipke, W.T., W.T., Ouchi, Ouchi, G.!. G.I. and Chou, Chou, J.T.: J.T.: Computer-assisted Computer-assisted prediction of CORRELATIONS of metabolism. metabolism. IN IN L. Goldberg Goldberg (Ed.), (Ed.), STRUCTURE-ACTIVITY STRUCTURE-ACTIVITY CORRELATIONS AS A PREDICTIVE. TOOL IN TOXICOLOGY. Hemisphere Publishing A PREDICTIVE.TOOL IN TOXICOLOGY. Hemisphere Publishing Corp., Corp., New York, New York, 1983, pp 151-169. Wipke, Wipke, \V.T., W.T., Ouchi, Ouchi, G. and Krishnan, Krishnan, S.: Simulation Simulation and and evaluation evaluation of of chemical synthesis--SECS. An An application intelligence chemical synthesis--SECS. application of of artificial artificial intelligence techniques. techniques. Art.ificial Artificial Intelligence Intelligence 10:999, 1978. E. A. -4. Feigenbaum Feigenbaum 212 Appendix Appendix B 5P41 5P41 RR00785-U RR00785-11 National National AIM AIM Project: Project: CLIPR CLIPR -- HIERARCHICAL HIERARCHICAL MODELS MODELS OF HUMAN HUMAN COGNITION COGNITION Principal Principal Investigators: Investigators: Walter Walter Kintsch, Kintsch, Ph.D. (KINTSCH@SUMEX-AIM) (KINTSCH@SUMEX-AIM) Peter Peter G. Polson, Poison, Ph.D. Ph.D. (POLSON@SUMEX-AIM) (POLSON@SUMEX-AIM) Computer Computer Laboratory Laboratory for Instruction Instruction in Psychological Psychological Research (CLIPR) (CLIPR) Department Department of Psychology Psychology University University of Colorado Colorado Boulder, Boulder, Colorado Colorado 80302 80302 (303) 492-6991 492-6991 Contact: Contact: Dr. Dr. Peter Peter G. Polson (Polson@SUMEX-AIM) (Polson@SUMEX-AIM) The CLIPR CLIPR Project Project isis concerned with with the modeling modeling of complex psychological psychological processes. It is comprised comprised of two two research groups. The The prose comprehension group has has processes. It completed completed a project project that that carries out out the microstructure microstructure text text analysis described by by Miller Miller a.nd and Kintsch Kintsch (1980), (1980) yielding yielding predictions predictions of the recall and readability readability of that that text text by human More recently, recently, this this group group has been been interacting interacting with with the Heuristic Heuristic human subjects. subjects. Programming Programming Project Project at Stanford, Stanford, using the AGE AGE and UNITS UNITS packages to build build a more complex model of the knowledge-based processes characteristic of prose comprehension. complex processes characteristic The The planning planning group is working working toward toward a model of the planning planning processes processes used used by by expert expert computer comput,er software software designers. designers. The initial initial processes processes involved involved in learning learning to use use computers computers and other other complex devices. SOFTWARE SOFTWARE AVAILABLE AVAILABLE ON ON SUMEX SUMEX A programs has been developed perform the microstructure A set of of programs developed to to perform microstructure text text analysis described in Kintsch Kintsch and van Dijk Dijk (Psychological (Psychological Review, 1978) and Miller Miller and Kintsch Kintsch (1980). The program accepts a propositionalized propositionalized text The program text as as input, input, and produces indices that that can be used. based in used to estimate estirnate the text's text’s recall recall and readability. readability. A A more complex complex model based AGE AGE and UNITS, UNITS, which which emphasizes emphasizes the knowledge-based aspects of of comprehension, is currently under development. currently under development. REFERENCES REFERENCES Jeffries, R., Turner, Turner, A.A., A.A., Polson, P.G. P.G. and Atwood, Atwood, M.A.: M.A.: The F’ Processes Jeffries, rocesses Involved in in Designing Des!'gning So Software. IN J.R. Anderson Anderson (Ed.), (Ed.), COGNITIVE COGNITIVE SKILLS SKILLS Inuoieled jt,ware. IN AND THEIR THEIR ACQUISITION. ACQUISITION. Hillsdale, Hillsdale, NJ, NJ, L. Erlbaum Erlbaum Assoc., 1981. (Forthcoming) AND 1981. (Forthcoming) Kieras, D.E. D.E. a.nd and Poison, Polson, P.G.: P.G.: The formal formal analysis analysis oof Kiera.s, j user complexity. Int. J. Man-Machine Studies, In Press. user complexity. Int. Man-Machine Studies, In Kintsch, W.: W.: On On modeling modeling comprehension. comprehension. Educ. Educ. Psychologist, Psychologist, 14:3-14, 1979. Kintsch, fI.,tiller, J.R. J.R. and Kintsch, Kintsch, W.: W.: Readability Readability and and recall recall oof short prose prose Miller, j short passages: A A. theoretical theoretical analysis. analysis. J. Experimental Experimental Psychology: Psychology: passages: Human Learning Learning and Memory, Memory, 1980. 1980. (In (In press) Human 213 213 A. Feigenbaum Feigenbaum E. A. 5P41 RR00785-11 Appendix Appendix B Rutgers Rutgers AIM AIM Project: Project: RUTGERS RUTGERS RESEARCH RESEARCH RESOURCERESOURCE COMPUTERS COMPUTERS IN IN BIOMEDICINE BIOMEDICINE Principal PI aincipal Investigators: Investigators: Saul Amarel, Amarel, Ph.D.[1982-83], Ph.D.[1982-831, Casimir Casimir Kulikowski, Kulikowski, Ph.D. Ph.D. Sholom Sholom M. Weiss, Ph.D.[H)83-84j Ph.D.1 1983-841 Department Department of Computer Computer Science Science Rutgers Rutgers University University New Brunswick, Brunswick, New Jersey 08903 (201) 932-3546 (AMAREL@RUTGERS) (AMARELQRUTGERS) (201) 932-2006 (KULIKO\VSKI@RUTGERS) (KULIKOWSKI@RUTGERS) (201) 932-2379 (WEISS@RUTGERS) (WEISS@IRUTGERS) The The Rutgers Rutgers Research Resource provides provides the research support support with with artificial artificial intelligence intelligence systems, and the computing computing support support with with its DEC2060 DEC2060 facility facility to a large number There are currently currently 86 investigators investigators number of biomedical biomedical scientists scientists and researchers. There associated with with the Resource. Research activities activities are concentrated concentrated in three three major major areas: expert expert medical medical systems, models for for planning planning and knowledge knowledge acquisition, acquisition, and general general AI AI systems systems development. development. One of of the most most significant significant achievements achievements in bringing bringing the work work of of the Resource to to bear bear on clinical clinical research and practice practice lies in the transfer transfer of of technology technology from from our our large DEC20 DEC20 machine machine to microprocessor microprocessor compatible compatible representations. representations. The The initial initial breakthrough breakthrough came with with the automatic automatic translation translation of of a serum serum protein protein electrophoresis electrophoresis interpretation interpretation model model so that that a version version could could be incorporated incorporated in an instrument instrument - the t,he scanning scanning densitometer densitometer (CliniScan) (CliniScan) produced produced by by Helena Helena Laboratories. Laboratories. After After testing, testing, it it was disseminated disseminated commercially, commercially, marking marking the the first first successful transfer transfer of of technology technology from from the Resource to being used in over to general general availability availability in the clinical clinical community. community. It It is now now being over one hundred hundred clinical clinical locations. locations. During period, we have started project with During the current current period, started a new project with long long term term implications implications for for the impact impact of of AIM AIM technology: technology: the development development of of a hand-held hand-held microcomputer microcomputer version version of of an expert expert consultation consultation system system for for front-line front-line health health workers. workers. In collaboration with collaboration with Dr. Dr. Chandler Chandler Dawson Dawson (UCSF), (UCSF), Director Director of of the \Vorld World Health Health Organization's Organization’s Collaborative Collaborat,ive Centre Centre for for the Prevention Prevention of of Blindness Blindness and Trachoma, Trachoma, we have developed prototype model primary eye care. This developed a prototype model for for consultation consultation on primary This has been orient.ed problems of oriented at problems of injury, injury, infection, infection, malnutrition malnutrition and cataract cataract in situations situations where where an opht,hamologist is unavailable. unavailable. In In most most developin, developingc nations, nations, the incidence incidence of of blindness blindness is ophthamologist 10% to 40% 40% higher higher than than in the the USA USA because of of these kinds kinds of of problems. problems. With With the the help of of 10% to grant from from the the CJSAID, USAID, we are developing developing the the systems needed for for management management of of eye a gra.nt by front-line front-line health health workers workers in developing developing nations, nations, and outlying outlying parts parts of of the the USA. USA. disease by disease Another significant significant technology technology transfer transfer experiment experiment involves involves a very very large Another consultation model. model. The The rheumatology rheumatology knowledge knowledge base developed developed by by our our collaborators collaborators consultation Drs. Lindberg Lindberg and Sharp Sharp at at the the University University of of Missouri Missouri has been transferred transferred by by us to to the the Drs. MC58000 microprocessor microprocessor based system, system, and and in the the past past year year testing testing has begun begun at at their their MC68000 sit.e. This This represents represents a major major step in bringing bringing the the results results of of artificial artificial int,elligence intelligence research research sit,e. to the point point where where clinical clinical researchers researchers who who do not not have access access to to large research research machines machines to will be able to to make make use of of the the results. results. We We are designing designing a specialized specialized rheumatology rheumatology will marhine which can carry out the same sophisticated reasoning that now the machine which carry out the sophisticated reasoning that now needs the Resource DEC20 DEC20 , but but will will cost cost little little over over $10,000. $10,000. Because the the transfer transfer has been Resource accomplished we can continue continue to to develop develop large scale models models using using the the full full facilities facilities of of the the accomplished Resource DEC“0 DEC20 but wit,h with the the confidence confidence that that they they can then then move move out out into into clinical clinical Resource - 9, but research environments environments when when completed. completed. research A. Feigenbaum Feigenbaum E. A. 214 5P41 5P41 RR00785-11 RR00785-11 Appendix Appendix BB REFERENCES REFERENCES Weiss, Weiss, S.M. S.M. and and Kulikowski, Kulikowski, C.A. C.A. A Practical Practical Guide Guide to to Designing Designing Expert Expert Systems, Systems, Rowman Rowman and and Allanheld, Ailanheld, H184. 1984. Kulikowski, Kulikowski, C.A. C.A. contributor contributor to the Knowledge Knowledge Acquisition Acquisition chapter chapter edited edited by B. B. Buchanan Buchanan in in the book Building Building Expert Expert Systems Systems (F. (F. HayesHayesRoth, Roth, et et aI., al., eds) eds) Addison-Wesley, Addison-Wesley, 1983 1983 (in (in press). Yao, Y. and KUlikowski, of Strategies of Reasoning Reasoning Kulikowski, C.A., C.A., It-fultiple Multiple Strategies for for Expert Expert Systems, Systems, Proc. Proc. Sixteenth Sixteenth Hawaii Hawaii International International Conference Conference on Systems Sciences, Sciences, pp. 510-514,1983.* 510-514 , 1983.) Kulikowski, in Expert Ezpert AI AI Medical Medical Consultation Consultation Systems: Systems: Kuiikowski, C.A.Progress C.A.Progresa in 1980 IQ880- 1988, 198.9, Proc. Proc. MEDINFO MEDINFO '83 ‘83 , pp. 499-502, Amsterdam, Amsterdam, August August 1983.* 1983.” Kastner, for Kastner, J.K., J.K., Weiss, S.M., S.M., and Kulikowski, Kulikowski, C.A.,An C.A.,An Efficient Efficient Scheme for Time-Dependent Time-Dependent Consultation Consultation Systems, Systems, Proc. Proc. MEDINFO MEDINFO '83, ‘83, pp.619-622, pp.619-622, 1983.* Kulikowski, Kuiikowski, C.A.Expert C.A.Ezpert Medical Medical Consultation Consultation Systems, Systems, Journal Journal of of Medical Medical Systems, v.7, pp. 229-234, 1983.* in Weiss, S.M., Ezpertise in S.M., Kulikowski, Kulikowski, C.A., C.A., and Galen, R.S.,Representing R.S.,Representing Expertise a Computer Computer Program: Program: The Serum Serum Protein Protein Diagnostic Diagnostic Program, Program, Journal Journal of of Clinical Clinical Laboratory Laboratory Automation, Automation, v.3, v.3, pp. 383-387, 1983. 1983.** Kastner, Expert System for Kastner, J.K., J.K., Weiss, S.M., S.M., and Kulikowski, Kulikowski, C.A., C.A., An An Ezpert System for Primary Eye Care, Front-Line Health Health Workers Front-Line Workers in in Primary Care, Proc. Proc. Seventeenth Seventeenth Hawaii Hawaii International International Conference Conference on Systems Sciences, Sciences, pp. 162-166, 1984. 1984.** Kulikowski, C.A. C.A. Knowledge Knowledge Acquisition Acquisition and and Learning Learning in in EXPERT, EXPERT, Kulikowski, Proc. 1983 \Vorkshop \Vorkshop on Machine Machine Learning, Learning, Univ. Univ. of of Iilinois,Champaign-Urbana Illinois,Champaign-Urbana Proc. 1983. Indicate by by an asterisk asterisk (*) (*) that that the the resource resource was given given credit. credit. Indicate 215 215 E. A. A. Feigenbaum Feigenbaum E. 5P41 RR00785-11 Appendix Appendix B National National AIM AIM Project: Project: SOLVER SOLVER -- PROBLEM PROBLEM SOLVING SOLVING EXPERTISE EXPERTISE Principal Principal Investigators: Investigators: Paul Paul E. Johnson, Ph.D. Ph.D. Center Center for for Research in Human Human Learning Learning 205 Elliott Elliott Hall Hall University University of Minnesota Minnesota Minneapolis, Minneapolis, Minnesota Minnesota 55455 (612) 373-5302 (PJOHNSON@SUMEX-AIM) (PJOHNSON@SUMEX-AIM) William William B. Thompson, Thompson, Ph.D. Ph.D. Department Department of Computer Computer Science 136 Lind Lind Hall Hall University University of Minnesota Minnesota Minneapolis, Minneapolis, Minnesota Minnesota 55455 (612) 373-0132 (THOMPSON@SUMEX-AIM) (THOMPSON@SUMEX-AIM) The The Minnesota Minnesota SOLVER SOLVER project project focuses focuses upon upon the development development of of strategies strategies for for discovering discovering and representing representing the knowledge knowledge and skill skill of of expert expert problem problem solvers. Although Although in the last 15 years considerable considerable progress has been made in synthesizing synthesizing the expertise expertise required for solving complex problems, most expert systems required for solving complex problems, most expert systems embody embody only only the limited limited amount amount of of expertise expertise that that individuals individuals are able to report report in a particular particular constrained constrained lan~uage language (e.g., production production rules). rules). What What is still still lacking lacking is a theoretical theoretical framework framework capable of reducing reducing dependence upon upon the the expert's expert’s intuition intuition or or on the near exhaustive exhaustive testing testing of possible organizations. organizations. Our Our methodology methodology consists of: (1) extensive extensive use use of of verbal verbal thinking thinking aloud aloud protocols protocols as as a source of of information information from from which which to to make inferences inferences about about underlying processes; (2) development underlying cognitive cognitive structures structures and processes; development of of computer computer models models as az a means of protocol studies; of testing testing the the adequacy adequacy of of inferences inferences derived derived from from protocol studies; (3) testing testing and refinement refinement of of the cognitive cognitive models models based upon upon the study study of of human human and model model performance in experimental problem-solving performance experimental settings. settings. Currently, Currently, we are investigating investigating problem-solving expertise physics, engineering, expertise in domains domains of of medicine, medicine, physics, engineering, management, management, and law. law. SOFTWARE SOFTWARE AVAILABLE AVAILABLE ON ON SUMEX SUMEX A A redesigned redesigned version version of of the Diagnoser Diagnoser simulation simulation model, model, named named Galen, Galen, has been implemented implemented on SUMEX. SUMEX. REFERENCES R.EFERENCES Johnson. P.E.: P.E.: What kind of of expert expert should should a sycltem system be? Johnson, What kind Medicine and and Philosophy, Philosophy. 8:77-97, 1983. J. Medicine Johnson. P.E., P.E.. Duran, Duran. A., A .. Hassebrock, Hassebrock. F., F .• Moller, Moller, J., Prietula, Prietula. M., M .. Johnson, Feltovich. P. and and Swanson, Swanson. D.: Expertise Expertise and and error error in in diagnostic diagnostic Feltovich, reasoning. Cognitive Cognitive Science 5:235-283, 5:235-283. 11981. 981. reasoning. Thompson. W.B., \V.B .. Johnson, Johnson. P.E. P.E. and Moen, Moen. J.B.: J.B.: Recognition-based Recognition-based Thompson, diagnostic reasoning. reasoning. Proc. Proc. Eighth Eighth IJCAI, IJCAI, Karlsruhe, Karlsruhe, West West diagnostic Germany. August, August. 1983. 1983. Germany, Sedlmeyer. R.L., R.L.. Thompson, Thompson. W.B. \V.B. and and Johnson, Johnson, P.E.: P.E.: Diagnostic Diagnostic Sedlmeyer, reasoning in in software software fault fault localization. localization. Proc. Proc. Eighth Eighth IJCAI, IJCAI. reasoning Karlsruhe. West West Germany, Germany. August, August. 1983. 1983. Karlsruhe, Feigenbaum A. Feigenbaum E. A. 216 216 5P41 5P41 RR00785-11 RR00785-1 I Appendix Appendix BB Johnson, Johnson. P.E., P.E., "The “The Expert Expert Mind: Mind: A A New New Challenge Challenge for for the the Information Information scientist." scientist,” in in Beyond Beyond Productivity: Productivity: Information In formation System System Development Development for for Organizational Organizational Effectiveness, Effectivenese, Th. Th. M. M. A. A. Bemelmans Bemelmsns (editor), (editor), Elsevier Elsevier Science Science Publishers Publishers B. V. (North-Holland), (North-Holland), 1984. 1984. 217 217 E. A. A. Feigenbaum Feigenbaum E. Appendix Appendix B 5P41 5P41 RR00785-11 RR00785- 11 National National AIM AIM Project: Project: Computer-Aided Computer-Aided Diagnosis Diagnosis of Malignant Malignant Lymph Lymph Node Diseases Diseases (PATHFINDER) (PATHFINDER) Principal Principal Investigator: Investigator: Bharat Bharat Nathwani, Nathwani, M.D. M.D. Department Department of Anatomical Anatomical Pathology Pathology City City of Hope National National Medical Medical Center Center Duarte, Duarte, California California (213) (213) 359-8111 359-8111 x 2456 2456 (NATHWANI@SUMEX-AIM) (NATHWANIQSUMEX-AIM) Lawrence Lawrence M. Fagan, M.D., M.D., Ph.D. Ph.D. Department Department of Medicine Medicine Stanford Stanford University University Medical Medical Center Center - Room Room TC135 TC135 Stanford, Stanford, California California 94305 (415) 497-6979 (F AGAN@SUMEX-AIM) (FAGANQSUMEX-AIM) \Ve to assist in the We are building building a computer computer program, program, called called PATHFINDER, PATHFINDER, diagnosis dia,gnosis of lymph lymph node pathology. pathology. The The project project is based at the City City of Hope National National medical medical center center in collaboration collaboration with with the Stanford Stanford University University Medical Medical Computer Computer Science Group. Group. A A pilot pilot version version of of the program program provides provides diagnostic diagnostic advice on 45 common common benign benign and malignant malignant diseases diseases of of the lymph lymph node based on 77 histologic histologic features. features. Our Our research plans are to develop a full-scale full-scale version version of of the computer computer program program by by substantially substantially increasing increa.sing the quantity quantity and quality quality of of knowledge knowledge and to develop techniques techniques for for knowledge knowledge represent.ation represent,ation and manipulation manipulation appropriate appropriate to this this application application area. The The design of of the program jCADUCEUS program strongly influenced influenced by by the INTERNIST INTERNIST/CADUCEUS program program has been strongly developed developed on the SUMEX SUMEX resource. SOFT\V ARE AVAILABLE S0FTWAR.E AVAILABLE ON ON SUMEX SUMEX PATHFINDER-A program is available for PATHFINDER-A version version of of the the PATHFINDER PATHFINDER program available for experimentation on the DEC 2060 computer. This version is a pilot experimentation DEC computer. This version pilot version program, and therefore version of of the program, therefore has not not been completely completely tested. tested. A. Feigenbaum Feigenbaum E. A. 218 218 Appendix Appendix B 5P41 5P41 RROOi85-U RR00785-11 Stanford Stanford Project: Project: EXPEX EXPEX -- EXPERT EXPERT EXPLANATION EXPLANATION Principal Principal Investigator: Investigator: Edward Edward H. Shortliffe, Shortliffe, M.D., M.D., Ph.D. Ph.D. Departments Departments of Medicine Medicine and Computer Computer Science Science Stanford University Medical Center Stanford University Medical Center - Room Room TC135 TC135 Stanford, Stanford, California California 94305 (415) 49i-6979 497-6979 (SHORTLIFFE@SUMEX-AIM) (SHORTLIFFE@SUMEX-AIM) EXPEX EXPEX isis a recent recent Stanford Stanford research project project that that is involved involved with with the development development of new representation This representation schemes schemes to facilitate facilitate knowledge knowledge acquisition acquisition and explanation. explanation. This includes includes not not only only the study study of fundamental fundamental representational representational formalisms formalisms but but also the encoding encoding of various various types types of knowledge, knowledge, such as as causal information information and user models. The The research effort effort deals with with medical medical domains domains and is being undertaken undertaken on SUMEX SUMEX or or on professional professional workstations workstations linked linked to the central central resource. Our Our interest interest in explanation explanation derives derives from from the insights insights we gained gained in developing developing explanatory that system system and its descendants, we expia,natory capabilities ca.pabilities for for the MYCIN MYCIN system. system. In that were able to generate generate intelligible intelligible explanations explanations by by taking taking advantage advantage of of a rule-based rule-based representation The limitations limitations of of the justifications justifications generated generated using MYCIN's MYCIN’s representation scheme. The explanation explanat.ion capabilities capabilities have become increasingly increasingly obvious, obvious, however, however, and have led to improved improved characterization characterization of of the kinds kinds of of explanation explanation capabilities capabilities that that must must be developed developed if EXPEX is devoted devoted to if clinical clinical consultation consultation systems are to be accepted accepted by by physicians. physicians. EXPEX the development practical and theoretical problem. development of of new practical theoretical insights insights into into this this problem. REFERENCES REFERENCES Teach, An analysis physician att-itudes Teach, R.L. R.L. and Shortliffe, Shortliffe, E.H.: E.H.: An analysis of of physician att,ltudes systems. Comput. regarding regarding computer-based computer-based clinical clinical consultation consultati.on systems. Comput. Biomed. Biomed. Res. 14:542-558, H18l. 1981. Explanatory power power for for medical Wallis, Wallis, J.\V. J.W. and Shortliffe, Shortliffe, E.H.: E.H.: Explanatory medical expert expert systems: Studies in the representation of causal relationships for clinical clinical systems: Studies in the representation of causal relationships for corlsultations. Meth. Meth. Info. Info. Med. Med. 21:127-136, 1982. consultations. Langlotz, C.P. C.P. and Shortliffe, Shortliffe, E.H.: E.H.: Adapting Adapting a consultation consultation La.nglotz, system to critique critique user user plans. plans. International International system Journal of of Man-Machine Man-Machine Studies, Studies, 19:479-496 (1983) Journal Tsuji, S. and Shortliffe, Shortliffe, E.H.: E.H.: Graphical Graphical access to the knowledge knowledge base Tsuji, of a medical medical consultation consultation system. system. Proceedings Proceedings of of AAMSI AAMSI Congress of 83, pp. pp. 551-555, San Francisco, Francisco, Ca., Ca., May May 1983. W., Clancey, Clancey, W. W. J., and and Rennels, G.: G.: Hasling, D. W., Hasling, Strateg£c explanations for a diagnostic consultation system. Strategic explanations for diagnostic consultation system. International Journal Journal of of Man-Machine Man-Machine Studies, Studies, Spring Spring 1984 (in (in press). International 219 E. A. A. Feigenbaum Feigenbaum 5P41 RR00785-11 RR00785- 11 Appendix Appendix B Stanford Stanford Project: Project: GUIDON/NEOMYCIN -GUIDON/NEOMYCIN KNOWLEDGE KNOWLEDGE ENGINEERING ENGINEERING FOR FOR TEACHING TEACHING MEDICAL MEDICAL DIAGNOSIS DIAGNOSIS Prinripal Principal Investigators: Invest,igators: \Villiam William J. Clancey, Clancey, Ph.D. Ph.D. 701 \Velch Welch Road Road Department Department of of Computer Computer Science Stanford Stanford University University Palo Palo Alto, Alto, California California 94304 (415) 497-1997 (CLANCEY@SUMEX-AIM) (CLANCEY@SUMEX-AIM) Bruce Bruce G. Buchanan, Buchanan, Ph.D. Ph.D. Computer Computer Science Department Department Margaret Margaret Jacks Hall Hall Stanford Stanford University University Stanford, Stanford, California California 94305 (415) 497-0935 (BUCHANAN@SUMEX-AIM) (BUCHANAN@SUMEX-AIM) SOFTWARE SOFTWARE AVAILABLE AVAILABLE ON ON SUMEX SUMEX GUIDON--A GUIDON--A system system developed developed for for intelligent intelligent computer-aided computer-aided instruction. instruction. Although Although it it was developed developed in the context context of of MYCIN's MYCIN’s infectious infectious disease disease knowledge knowledge base, base, the tutorial tutorial rules will will operate 0perat.e upon upon any any EMYCIN EMYCIN knowledge knowledge base. base. NEOMYCIN--A consulation NEOMYCIN--A consulation system system derived derived from from MYCIN, MYCIN, with with the knowledge knowledge base base greatly In In contrast contrast with with MYCIN, MYCIN, greatly extended extended and reconfigured reconfigured for for use in teaching. teaching. dia,gnostic dia.gnostic procedures, procedures, common common sense sense facts, facts, and disease disease hierarchies hierarchies are factored factored out out of of the basic finding/disease finding/disease associations. associations. The The diagnostic diagnostic procedures procedures are abstract abstract (not (not specific specific to to any a,ny problem problem domain) domain) and model model human human reasoning, reasoning, unlike unlike the exhaustive, exhaustive, top-down top-down approach approach implicit implicit in MYCIN's MYCIN’s medical medical rules. This This knowledge knowledge base base will will be used in the GUIDON2 GUIDON2 family family of of instructional instructional programs. programs. REFERENCES REFERENCES Clancey, Clancey, \V.J.: W.J.: Overview Overview of of GUIDON. GUIDON. In In A. A. Barr Barr and E.A. E.A. Feigenbaum Feigenbaum (Eds.), (Eds.), THE Vol. THE HANDBOOK HANDBOOK OF OF ARTIFICIAL ARTIFICIAL INTELLIGENCE, INTELLIGENCE, Vol. 2. William William Kaufmann Kaufmann Assoc., Los Altos, Altos, CA, CA, 1982. (Also (Also to to appear appear in J. of of Computer-based Computer-based Instruction) Instruction) Clancey, Methodology for for building system. Clsncey, W.J.: W.J.: Methodology building an intelligent intelligent tutoring tutoring system. In Kintsch, Kintsch, Polson, Poison, and Miller, Miller, (Eds.), (Eds.), METHODS METHODS AND AND TACTICS TACTICS IN IN COGNITIVE COGNITIVE SCIENCE. NJ. (In SCIENCE. L. Erlbaum Erlbaum Assoc., Hillsdale, Hillsdale, NJ. (In press) (Also (Also STAN-CS-81-894, STAN-CS-81-894, HPP HPP 81-18) 81-18) Clancey, Acquiring, representing, Clancey, W.J.: W.J.: Acquiring, representing, and and evaluating evaluating a competence competence model of diagnosis. In Chi, Glaser, and Farr NATURE model of diagnosis. In Chi, Glaser, Farr (Eds.), (Eds.), THE THE NATURE OF EXPERTISE. In preparation. HPP-84-2. OF E?
Source Exif Data:File Type : PDF File Type Extension : pdf MIME Type : application/pdf PDF Version : 1.6 Linearized : No XMP Toolkit : Adobe XMP Core 4.2.1-c041 52.342996, 2008/05/07-21:37:19 Format : application/pdf Creator : StreetI Description : Title : Create Date : 2001:07:17 15:29:44Z Creator Tool : Acrobat Capture Server 2.01 Modify Date : 2018:03:20 08:52:17-07:00 Metadata Date : 2018:03:20 08:52:17-07:00 Keywords : Producer : Adobe Acrobat 9.0 Paper Capture Plug-in Document ID : uuid:128fb35a-42bb-1040-ae6a-5143bd225111 Instance ID : uuid:25363a4d-6f61-1f46-a4f0-d61979dbc07d Page Layout : SinglePage Page Count : 233 Author : StreetI Subject :EXIF Metadata provided by EXIF.tools