Section 6.3.1.2 Core Research Plans C. Recovery from Execution Failure Once a problem solver observes a failure, there are a variety of actions it can take. If adequate monitoring has been performed, the source of the failure should be immediately evident. If not, the program must localize the problem in order to correct it. Once the failure is diagnosed, several responses are possible. The problem solver may scrap its efforts and start from scratch; or it may pick up from the failed state and produce another plan. In some cases, the original plan may be used but only after the effects of the failed attempt are undone. We would like to explore the techniques for dealing with failure in a variety of settings from the laboratory to the computer system, and we would like to study the trade-offs involved where several approaches are possible. E. A. Feigenbaum 76 Privileged Communication Core Research Plans Section 6.3.1.3 6.3.1.3 Causal Models Medical and scientific reasoning depend on exploiting causal relationships. We have encoded causal knowledge in production rules and other representations without separating it from empirical associations. This was a successful pragmatic approach. However, we recognize the importance of representing and manipulating causal models as a separate kind of knowledge in our reasoning programs, knowledge acquisition systems and tutoring programs. We propose using the VM program as a test-bed and point of takeoff for this research. The VM program provides real-time interpretation of the clinical Significance of measured data in the ICU. The project has considered the relation between three related sets of abstract clinical data: physiological information, measurements provided by a monitoring system, and diagnostic parameters used in patient care. For example, VM uses a patient's respiratory rate, a measured parameter, in interpreting his effort of breathing, a diagnostic parameter. VM now includes a limited set of physiological parameters, those directly related to measured data. Associations between measured, diagnostic parameters and physiology are now represented in VM when the association between these parameters is close and apparent. diagnostic parameters / \ / \ measurements ---- physiological from monitoring information system The proposed research will increase the number of physiological parameters used in the system and increase the number of interactions among the three kinds of parameters which are represented in the system. Specifically, this research will attempt to represent causal relations among physiological parameters (based on a physiological model) and among measurement parameters (based on a model of instrument function). Privileged Communication 77 E. A. Feigenbaum Section 6.3.2 Core Research Plans 6.3.2 Knowledge Utilization and Tools for Building Expert Systems 6.3.2.1 Attempt to Generalize (AGE) The AGE system is currently supported under SUMEX core research and its progress and future plans are described in Section 9.1.1 on page 137. 6.3.2.2 I Handbook The AI Handbook is also supported currently under SUMEX core research and its progress and future plans are described in Section 9.1.2 on page 145, E. A. Feigenbaum 78 Privileged Communication Core Research Plans Section 6.3.2.3 6.3.2.3 Research in Automated Consultation about Expert Systems One of the drawbacks of knowledge based systems is that they are often difficult to use. Consider, for example, a scientist trying to solve a problem with a computer system he does not fully understand, and assume that he has encountered a problem due to his lack of knowledge of that system (say MACSYMA or MOLGEN). For example, he may be unaware of the Capabilities available, not know the system's vocabulary, or he might get an answer he didn't expect. The simplest way for him to acquire just the information he needs is to ask a consultant for help. Consultation is a method widely used in computer centers; and, as complex programs become more pervasive and more complex, the need for consultants will grow. Unfortunately, consultants are scarce, expensive, and often unavailable when needed. One partial solution to this dilemma is in the form of automated consultation about the use of complex programs. Recent work by Genesereth in building an automated consultant for MACSYMA demonstrates the feasibility of the approach, but there are many problems to be solved before such consultants are put into general use. Genesereth'’s program deals primarily with a user's violated expectations about a system and tries to uncover and correct the misconception responsible for those expectations. It assumes that the user's actions are rational, i.e. that he has some plan for achieving his goal. This plan explains why he chose the operations he did in terms of his beliefs about those operations. The key to the identification of the user's misconception is the recognition and debugging of this plan. In the years ahead, we propose to extend and develop this "plan recognition" approach and apply it to some of the computer capabilities available in the SUMEX resource. The importance of automated consultation should not be overlooked. In general, a consultant is necessary whenever one is faced with a problem solving situation in a domain one does not fully understand. The lack of knowledge may be incidental, as it is when the domain is fairly simple but time constraints make it impossible for the individual to learn all that is necessary (e.g. with a simple text editor). Or, it may be essential, as when the domain is very complex and the user can't possibly learn everything (e.g. chemistry or MACSYMA). Privileged Communication 79 E. A. Feigenbaum Section 6.3.2.4 Core Research Plans 6.3.2.4 EMYCIN EMYCIN is a tool for building consultation systems within a backward- chaining framework. For small domains in which experts' judgmental knowledge is expressible in conditional rules, EMYCIN can provide rapid feedback on the adequacy of a rule set for providing reliable consultations. The INTERLISP version of EMYCIN is ready for use by others now. Future work includes the following: (1) Translation of EMYCIN for broader export (2) Incorporation of strategic and structural information to integrate the information needed for tutoring and for acquiring new knowledge The development of GUIDON, a case method tutor for EMYCIN knowledge bases, has given us a new perspective on the nature of the expertise that we have captured in our programs, and suggests guidelines for both representation and acquisition of knowledge. In particular, we have found that rules conveniently separate relationships into readily accessible associations, but an adequate knowledge base for teaching and acquisition requires the addition of structural knowledge (clusters and patterns), Support knowledge (underlying causal mechanisms), and strategical knowledge {managerial approaches). The strategical model is expressed in terms of rules in which the goal or action part is a task to carry out and the premise part consists of steps for achieving the goal. The strategical model will provide the foundation for a new version of EMYCIN which will encourage EMYCIN clients to incorporate in their specialized consultation systems the knowledge we have found to be useful for teaching. We believe the strategical rules will be useful for controlling inferences as well as for teaching. Implementation of this idea requires that MYCIN's rule interpreter be modified slightly to recognize that some rules describe "tasks" that may be done repetitively, unlike “inference rules" which it only considers once for any case. With the addition of structural knowledge described below, MYCIN's backward-chaining interpreter can then be used to do hypothesis formation with focusing and non- exhaustive search. Structural knowledge consists of clusters and patterns of rules and parameters~--distinctions made by the strategical rules in their control of diagnostic reasoning. The central form of structural knowledge will be a taxonomic classification of the problem space. In MYCIN this will take the form of parameters that are hierarchically related to one another and share properties. One portion of the classification is shown below. E. A. Feigenbaum 80 Privileged Communication Core Research Plans Section 6.3.2.4 INFECTIOUS-DISEASE / \ / \ / \ MENINGITIS other infections... / \ / \ / \ / \ TYPE .ACUTE TYPE.CHRONIC / \ / | \ / \ / | \ BACTERIAL VIRAL FUNGAL TB PARTIALLY-TREATED-BACTERIAL | / \ | / \ Forming this classification involves regrouping the existing rule set, creating a new parameter for each node in the hierarchy. This design of taxonomic organization and inheritance of properties wilt make MYCIN's representation more "frame-like,” while preserving the use of rules to make judgmental associations among the parameters. Because the strategical rules embody a weak model of diagnostic behavior, we believe that they constitute a backbone that will be useful for multiple problem areas. In particular, the strategical backbone could be used to structure a knowledge acquisition dialogue. In addition to encouraging a taxonomic classification of parameters, the strategical rules indicate what other kinds of knowledge the expert building an EMYCIN system will have to specify. For example, it is important to detail the knowledge that suggests a broad category of problems that merit attention in a particular case ("triggering associations") and knowledge to adequately discriminate a case on the basis of the taxonomic distinction. Representing the diagnostic and strategical knowledge in a uniform formalism of rules and parameters, and using an accepted backbone of strategical knowledge, will enable us to use GUIDON for teaching from any new EMYCIN-based program without needing to reorganize the consultation knowledge base. The teaching program will be able to teach a student how to approach cases, while the consultation program will direct its problem- solving according the same approach, one that might be more acceptable to physicians because it is patterned after their methods for solving problems. Privileged Communication 81 E. A. Feigenbaum Section 6.3.3 Core Research Plans 6.3.3 Knowledge Acquisition Our research on knowledge acquisition to date has largely focused on adding new knowledge to an existing knowledge base. A long-term effort is proposed in which we focus on acquiring the structure and contents of a whole knowledge base. The keystone of our approach to knowledge acquisition is the belief that there is a substantial overlap in the knowledge of many different task domains. We are not referring here to superficial facts and rules (say of the physical world) but rather to the abstract structure implicit in even quite disparate domains. For example, the notion of a hierarchy is found in biological taxonomy, the classification of geologic time, and business organization charts. The advantage of recognizing such abstract structures is that they often possess efficient representations and efficient algorithms for reasoning about them. In the past this commonality has not been exploited. One reason is the difficulty of representing these abstract structures in a form directly useable in different domains. Another problem is the difficulty of finding and piecing together the structures appropriate to a novel domain. We believe there is an elegant solution for these problems via the notions of abstraction and simulation structure described below, and we propose to develop a library of useful abstractions together with their specialized representations and algorithms from which a knowledge engineer can pick and choose in assembling expert programs. More specifically, we propose to expend our effort in four major directions: (1) encoding useful abstractions and simulation structures, (2) exploring the use of abstractions in checking the consistency and completeness of knowledge bases, (3) automated selection of simulation Structures, (4) the use of abstractions in understanding analogy and the use of analogies in identifying abstractions. (1) A Library of Abstractions and Simulation Structures There are an infinite number of possible abstractions. What motivates us to talk of a finite library is the fact that certain abstractions have data representations or algorithms that are particularly efficient or powerful. Some examples are trees, partial orders, rings, groups, and monoids. We propose to differentiate simulation structures on the basis of their representational economy and deductive power. For some Structures, this economy and power outweighs the uniformity of semantic networks and frames. We intend to include only those abstractions for which this is the case. A certain amount of theoretical work must precede the construction of this library. We must devise an adequate language for describing simulation structures and develop data and algorithm representations that facilitates their interface and direct application in new domains. The recent work on abstract operations by Barton, Genesereth, Moses, and Zippel should help in this effort: E. A. Feigenbaum 82 Privileged Communication Core Research Plans Section 6.3.3 (2) The Use of Abstractions in Checking Consistency and Completeness An abstraction prescribes a set of axioms that must be satisfied by all its models. These axioms can be used to check the consistency and completeness of the assertions a knowledge engineer makes in describing his task domain. For example, if a knowledge representation system suspected that a group of assertions was intended to describe a hierarchy, it could detect inconsistent data, such as cycles or multiple parents, and incomplete data, such as nodes without parents. The abstractions appropriate to the task domain are determinable from a number of sources. The user may directly name the abstraction or describe it with an analogy; or the system may be able to infer it from partial information, (3) Modeling 7 The use of models is a time-renowned problem solving technique. For example, architects and ship builders use models to get answers that would be too difficult to obtain using purely formal methods. We would like to draw an analogy between the architect's use of a physical model and the expert system's use of a simulation structure, In both cases the / advantages to be gained are power and efficiency in reasoning about their domains. Most knowledge representation systems store assertions in a uniform, domain-independent formalism like predicate calculus or semantic networks or frames. While there are advantages to uniformity and domain independence, these representations are in many cases considerably less efficient than specialized data structures, and the associated algorithms are often less efficient and less powerful. We are proposing to develop a systematic way of describing when well-known data representations and well- known algorithms are applicable and to devise a program able to employ simulation structures automatically in representing knowledge, given the abstractions it satisfies. (4) Analogies Many analogies are best understood as statements that the situations being compared share a common abstraction. For example, when one asserts that the organization chart of a corporation is like Linnaean taxonomy, what he is saying is that they are both hierarchies. This view of analogy can be turned around and used to help novice users of our abstraction library in finding appropriate entries. Imagine an engineer describing the classification of time in geology (epochs, eras, periods, etc.) who can tell the system that his knowledge base is like that of biological taxonomy and have it infer and use the hierarchy abstraction. In order to realize this goal, a number of problems must first be solved. The fundamental problem is completing a partial interpretation of Privileged Communication 83 E. A. Feigenbaum Section 6.3.3 Core Research Plans an abstraction. Once we have a method for completing interpretations, analogy understanding (or at least the bit of it we are considering) becomes easy. The system merely checks each of the abstractions of the comparison domain, testing to see whether it is applicable. Sometimes the system may not have a suitable prestored abstraction, and this process will fail. Understanding an analogy in this situation requires the invention of a new abstraction. We are interested in applying and extending the concept formation techniques of Hayes-Roth, Mitchell, and Dieterrich and Michalski in building a program to formulate new abstractions automatically. Of course, a new abstraction will not initially have any specialized data structures or algorithms, but it can provide the next system builder with the techniques developed by the originator. —E. A. Feigenbaum 84 Privileged Communication Core Research Plans Section 6.3.4 6.3.4 Explanation Our motivation for making explanation a primary focus of our research is a belief that expert systems will not be accepted by physicians or scientists unless the systems are able to justify the decisions they make. When important real world domains are involved, human decision makers are loathe to consult machines unless they understand and agree with the basis for the advice. This constraint not only forces us to consider mechanisms for generation of explanations, but it also impacts on the design of the underlying reasoning and representation techniques used by the rest of the consultation system. In the case of MYCIN and its descendents, we have been able to generate intelligible explanations by taking advantage of our rule-based representation. Rules can be translated into English for display to a user, and their interactions can also be explicitly demonstrated. By adding mechanisms for understanding questions expressed in simple English, we were able to create an interactive system that allowed physicians to ‘convince themselves that they agreed with the basis for the program's recommendations. MYCIN's explanation capabilities have been thoroughly discussed elsewhere [26]. ° MYCIN's explanation capabilities were generalized in EMYCIN and thus became available for any EMYCIN consultation system. They were further modified and utilized in both TEIRESIAS and GUIDON. Although we had experienced problems using MYCIN's rules for certain kinds of explanations (e.g., control mechanisms that were sometimes encoded in rules, or algorithmic knowledge such as the mechanisms for drug selection), it was in the setting of GUIDON that the inadequacies of MYCIN's approach became most apparent. Consider, for example, a simple MYCIN rule such as: If: the patient is less than 8 years old Then: don't give tetracycline This rule is totally adequate for MYCIN's decision making task, and would be understood by most physicians if it were used in an explanation, but it is obvious to a casual observer that it contains a giant leap in logic. It is accordingly difficult for GUIDON to teach this rule to a novice medical Student because the underlying pathophysiologic knowledge (i.e., that tetracycline is deposited in the developing bone and teeth of youngsters, weakening the former and disfiguring the latter) is not explicitly represented in MYCIN. Examples such as this one emphasize that a variety of knowledge forms are necessary if an intelligent system is to customize its explanations to the individual who is using the program. Underlying structural and causal relationships are generally required in addition to the high Jevel judgmental rules that had contained almost all of the domain knowledge in MYCIN and the other EMYCIN systems. During the second half on 1979 we formed a weekly seminar group to analyze the characteristics of good explanations. We generally tried to keep our discussions separate from computer science issues, concentrating instead on the psychology of explanation and planning to return eventually Privileged Communication 85 E. A. Feigenbaum Section 6.3.4 Core Research Plans to consider ways in which our developing theory might be implemented in knowledge-based consultation systems. Although there are several subproblems, it was agreed that the problems of explanation can generally be divided into four categories: (1) modeling the knowledge of the system user; (2) selecting a response strategy; (3) modeling contextual information regarding the interaction; and (4) understanding the question. One goal of our proposed work, then, is to build an explanation system which explicitly addresses all four of these topics. We shall briefly discuss each point: (1) Modeling the User's Knowledge: GUIDON and other ICAI systems have recognized the need to keep an internal model of the student, i.e., what he has shown he knows, what you have already told him, and perhaps a record of where his greatest weaknesses lie. Similarly, it is clear than an expert human consultant customizes his explanations so that they can be understood by the person requesting the consultation (and are thereby maximally convincing). The expert starts with certain suppositions about his client's knowledge (e.g., a teacher may presume his student is starting from scratch, but a cardiologist will assume that another physician requesting advice probably already knows a fair amount of cardiology). The default presumption is modulated, however, as the interaction proceeds and the client demonstrates his strengths or weaknesses. We have recently begun some experiments to investigate methods for encoding, along with the domain knowledge, the complexity and importance of that knowledge. These two parameters seem to be independently important in deciding whether to include a given reasoning step in an explanation. "Key" points (i.e., those that are highly important) probably should be mentioned even if they are not complex and are likely to be known to the user, On the other hand, less important but complex items probably need not be mentioned unless an expert user is really pressing for details of a decision pathway. Thus, static measures of complexity and importance can be compared with user descriptors that are initially assigned by default (depending upon the status of the user, e.g., expert vs. student), but are later altered dynamically in response to the course of the dialog and what it has revealed about the user's background knowledge. These ideas have been encoded in a small computer program which uses a limited knowledge base of rules and associations from the domain of pharyngitis (sore throats). We have experimented with a semantic network representation in which the nodes are values of attributes and rules are only one form of link between nodes. Ati nodes and rules have complexity and importance measures associated with them. An "opinion" regarding a specific patient can be represented as a subset of the nodes in the network, plus the links between them that account for how it has been determined which nodes are active. In this setting, a question tends to ask how it has been determined that a given node is active for a given patient. The appropriate explanation could be very complex if an effort were made to explain every link leading from data observations to the node descriptor in question. A customized explanation is therefore generated E. A. Feigenbaum 86 Privileged Communication Core Research Plans Section 6.3.4 based on three variables which can be dynamically manipulated by the program: (1) the focus of the dialog (e.g., broad-based vs. localized), (2) the expertise of the user, and (3) the degree of generality which is appropriate. These three variables are clearly not independent, and we are experimenting with ways to have their values manipulated in a reasonable fashion as the dialog proceeds. This early effort will provide the basis for further discussions in Year 1 of the proposed work. We have been fortunate to enlist the collaboration of an endocrinologist at Stanford, Dr. Larry Crapo, who is eager to work with us on building an endocrinology knowledge base. It is likely that we will select the pathophysiology of thyroid disease, or of the pituitary adrenal axis. Both these domains are appealing for computer- based representation because the relationships are well-understood and there are some challenging problems of feedback homeostasis that will need to be represented. During Year 02 we will encode this knowledge base in detail and begin experiments on the generation of explanations using the kinds of techniques outlined above. (2) Selecting A Response Strateqy: Our explanation efforts to date have tended to be simple reiterations of individual reasoning steps, but it is clear that experts and teachers use several alternate strategies for conveying their ideas or key facts. Many of these techniques draw upon common sense world knowledge (e.g., analogies with familiar concepts outside the domain), but we have thus far failed to capitalize on these teaching strategies in our work. Thus another goal of the work that lies ahead will be to develop structures for drawing parallels or otherwise representing the strategies used by good “explainers." (3) Modeling Contextual Information Regarding the Interaction - We have already mentioned some of the ways in which contextual information may be useful in determining the best way to answer a question. For example, a more accurate model of the user's knowledge can be developed over time, and the extent to which a given conversation is focused on a particular local topic can be assessed. Note that we are emphasizing here issues other than those related to natural language understanding; computational linguists also often cite the need to record contextual dialog information in order to handle problems such as anaphora. An understanding of the "flow" of a dialog is also important in understanding the meaning of subsequent questions, as we discuss below. (4) Understanding The Question This issue interfaces with the problem of natural language understanding, but we view it in a somewhat different light. We emphasize instead the ways in which the model of the user and contextual information may allow us to disambiguate questions. To draw from a medical example Privileged Communication 87 E. A. Feigenbaum Section 6.3.4 Core Research Plans that we have frequently discussed, consider the following scenario. A reasoning program for pharyngitis diagnosis and management has just diagnosed strep throat and recommended penicillin and the user asks the question "Why would you give penicillin?” In the most obvious case, one might imagine a response that itemizes the risks of streptococcal infections and the reasons for treating early with penicillin. Similarly, one might expect a more detailed response for a student and a quick summary for a physician using the system. However, an alternate interpretation is that EVERY physician knows the theoretical reasons for giving penicillin in strep pharyngitis, and that if the user is a physician and is asking the question then he must be asking something different than the simple informational question. In this case the query might be interpreted as a challenge (one that might have been conveyed by tone of voice if it had been asked of a human consultant). Apparently the user has reason to doubt that penicillin was the appropriate agent in this case, or thinks that no drug was required. Other background information and contextual knowledge should also help, and an intelligent program might thereby answer the question in a given case in any of the following ways: "Because the patient has pre-existing rheumatic heart disease." "Because I doubt that he is allergic to penicillin, even though he reported that he is.” "Because he is unreliable and I am afraid I will not be able to reach him to call him back if his strep culture comes back positive." "Because I tend to treat conservatively and give penicillin for strep throat even though I know there hasn't been a case of rheumatic heart disease in California in over 10 years." Note how different these kinds of explanations are from the simple justification that a program such as MYCIN might have given: "Because streptococcal pharyngitis may be followed by rheumatic myocarditis or glomerulonephritis, mediated by immune complexes, and I can prevent this complication by giving penicillin (to which streptococci are uniformly sensitive)." The ideal intelligent assistant should be able to determine from knowledge of the user, the domain, the individual case, and the context of the dialog, which of the preceding responses is most appropriate. We will attempt to identify methods for giving our program this kind of capability. E. A. Feigenbaum 88 Privileged Communication Available Facilities 7 Available Facilities The existing SUMEX-AIM computer and communications facilities have been described in earlier sections. The number of personnel to support this follow-on work will remain at approximately the same level as before so no additional office space will be required. The additional equipment (VAX's, file server, and PWS's) will be accommodated in the existing SUMEX machine room, a portion of the Pine Hall machine room allocated to Prof. Feigenbaum, and in existing individual office areas. Technician support and hardware development for this equipment will be housed in the existing SUMEX electronics laboratory. Privileged Communication 89 E. A. Feigenbaum Literature Cited 8 10, 11. Literature Cited Feigenbaum, E.A., The Art of Artificial Intelligence: Themes and Case Studies of Knowledge Engineering, Proceedings of the 1978 National Computer Conference, AFIPS Press, (1978). Nilsson, N.J., Principles of Artificial Intelligence, Tioga Publishing Company, Palo Alto, California (1980). Winston, P.H., Artificial Intelligence, Addison-Wesley Publishing Co., (1977). Nilsson, N.J., Artificial Intelligence, Information Processing 74, North-Holland Pub. Co. (1975). Barr A. and Feigenbaum, E.A. (Eds.}, The Handbook of Artificial Intelligence, Stanford University Department of Computer Science, forthcoming. Boden, M., Artificial Intelligence and Natural Man, Basic Books, New York, (1977). McCorduck, P., Machines Who Think, W.H. Freeman and Co., San Francisco (1979). Coulter, C. L., Research Instrument Sharing, Science, Vol. 201, No. 4354, August 4, 1978. Stefik, M., An Examination of a Frame-Structured Representation System, Proceeding of the Sixth International Joint Conference on Artificial Intelligence, Vol. 2, 845, August 1979. Metcalfe, R.M. and Boggs, D.R., Ethernet: Distributed Packet Switching for Local Computer Networks, Comm. ACM, Vol. 19, No. 7 (July 1976). Shoch, J.F. and Hupp, J.A., Performance of an Ethernet Local Network -- A Preliminary Report, Proceedings of the Local Area Communications Network Symposium, Boston,May 1979. E. A. Feigenbaum 90 Privileged Communication 12. 13. 14. 15, 16. 17. 18. 19, 20. 21. 22. 23. 24. Literature Cited Taft, E.A., Implementation of PUP in TENEX, Internal XEROX PARC memorandum, June 1978. 9. Boggs, D.R., Shoch, J.F., Taft, E.A., and Metcalfe, R.M., PUP: Internetwork Architecture, XEROX PARC report CSL-79-10, July 1 An 97 Wilcox, C. R., Jirak, G. A., and Dageforde, M. L., MAINSAIL - Language Manual, Stanford University Computer Science Report STAN-CS-80-791 (1980). Wilcox, C. R., Jirak, G. A., and Dageforde, M. L., MAINSAIL - Implementation Overview, Stanford University Computer Science Report STAN-CS~80-792 (1980). Mead, C. and Conway, L., Introduction to VLSI Systems, Addison- Wesley Publishing Co. (1980). Rosen, B., PERQ: A Commercially Available Personal Scientific Computer, COMPCON 1980. Thacker, C. P., McCreight, E. M., Lampson, B. W., Sproull, R. F Boggs, D. R., ALTO: A Personal Computer, Computer Structures: Readings and Examples (Siewiorek, Bell, and Newell, eds.), 1979. -, and McDaniel, The Dorado: A Compact High-Performance Personal Computer for Computer Scientists, COMPCON 1980. Greenblatt, R., MIT's LISP Machine, COMPCON 1980. Ward, S. and Terman, C., An Approach to Personal Computing, COMPCON 1980. Lenat, D., “AM: An Artificial Intelligence Approach to Discovery in Mathematics as Heuristic Search, Ph.D. Dissertation, Stanford University, July 1976. Stefik, M.J., Planning With Constraints, Ph.D. dissertation, Stanford University, January 1980. Friedland, P.£., Knowledge-Based Experiment Design In Molecular Genetics, Ph.D. dissertation, Stanford University, October 1979. Privileged Communication 91 E. A. Feigenbaum Literature Cited 25. Sacerdoti, £.D., Problem Solving Tactics, Invited Lecture, Proceedings of the Sixth International Joint Conference on Artificial Intelligence, IJCAI-79. Available from Computer Science Dept., Stanford University, August, 1979. 26. Scott, A.C., Clancey, W., Davis, R., and Shortliffe, E.H.,Explanation Capabilities of Knowledge-Based Production Systems, American Journal of Computational Linguistics, Microfiche 62, Knowledge-Based Consultation Systems, 1977. E. A. Feigenbaum 92 Privileged Communication Biographical Sketches The following are biographical sketches for all professional personnel contributing to the SUMEX-AIM resource project. These do not include sketches for any of the individual collaborating project investigators. E. A. Feigenbaum 94 Privileged Communication SECTION 11 — PRIVILEGED COMMUNICATION BIOGRAPHICAL SKETCH (Give the following information for all professional personnel listed on page 3, beginning with the Principal Investigator. Use continuation pages and follow the same general format for each person, } NAME ACHENBACH, Michael W. TITLE System Programmer BIRTHDATE (Ma, Day, Yr.) August 2, 1952 PLACE OF BIRTH (City, State, Country} Los Angeles, California, U.S.A. PRESENT NATIONALITY (f/f non-U.& citizen, indicate kind of visa and expiration date) U.S. Citizen SEX £0 Mate C Female EDUCATION (8egin with baccalaureate training and include postdoctoral) YEAR IENTIFIC INSTITUTION AND LOCATION DEGREE CONFE MRED Soe Stanford University B.S. , 1974 Physics Stanford University M.A. 1975 Education HONORS MAJOR RESEARCH INTEREST Network communications, ‘Small machines RESEARCH SUPPORT (See instructions} ROLE IN PROPOSED PROJECT System Programmer RESEARCH AND/OR PROFESSIONAL EXPERIENCE (Starting with present position, Jist training and experience relevant to area of project. List all or most representative publications, Do not exceed 3 pages for aech individual.) 1978 - present System Programmer, SUMEX Computer Project, Department of Genetics, Stanford University School of Medicine 1975 - 1978 Scientific Programmer, Instrumentation Research Laboratories, Department of Genetics, Stanford University School of Medicine 1975 Scientific Programmer, Institute for Mathematical Studies in the Social Sciences, Stanford University PUBLICATIONS Smith, D.H., Achenbach, M., Yeager, W.J., Anderson, P.J., Fitch, W.L., and Rindfleisch, T.C.: Quantitative Comparison Gas Chromatographic/ Mass Spectrometric Profiles of Complex Mixtures, Anal. Chem., 49, 1623, 1977. NIH 398 (FORMERLY PHS 396) Rev. 1/73 E.A. Feigenbaum #U.S. GOVERNMENT PRINTING OFFICE: 1977—241-161:3024 95 Privileged Communication SECTION !! — PRIVILEGED COMMUNICATION BIOGRAPHICAL SKETCH (Give the following information for all professional personnel listed on page 3, beginning with the Principal Investigetor. Use continuation pages and follow the same general format for each person} NAME TITLE BIRTHDATE (Ma, Day, Yr.) AIELLO, Nelleke T.G.K. Scientific Programmer March 21, 1949 PLACE OF BIRTH (City, State, Country) PRESENT NATIONALITY {if non-US. citizen, SEX indicate kind of visa and expiration date) Amsterdam, The Netherlands U.S. Citizen ClMeie ffitFemale EDUCATION (Begin with baccalaureate training and include postdoctoral) YEAR SCIENTIFIC INSTITUTION AND LOCATION DEGREE CONFERRED FIELD University of California, Santa Cruz B.A. 1971 Mathematics University of California, Santa Cruz B.A. 1971 Information and Computer - Science University of Utah, Salt Lake City M.S. 1972 Computer Science HONORS Departmental Honors, Information and Computer Science, University of California -Grown College Honors, University of California “MAIOR RESEARCH INTEREST ROLE IN PROPOSED PROJECT ~Butiding intelligent systems “Hrowtedge engineering “RESEARCH SUPPORT (See instructions} - Scientific Programmer RESEARCH AND/OR PROFESSIONAL EXPERIENCE (Starting with present position, jist training and experience relevant to ares of project List all or most representative publications. Do not exceed 3 pages for sech individuel.} 1977 - present Scientific Programmer, Heuristic Programming Project, Computer Science Department, Stanford University 1972 - 1977 Programmer, Bolt, Beranek and Newman, Inc. 1973 - 1975 Teaching Assistant, Structured Programming, University of Summers California Extension Summer 1972 Teaching Assistant, Compiler Writing, University of California Extension 1971 Programmer, Shell Benelux Centre, De Hage, The Netherlands PUBLICATIONS (See continuation page) W1H 398 (FORMERLY PHS 398) Rev. 1/73 E. A. Feigenbaum 96 US. GOVERNMENT PRINTING OFFICE: 1977-—241-161:3024 Privileged Communication BIOGRAPHICAL SKETCH — AIELLO, Nelleke T.G.K. PUBLICATIONS 1. Aiello, N: An Analysis of Notations for Music Applicable to the Digital Control of Electronic Musical Instruments. Masters Thesis, University of Utah, 1972. Collins, A.M., Warnock, E.L., Aiello, N., and Miller, M.L.: Reasoning from Incomplete Knowledge, In D. Bobrow and A.M. Collins (Eds.) REPRESENTATION AND UNDERSTANDING STUDIES IN COGNITIVE SCIENCE, New York, Academic Press, Inc., 1975. Nii, H.P. and Aiello, N.: AGE (Attempt to Generalize): Profile of the AGE-O System. Stanford Heuristic Programming Project Memo HPP-78-5 (Working Paper), June 1978. Nii, H.P. and Aiello, N.: AGE: A knowledge-based program for building knowledge-based programs. Proc. of IJCAI-6, pp 645-655, 1979, Aiello, N., Nii, H.P. and White, W.C.: The Joy of AGE-ing: An Introduction to AGE-1l. Stanford Heuristic Programming Project Memo (work in progress), May 1980. E. A. Feigenbaum 97 Privileged Communication SECTION II — PRIVILEGED COMMUNICATION BIOGRAPHICAL SKETCH {Give the following information for all professional personnel listed on page 3, beginning with the Principal Investigator. Use continuation pages and follow the same general format for ech person) BIRTHDATE (Ma,, Day, Yr.) NAME TITLE Adjunct Professor July 7, 1940 BUCHANAN, Bruce G. . Computer Science PLACE OF BIRTH (City, State, Country} PRESENT NATIONALITY (/f non-U.S. citizen, SEX indicate kind of visa and expiration date) St. Louis, Missouri, U.S.A. U.S. Citizen Gd Mate (2 Femate EDUCATION (8egin with baccalaureate training and include postdoctoral) YEAR SCIENTIFIC INSTITUTION AND LOCATION DEGREE CONFERRED FIELD Ohio Wesleyan University B.A. 1961 Mathematics Michigan State University M.A. 1966 Philosophy Michigan State University Ph.D. 1966 Philosophy HONORS (see continuation page) MAJOR RESEARCH INTEREST Axtificial Intelligence RESEARCH SUPPORT (See instructions) tsee continuation page) RESEARCH AND/OR PROFESSIONAL E XPERIENCE (Starting with present position, list training and experience refevant to area of project List all ROLE IN PROPOSED PROJECT Or most representative publications, Do not exceed 3 pages for each individual.] 1976 - present 1972 - 1976 Stanford University 1966 ~ 1971 Stanford University PUBLICATIONS ( see continuation page) Technical Director of Core Research Adjunct Professor, Computer Science Department, Stanford University Research Computer Scientist, Computer Science Department, Research Associate, Artificial Intelligence Project, WiH 398 (FORMERLY PHS 998) Rav. 1/73 E. A. Feigenbaum aus GOVERNMENT PRINTING OFFICE. 1977—241-161:3024 98 Privileged Communication BIOGRAPHICAL SKETCH — BUCHANAN, Bruce G. RECENT HONORS Editorial Board, Artificial Intelligence: An International Journal American Association for Artificial Intelligence — Organizing Committee, Program Committee and Membership Chairman Chairman of Program Committee, IJCAI-79 (International Joint Conference on Artificial Intelligence, Tokyo, 1979) Invited Colloquium Speaker: University of Maryland Carnegie-Mellon University Rutgers University University of California at Berkeley Michigan State University Invited Speaker: AISB Annual Conference (Amsterdam, July 1980) Workshop on the Logic of Discovery and Diagnostics in Medicine (Pittsburgh, October 1978) Douglass College Seminars for Faculty (Rutgers University, 1978) Workshop on Pattern Directed Inference Systems (Honolulu, 1977) Recipient, National Institutes of Health Career Development Award (1971-1976) MEMBERSHIPS American Association for Artificial Intelligence (AAAT) Cognitive Science Society Association for Computing Machinery (ACM), SIGART Philosophy of Science Association RESEARCH SUPPORT Funding Current Project % of Gr ant Grant No. fitle of Project Year Period Effort Agency 1P01 LM Research Program : $ 99,484 $497,420 10 NLM 03395-01 Biomedical Knowledge (7/79-6/80) (7/79-6/84) Repr esentation MCS-7 903 75 3 Knowledge-Based $ 73,659 $ 73,659 10 NSF Consultation (7/79-6/80) (7/79-6/80 Systems + 6 months) NOOO14-79- Exploration of $396,325 $396, 325 10 ONR C~0 302 Tutoring and (3/79-3/82) (3/79-3/82) Prob. Solv. Strategies in Intelligent Com puter~Aid ed Instruction MDA 903-80- jeur istic $496,256 $1,613,588 40 ARPA C-0107 Programming (10/79-9/80) (10/79-9/82) Project 5R24 RROO612- Resource-Related $221,255 $641,419 5 NIH 10 Research — (5/80-4/81) (5/80-4/83) Computers and Chemistry E. A. Feigenbaum 99 Privileged Communication BIOGRAPHICAL SKETCH — BUCHANAN, Bruce G. Selected Publications Edward H. Shortliffe, Bruce G. Euchenen, and Edward A. Feigenbeum, “Knowledge Engineering for Medical Decision Making: A Review of Computer-Based Clinical Decision Aids," Proceedings of the IEEE, September, 1679. Bruce €. Buchanen, “Issues of Representation in Conveying the Scope end Limitations of Intelligent Assistant Programs." In J.E. Hayes, D. Michie, and L.I. Mikulich (cds.), Mechine Intelligence 9: Machine expertise and the humen interface. New York: Jcehn Wiley, 1o79. Eruce G. PBuchenen and Edward A. Feigenbaum, "DENDRAL and Meta-DENDRAL: Their Applications Dimension," Artificial Intelligence 11, 5, 1978. Bruce G. Buchanén, Tom M. Mitchell, Reid G. Smith and C. Richard Johnson, Dr., "Models of Learning Systems," in J. Belzer (ed.), Encyclopedia of Computer Sciences end Technology, New York: Marcel Dekker, Inc., 1978, Vol ll. Randall Pavis end Pruce G. Buchanén, “Meta-Level Knowledge: Overview end Applications," Proceedings of the Fifth IJCAI,1,926, August 1977. Bruce G. Buchanan end Tom Mitchell. "“Model-Directed Learning of Production Rules," in D.A. Waterman and F. Hayes-Roth (eds.), Pattern Directed Inference Systems, New York: Academic Press, 1978. Bruce G. Buchanan énd Dennis Smith, “Computer Assisted Chemical Reasoning," in E.V. Ludena, N.H. Sébelli and A.C. Wahl (eds.), Computers in Chemical Education and Research, New York: Plenum Press, 1977, p. 461. Randall Devis, Bruce Buchénan, Edwerd Shortliffe, “Production Rules es a Representation of a Knowledge-Based Consultation Program," in Artificial Intelligence, &, 1, February 1¢77. Bruce G. Puchenén, D.H. Smith, W.C. White, R.J. Gritter, E. Feigenbeun, J. Lederberg, and C. Djerassi, “Application of Artificial Intelligence for Chemical Inference XXII. Automatic Rule Formation in Mass Spectronomy by Means of the Meta-DENDRAL Program," Journal of the American Chemical Society, 8, 6168, 1976. E. A. Feigenbaum 100 Privileged Communication