Paper presented at the IPN Symposium 7: Formalized Theories of Thinking and Learning and Their Implications for Science Instruction.

In August of 1980, Stanford University was the site of the annual workshop on artificial intelligence in medicine (AIM). This specialized area of medical computer science research had been born almost ten years earlier with the near-simultaneous development of AIM research groups at Massachusetts Institute of Technology (in collaboration with physicians from the Tufts New England Medical Center), the University of Pittsburgh, Rutgers University, and Stanford. These small groups of computer scientists working in the field were drawn together naturally by their common interests and by the establishment of the SUMEX-AIM network (Stanford University Medical EXperimental Computer for Artificial Intelligence in Medicine). This computing resource was established by the Biotechnology Resources Program of the NIH in 1974 and consisted of a pair of computers, one at Rutgers and one at Stanford, linked by a communications network. The funding for SUMEX-AIM not only provided computing power for researchers exploring the potential of artificial intelligence techniques in medicine but also established a series of annual workshops so that the investigators could gather to,share their insight, results, and ideas regarding approaches to the difficulties they encountered.

How do practicing physicians make clinical decisions? What techniques can we use in the computer to produce programs that exhibit medical expertise? Our interest in these questions is motivated by our desire: 1. to provide (by computer) expert medical consultation to general practitioners or paramedical personnel in communities where such consultation is normally unavailable; 2. to come to understand the reasoning processes of expert doctors so that we may improve the teaching of their skills to medical students; and 3. to advance the techniques of artificial intelligence, especially as applied to medicine (AIM), to support our other goals. In other publications, we have described research by our group on programs to take the history of the present illness of a patient with renal disease (Pauker and Gorry, 1976; Szolovits and Pauker, 1976) and to advise the physician in the administration of the drug digitalis to patients with heart disease (Gorry et al., 1978; Silverman, 1975; Swartout, 1977).

JANICE S. AIKINS Dr. Aikins received her Ph.D. in computer science from Stanford University in 1980. She is currently a research computer scientist at IBM's Palo Alto Scientific Center. She specializes in designing systems with an emphasis on the explicit representation of control knowledge in expert systems. ROBERT L. BLUM Dr. Blum received his M.D. from the University of California Medical School at San Francisco in 1973. From 1973 to 1976 he did an internship and residency in the Department of Internal Medicine at the Kaiser Foundation Hospital in Oakland, California, where he was chief resident in 1976.

To test the program during its development, MyeTs and his students would select especially difficult cases for considemtion, often ones drawn fTOm published clinical pathological confeTences in medical journals. AfteT seveTal years of testing and rf:finement of the knowledge base, the study outlined in the following chapteT was peTformed. To document the strengths and weaknesses of the pTogmm, the gTOUP performed a systematic evaluation of the pTOgTam's capabilities.

In the present paper, a general approach to structuring medical knowledge for computer-aided diagnosis and therapy is presented. We have developed a representation that models disease processes as a causal-associational network (CASNET). This model-based method has been used successfully in designing a consultation program for the diagnosis and long-term treatment of the glaucomas. The consultation program uses a set of general decision-making strategies in conjunction with a class of causal-associational models (Kulikowski and Weiss, 1971; Weiss, 1974). In this paper, examples will be given from a CASNET model of glaucoma.

The format is identical to that of Figure 6-2.

System Overview 101 A second constraint was the need to design the program to accommodate a large and changing body o.f technical knowledge. It has become clear that large amounts of task-specific knowledge are required for high performance and that this knowledge base is subject to significant changes over time (Buchanan and Lederberg, 1971; Green et al., 1974). Our choice of a production rule representation was significantly influenced by such features of the knowledge base. A third demand was for a system capable of handling an interactive dialogue and one that was not a "black box." This meant that it had to be capable of supplying coherent explanations of its results, rather than simply pririting a collection of orders to the user.

By providing an environment for encoding knowledge, editing the evolving knowledge base, and testing programs, these systems provide techniques and tools that promise to be very versatile in helping to design new medical expert systems. While the earlier chapters in this volume provide motivation for applying artificial intelligence techniques to medicine, comparing the methods to those of traditional algorithmic programming and statistics, in this paper Kulikowski presents the knowledge-based perspective as a whole. This serves as a prelude to detailed discussions of particular consultation systems (Chapters 5, 6, 7, and 8) and to Szolovits and Pauker's analysis of medical reasoning in the context of these programs (Chapter 9).

The ultimate systems art' still probably many decades away, but existing techniques help define a subset of problems with which we are alread,'V prepared to deal.