Reference: Shortliffe, E. H. Medical Expert Systems Research at Stanford University. September, 1986.
Abstract: The Knowledge Systems Laboratory (KSL) at Stanford University is now midway through its second decade of research into biomedical applications of artificial intelligence. The KSL was known as the DENDRAL project in the late 1960's, and as the Heuristic Programming Project during the 1970's and early 1980's. It is a research laboratory dedicated to the notion that the most challenging fundamental issues in artificial intelligence are often discovered and effectively addressed in the setting applications research for compex problem soving tasks in biology and engineering. Clinical medicine has been a particularly fruitful area for such basic research. Biomedical AI research in the KSL has been supported in large part by the SUMEX Computer Project, a federally funded computing resource physically located at Stanford but shared via networks by medical computer scientists at several other institutions. Four experimental systems are generally regarded as having started the research field of artificial intelligence in medicine in the early 1970's [25, 3]. These were: *MYCIN, a program developed at Stanford to advise physicians on antimicrobial selection for patients with bacteremia or meningitis [23, 2] *the Present Illness Program (PIP), a system from MIT and Tufts-New England Medical Center that gathered data and generated hypotheses about disease processes in patients with renal disease [20] *INTERNIST-1, a large system from the University of Pittsburgh to assist with diagnosis of complex problems in general internal medicine [17] *CASNET, an ophthalmology advisor from Rutgers University designed to assess disease state and to recommend management for patients with glaucoma [27] All four drew on AI techniques, emphasizing the encoding of large amounts of specialized medical knowledge acquired from the clinical literature and from expert collaborators. Each was influenced by earlier AI work on generalized problem solving techniques, and two of the systems (PIP and Internist-1) explicitly modeled hypothetico-deductive behavior [6, 13], the familiar process by which physicians formulate tentative hypotheses rapidly after obtaining the first few pieces of information about a patient and then let those hypoteses (typically a differential diagnosis) guide further data collection and problem solving. The initial explorations into medical AI during the 1970's have given way in the current decade to more focussed research on central issues that characterize work not only in biomedicine but in all applications areas in which knowledge-based reasoning techniques are being applied. In this presentation, I will summarize some of these key issues and briefly describe a few of the biomedical research projects at Stanford's KSL that are addressing these central problems. The issues range from pragmatic to theoretical, and will be discussed the the following order: 1. The logistics of human-computer interaction 2. Representing and reasoning with temporal relationships 3. Acquiring and encoding knowledge 4. Model-based reasoning 5. Reasoning with constraints 6. Synthesizing AI and Bayesian statistics 7. Reasoning from published analyses of empirical data The first four of these topics are all being studied in different subprojects of a large effort know as ONCOCIN. I will therefore begin with a brief description of ONCOCIN and then proceed to a descussion of each of the seven research areas mentioned.
Notes: Working Paper 13 pages.