procascamc00006-0199

A PROGRESS REPORT ON ARTIFICIAL INTELLIGENCE: HOSPITAL APPLICATIONS AN D A REVIEW OF T H E ARTIFICIAL INTELLIGENCE MARKETPLACE Lawrence M . Brenkus, M.D. Arthur D , Little, Inc. Acorn Park Cambridge, Massachusetts 02140 ABSTRACT Artificial intelligence applications ar e finally beginning t o move from t h e university research laboratory into commercial use. Before t h e en d of the century, this n e w computer technology will have profound effects on ou r work, economy, and lives. At present, rela- tively f ew products have appeared i n th e hospital, bu t we ca n anticipate significant product offerings i n i n- strumentation a nd affecting hospital administration within 5 years, Knowledge-based expert systems represent a n e w software technology which permits programs to simulate human expertise through th e us e of a large knowledge base and reasoning methods. These programs have recently become a topic of lively interest because of their increas- i n g capabilities, a n d their recent intro- duction into t he commercial marketplace as real-world products. Th e most famous program of this type operating i n th e medical domain i s probably INTERNIST, a diagnostic consulting program developed at University of Pittsburgh. A test of i ts performance w a s published recently i n th e N e w England Journal of Medicine; th e performed approximately as well as t h e human physicians i n solving difficult C P C cases. At present, ou r technology f or building expert systems is still limited. Each program i s a custom effort; no fully adequate off-the-shelf system building tools exist yet, Furthermoref they per- form adequately only i n a narrow domain of expertise, unlike a human expert w h o can often b e a generalist. T he programs are also unable t o learn an d require human intervention to a d d n e w rules to refine th e system. Thus, a fully developed "in- dustrial strength" expert system i s quite expensive - - commercially useful products have required about 5 man-years of effort a n d about a million dollars expenditure. (Prototype development i s much less costly, b u t these figures are typical to develop a sizeable program with true human expertise i n a field.) Even with these limitations there a r e many problems which justify t he u s e of expert systems technology because of th e large savings which c a n be realized. Typically these high-value applications a re used within a company to save money on i ts i n- ternal operations. Examples already d e- Velop include a n expert underwriter f or a n insurance firm, a sales adviser w h o aids i n placing orders correctly, a shop- floor scheduling expert, a nd a n oi l wel l instrument interpreter. Each of these applications involve t he "base busine.s" of th e developing company. They are e x - pected to pay back their development costs within th e first year of use. While many expert systems like INTERNIST have been developed by university r e - searchers, none of them have ever been commercialized or have even moved success- fully f rom their original research site. A simple explanation for thi s i s that th e "doc i n a box" represents a n ideal r e- search project, but does n ot offer th e direct cost savings required to be a suc- cessful commercial expert system. These academic prototypes were designed t o be used b y physicians, but there is little evidence that doctors even want a stand- alone diagnostic consulting system. Older physicians ar e often threatened b y this 1 8 2 CH2090-9/84/0000/0182$01.00 © 1984 IEEE

procascamc00006-0199

Documents