Peter E. Hart
Ricoh Silicon Valley, Inc.
The intellectual, research funding, and commercial history of artificial intelligence includes both boom and bust periods in the decades since its inception. During much of its history, AI as a whole has been widely-- though inaccurately-- identified with expert systems, no doubt because expert systems have been the most widely-commercialized AI technology.
I will describe the life and times of three expert systems for which I bear some responsibility: Prospector in the 1970's, Syntel in the 1980's, and Fixit in the 1990's. These systems, which to some extent can be regarded as surrogates for the AI activities of their day, experienced very different fates. Respectively, they were a resounding technical success, a large-scale commercial failure, and a modest technical and commercial success.
Truth in Abstracts disclaimer: The emphasis will not be on the technical details of these systems, nor will it be on a broad scholarly review of the history and current state of AI. Rather, this will be a purely personal view of the conditions surrounding the failure and success of AI in the real world, and the lessons that might be drawn from them.
Peter E. Hart, the Chairman and President of Ricoh Silicon Valley, Inc., has done AI research and has managed corporate R & D organizations since completing his Ph.D. at Stanford in 1966. His AI research outside of expert systems has contributed to a theoretical understanding of the nearest-neighbor classifier, to the development and analysis of the A* graph-searching algorithm, and to the invention of the modern form of the Hough transform for image analysis. He has co-founded two companies, directed three research centers, and also is the co-author of what might be the oldest AI-related textbook still in print (Duda & Hart, 1973).