In just one decade the field of expert systems, a subfield of artificial intelligence, has progressed from an intellectual endeavor pursued by researchers at universities to commercial systems used by major corporations. Expert systems have been developed by Digital Equipment Corporation to configure VAX computers (XCON). SRI International to perform geological exploration (PROSPECTOR), the U.S. Navy to train personnel in the operation of steam propulsion plants (STEAMER), and Intellicorp to help molecular biologists perform nucleotide sequence analysis (SEQ), just to name a few. The hopes and dreams of everyone have been raised by the performance of these and other systems and by the future potential of what expert systems could do. But are these aspirations to be realized? Will expert systems prove to be “white knights” assisting us In solving our problems or will they be “wolves in sheep's clothing” tormenting us as we strive towards our goal?
This paper examines issues of concern in the development of expert systems: characteristics of the problem domain being attacked, attributes of the experts on which the system will be based, and requirements of the development process. Examples from the medical and automotive domains are discussed to illustrate the potential problems inhibiting the development, usefulness, and effectiveness of expert systems. Finally, some research directions to solve these problems are presented.