How Expert Should an Expert System Be?

Abstract

A computer system which aids computer engineers in fault diagnosis is described. The system, c alled CRIB ( Computer Retrieval Incidence Bank ) is shown to fit into the class of pattern-directed i nference systems. Emphasis is placed on the "bef o re" and "after" phases of system generation and it is shown why, to be called an expert system, these phases are important. The forms of knowledge used i n CRIB are shown to be adequate for diagnosis and yet possess little of the structural or functional knowledge of more advanced expert systems. Summaries are given of the three phases of implementation: elicitation, implementation of knowledge structures, validation and improvement. The idea o f an expert system as a "model of competence" i s mentioned and the transferrance of the system architecture to software diagnosis, using the same model, is described. There are short discussions of system performance and the nature of expert systems.