A methodology for constructing expert systems

Abstract

AbstractConstruction of Expert Systems has so far been seen as a craft or an art, not a science. This paper attempts to improve this situation by deducing some aspects of a methodology for constructing Expert Systems from a model of an expert's expertise. It suggests that the traditional methods, based on extracting problem‐solving rules from an expert and encoding them directly into a suitable knowledge representation have certain disadvantages, while attempting to extract a causal model from the expert overcomes many of these. The simple Plausible Inference Expert System shell, of which several are now commercially available, is often very suited to the construction of causal models, and some practical issues concerning the construction of causal models are discussed.The suggested methodology has been found to work in several Expert System projects in I.C.I., though it is still being developed in the light of experience. In particular, the importance of the top‐down design of an Expert System is stressed, though it is not yet based on theoretical considerations.