Case-Based Knowledge Representation

The representation of knowledge in terms of rules is fraught with theoretical problems, such as the justification of induction, the "right" way to do it, and the revision of knowledg ein face of contradictions. In this paper we argue that these problems, and especiallyt the inconsistency of "knowledge," are partly due to the fact that we pretend to know what in fact cannot be known. Rather than coping with thep roblems that explicit induction raises, we suggest to avoid it. Instead of formulating rules which we supposedly "know," we may make do with the knowledge of actual cases from our experience. Starting from this viewpoint, we continue tooderive Case-Based Decision Theory (CBDT), and propose it as a less ambitious, yet less problematic theory of knowledge representation. CBDT deals with decision making under uncertainty, and can be viewed as performing implicit induction, that is, as using past experience to make decisions, without resorting to the explicit formulation of rules. We discuss two levels on which implicit induction takes place, and the corresponding two roles that "rules" may have in case-based decision making. We also discuss the process of learning and the concept of "expertise" as they are reflected in our model.