Using Decision Theory to Justify Heuristics

Proceedings of the AAAI Conference on Artificial Intelligence, 5

Volume

Science

Search

We present a method for using decision theory to evaluate the merit of individual situation -- action heuristics. The design of a decision-theoretic approach to the analysis of heuristics is illustrated in the context of a rule from the MYCIN system. Using calculations and plots generated by an automated decision making tool, decision-theoretic insights are shown that are of practical use to the knowledge engineer. The relevance of this approach to previous discussions of heuristics is discussed. We suggest that a synthesis of artificial intelligence and decision theory will enhance the ability of expert systems to provide justifications for their decisions, and may increase the problem solving domains in which expert systems can be used.