Numerical and symbolic approaches to uncertainty management in AI

摘要

Dealing with uncertainty is part of most intelligent behaviour and therefore techniques for managing uncertainty are a critical step in producing intelligent behaviour in machines. This paper discusses the concept of uncertainty and approaches that have been devised for its management in AI and expert systems. These are classified as quantitative (numeric) (Bayesian methods, Mycin's Certainty Factor model, the Dempster-Shafer theory of evidence and Fuzzy Set theory) or symbolic techniques (Nonmonotonic/Default Logics, Cohen's theory of Endorsements, and Fox's semantic approach). Each is discussed, illustrated, and assessed in relation to various criteria which illustrate the relative advantages and disadvantages of each technique. The discussion summarizes some of the criteria relevant to selecting the most appropriate uncertainty management technique for a particular application, emphasizes the differing functionality of the approaches, and outlines directions for future research. This includes combining qualitative and quantitative representations of information within the same application to facilitate different kinds of uncertainty management and functionality.