HANDLING INCOMPLETE KNOWLEDGE IN ARTIFICIAL-INTELLIGENCE

In this paper we first discuss the important role of nonmonotonic reasoning for Artificial Intelligence. After presenting some simple forms of nonmonotonicity as they arise in various well-known AI systems we present in Section 2 some of the most important existing nonmonotonic logics: McCarthy's circumscription, Moore's autoepistemic logic, and Reiter's default logic. Section 3 examines an approach in which default reasoning is reduced to reasoning in the presence of inconsistent information. The approach is based on the notion of preferred maximal consistent subsets. It is shown that these preferred subsets can be defined in such a way that it is possibly to represent priorities between defaults adequately. Section 4 briefly discusses the problem of implementing nonmonotonic systems.