Investigations into a Theory of Knowledge Base Revision

Abstract

A fundamental problem in knowledge representation is how to revise knowledge when new, contradictory information is obtained. This paper formulates some desirable principles of knowledge revision, and investigates a new theory of knowledge revision that realizes these principles. This theory of revision can be explained at the knowledge level, in purely model-theoretic terms. A syntactic characterization of the proposed approach is also presented. We illustrate its application through examples and compare it with several other approaches. 1 Introduction At the core of very many AI applications built in the past decade is a knowledge base --- a system that maintains knowledge about the domain of interest. Knowledge bases need to be revised when new information is obtained. In many instances, this revision contradicts previous knowledge, so some previous beliefs must be abandoned in order to maintain consistency. As argued in [Ginsberg, 1986], such situations arise in diverse areas such...