KNOWLEDGE BASE REFINEMENT AND THEORY REVISION

Abstract

This chapter reviews a few intuitions, beliefs, and speculations concerning the nature of theory revision, its relation to knowledge base refinement, and related topics. The terms knowledge base and knowledge-base refinement seem, for better or for worse, inextricably intertwined with the expert systems movement. The terms theory and theory revision seem to evoke an expectation of something having to do with the sciences or at least with something deeper than the content of a typical knowledge base. While any theory must include a declarative component, that is, the axioms of the theory, it might be appropriate to view a theory as also containing, like an expert system, a procedural component which, among other things, enables someone who learns the theory to use it in problem solving. The procedural component is easily identified in expert systems as these are artificial systems designed according to a certain paradigm. Bridge laws can be thought of as go betweens from the theoretical terms of a theory to either observational terms, or perhaps, the theoretical terms of another theory. A good deal of scientific inquiry and technological progress involves the discovery and use of bridge laws. Given that bridge laws are a part of what makes theories useful in problem solving, they are an ideal candidate for inclusion in the procedural component of a theory. There is a mapping from the terms of the old theory to the new theory such that the laws of the old theory are valid statements in the new theory; this essentially is theory revision. There are basically three issues or processes of importance in knowledge-base refinement: (1) localization, (2) refinement or revision generation, and (3) testing and selection of proposed revisions.