A Consistency-Based Approach to Knowledge Base Refinement

In this work, we outline an approach to incrementally building knowledge-based systems based on tightly controlling the order of evaluation of the knowledge components of the system. The order of evaluation is based on two relations, sequence and correction, that correspond to two kinds of changes that an expert may wish to make to a knowledge base. A consistency notion based on past performance is also introduced. Knowledge base refinement is structured so that newly added knowledge has one of these relations with existing knowledge components in the system and the consistency condition is preserved. We further propose that the knowledge components added might be any knowledge-based systems or programs rather than rules. This proposal is a generalisation of the Ripple-Down Rule incremental approach to building knowledge-based systems.