Interactive diagnosis and repair of decision-theoretic models

Abstract

The authors present an intuitive, interactive method for diagnosing and repairing a formal decision-theoretic model, and describe the role of this method in interpretive value analysis (IVA), a general framework for modeling choices in expert systems that is at once formal and transparent. IVA's interpretation concepts provide an internally consistent vocabulary based on value theory for talking about decisions. IVA's explanation component explains the operation-by-operation computation of a comparison of two alternatives, using interpretation concepts, operation explainers, and the difference-function topology. IVA's diagnosis-and-repair method uses this explanation component, along with modules for suggesting, repairing, and verifying the integrity of parameter values, to assist users with modifying models that deviate from their preferences. Although previous approaches to modeling choices in expert systems generally have sacrificed formal specification for transparent operation, this approach suggests that knowledge engineers can retain the benefits of decision theory's formal foundation without compromising intuitive explanation and knowledge acquisition.