A conceptualization of preferences in non-monotonic proof theory

Abstract

Formalizing non-monotonic reasoning is a significant problem within artificial intelligence. A number of approaches have been proposed, but a clear understanding of the problem remains elusive. Given the diversity of proof theoretic approaches, we argue the need for frameworks for elucidating key concepts within non-monotonic reasoning. In this paper we consider the preferences, implicit and explicit, that can be seen in a disparate range of non-monotonic logics. In particular, we argue the case for an analysis based on Labelled Deductive Systems for viewing existing approaches to formalizing non-monotonic reasoning, and for identifying new approaches. For this we introduce the family of prioritized logics — each member being a defeasible logic defined in terms of labelled deduction -that forms the basis of a framework for viewing the nature and mechanization of non-monotonic reasoning.