A General Framework for Ampliative Inference Patterns

Abstract

Non trivial reasoning from contradictory premises is being acknowledged as one of the most important features in intelligent systems. Expert systems, planners and schedulers, and diagnosers, are almost always faced to potentially fallacious information, errors, uncertainty, and difference of opinions. In thesecases, we expect that the reasoning systems will not collapse. Instead, the rational expected behavior is to isolate the source of contradiction. Several systems for reasoning from contradictory premises have been advanced, usually within acontext of strict, monotonic knowledge. In this work we investigate how defeasible knowledge can be also handled in these systems. The key idea is to represent pieces of defeasible knowledge ordered within anepistemic importance relation. A semantic characterization is provided, and a sound and complete procedure to compute conclusions is also given. Then, we show how nonmonotonic reasoning and other patterns of ampliative inference like abduction and induction can be adequately recast within the general pattern of reasoning from contradiction. We discuss some applications, in particular, a brief formalization of scientific research programmes.