Causal reasoning in graphical models

Abstract

This paper presents the problem of the identification of the causal relations that agents, in front of a sequence of reported events, may attribute on the basis of their beliefs on the course of things and available pieces of information. In particular, we focus on graphical models exploiting the idea of "intervention", initially proposed in the probability framework by Pearl, and developed in the more qualitative setting of the theory of possibilities within the french national project called MICRAC. We show that interventions, which are very useful for representing causal relations between events, can be naturally viewed as a belief change process. This paper also provides an overview of main compact representation formats, and their associated inference tools, that exist in a possibility theory framework.