A justification-based theory of explanation
Eunok Paek · 1992
A notion of explanation has been a fundamental concept for various artificial intelligence problems. Despite its fundamental importance, little has been done to provide a formal theory of explanation. Here we present a formal and unifying theory of explanation which is justification-based. Given a background theory and an observation, we will define when a proposition explains the observation with respect to the background theory. Our approach states that the addition of an explanation to the background theory should provide a better justification for the observation in question than the background theory alone. Our approach also provides various ways to define preference criteria on explanations. As we apply our theory of explanation to a particular domain, domain-specific characteristics allow us to further refine and develop the theory; we will explore the role of this domain-specific information in the areas of causal reasoning and plan recognition. In both areas, we will focus on the phenomenon called causal asymmetry which results because two causes of an observation interact differently than two consequences of a common cause. The justification-based approach provides a better qualitative account of causal asymmetry than previously proposed formalism and also demonstrates its generality by showing that it produces the desired results in plan recognition.