Inferring causal explanations

Abstract

A popular approach to explanations amounts to backward chaining over logical implications encoding causal links. However, the resulting explanations are often unsatisfactory from a common-sense point of view. We dene a framework allowing us to distinguish causal implication from mere logical implication. Causal explanations are then deduced through two inference schemes so that explaining is in some way "less than implying" and "more than implying". Finally, we show how our approach applies to diagnostics.