Where Are the Explanations in Causal AI? Integrating Pearl's Definitory Account of Causality with Causal Explanations

Abstract

In artificial intelligence, Pearl's influential work conceives causality as a feature allowing for reasoning about sophisticated queries such as the consequences of actions, while philosophical accounts primarily understand causality as providing explanations for why things happen. This contribution proposes the framework of definitory causal models to merge these two perspectives. Here, Datalog programs represent Boolean causal relationships that can be composed into causal explanations.