Adaptation-Based Explanation: Explanations as Cases

Abstract

This chapter discusses a case-based approach to the problem of learning new explanations. Adaptation-based explainer (ABE) applies a case-based approach to the problem of constructing explanations. It brings together a few insights of explanation-based learning (EBL) and case-based reasoning (CBR), and addresses questions about each framework with solutions suggested by the other. ABE cuts down the amount of work needed to produce explanations by reusing explanations that have worked in the past. By relying heavily on past explanations to build new ones and by using the causal knowledge represented in those explanations to adapt them to new situations, this approach brings together a few important insights in explanation-based learning and case-based reasoning. EBL systems bring causal knowledge to bear on a problem via an explanation of why an instance of a class actually is an instance of that class. The burning question for EBL is regarding the construction of the explanations needed, without having the cost of building those explanations outweigh the savings that accrues from having them. CBR is based on appealing to memory of past cases as a more efficient and reliable way to reason about new situations than relying merely on abstract rules. The biggest challenge within CBR is regarding the application of the knowledge contained in a case to a new situation that is a close but not an exact match to the original case.