A framework for explaining decision-theoretic advice

摘要

We present strategies for automatically explaining decisions based on Multiattribute Value Theory, the branch of decision theory that addresses tradeoffs among multiple competing objectives. Our explanations are at once empirically motivated and formally sound, retaining the advantages of both artificial-intelligence and decision-theoretic representations for modeling decisions. We demonstrate the explanation strategies with implemented examples in the domains of marketing, process control, and medicine. Although previous approaches to modeling decisions in expert systems often have sacrificed formal specification for transparent operation, our methodology suggests that knowledge engineers can retain the benefits of decision theory without compromising intuitive explanation.