An expert system is designed to solve hard problems usually requiring considerable expertise. Users are more likely to accept such a system if they can ask for and receive explanations of the system's conclusion. Our contention is that improved explanations can be generated by a knowledge-based explanation system (loosely coupled to the expert system) that actively searches for support of the expert system's conclusion. Current methods of explanation derive their informtion almost entirely from a trace of the rules used during the expert system's problem-solving process. Additional information, when present at all, is introduced through justifications of these rules. We propose to build a system that treats the generation of explanations as a problem-solving activity largely distinct from the processes initially used to solve a problem. This system will move expert system explanation from a passive (unintelligent) task of presenting rule traces to an active (intelligent) task of searching for explanations in support of the conclusion of the problem-solving expert system. At the center of our knowledge-based explanation techniques (the knowledge of how to explain). This knowledge will influence not only the way information is presented to the user, but also what information is presented tothe user. With this ability to determine both the content and form of the explanation, our system will have increased flexibility over previous explanation systems. This increased flexibility will be used to support the creation of explanations that more naturally support the expert system's conclusion.
