Recognizing and Revising Unconvincing Explanations

Abstract

Expert system explanations sometimes fail to satisfy their users. The standard assumption is that these failures arise because the user did not completely understand the system's explanation. However, users may also be left unsatisfied because they are unconvinced by the explanation provided. Advice‐giving systems should therefore eventually be extended to recognize when and why an explanation has failed to convince the user and to elaborate or revise explanations to make them more convincing. This paper provides an initial model of this process for systems that give advice about choosing between multiple alternatives for accomplishing a particular task. In particular, we discuss the model underlying a prototype system to advise novice computer users about how best to accomplish tasks on the UNIX operating system.