The Correcting Machine: toward a conversational advisory system for UNIX

The author presents a model of the process of providing corrective explanatory responses to misconceptions made by novice UNIX users. To provide these responses, the advisor must be able to formulate justifications for their own beliefs, and to understand the belief justifications of the user. Both tasks require abstract knowledge about belief justification. This knowledge is represented in a collection of justification patterns that capture knowledge about possible ways to justify different types of plan-oriented beliefs. This approach differs from earlier efforts in two ways. First, its knowledge about belief justification depends only on the abstract planning structure of the different types of beliefs being justified, and not on the specific domain (knowledge about events) or the specific task (arguing or correcting misconceptions). Second, it demonstrates how the same knowledge about belief justification can be used to build a model of the user´s belief and to provide an appropriate advisor response