On the application of rule-based techniques to the design of advice-giving systems

Abstract

This article attempts to assess how much is known about building systems whose advice actually benefits users. We review current approaches to the provision of on-line help, and suggest that the most promising are those which represent a user's intentions explicitly. Following this lead, we examine recent work on speech acts, planning and meta-level inference for clues as to how a user's inputs could be interpreted in the context of his current aims and activities. We conclude that the appropriate context of interpretation for an input is supplied by hypotheses concerning the current state of a user's plan. Next we suggest that the techniques developed in rule-based systems could be used to implement an advisor capable of generating and revising plan hypotheses, and outline what we consider to be the outstanding problems of control associated with such an implementation. Finally, we show how such a system might help a user to structure his activity, so that he can iterate towards his goal while avoiding common errors.