Reference: Langlotz, C. P. & Shortliffe, E. H. The Critiquing Approach to Automated Advice and Explanation: Rationale and Examples. Ellis Horwood, London, 1989.
Abstract: One of the most important lessons from the past decade of medical AI research is that expert-level decision-making performance does not guarantee user acceptance. Several medical advice systems demonstrate the ability to provide advice comparable to that of experts. Nevertheless, proven advice systems have gained only limited acceptance in the medical setting. The lack of explanation capabilities may be a substantial impediment to the integration of advice systems into the practical setting. We highlight a unique approach to the problem of explaining automated advice--the critiquing model. We begin with a review of explanation systems that exemplify successful and popular explanation techniques. Then, we focus on the critiquing model and describe several examples of its use. In conclusion, we introduce a simple classification of explanation systems, and illustrate how each of the systems we have discussed fits into the taxonomy.