A Perspective on Explanation in Diagnosis

This position paper argues that causal explanation in diagnostic tasks are more easily achieved in fault-based or failure-driven reasoning versus model-based reasoning. Faultbased or failure-driven diagnosis is more of a contextual task and can more easily be used to support user interaction through explanation than model-based diagnosis. A problem with fault-based reasoning that is often cited is the difficulty in reasoning about novel faults. In response to this key problem researchers in diagnostic systems are examining model-based reasoning. The latter approach is computationally expensive and in an effort to make it more efficient it is sometimes combined with fault-based reasoning techniques. The paper elaborates on explanations in RATIONALE and reasoning in JETA, both faultbased diagnostic systems. It also describes model-based diagnosis. Some of the issues and approaches needed to achieve causal explanation in model-based reasoning by integrating it with fault-based reasoning conclude the paper.