COMPASS is more complete in its analysis and thus may outperform the experts even for the experience-based portion of the system.)
There are some risks in basing the expert system's analysis on unproven knowledge, that is, system knowledge that is not based on the experience of the expert in the field. The resulting analyses will almost certainly contain some inaccuracies, and these may be very hard to detect. In our domain, maintenance of a highly redundant switching system, there was little risk of any major disaster. Any undetected problem can only get worse until it is eventually obvious enough to be detected. Even the least optimal ordering of maintenance actions will eventually solve a detected problem if the set of actions on the list is correct. The advantage of using the portion of COMPASS founded on expertise rather than experience is that there is a strong possibility that problems will be detected earlier -- in the SMA rather than the NR stage. The number of customer-affecting problems present in the switch is thus reduced, resulting in improved quality of service. So in this particular case, the payoff far outstrips the risk.
Chandrasekaran B., & Mittal S. ( 1982). "Deep versus compiled knowledge approaches to diagnostic problem solving". Proceedings of the National Conference on Artificial Intelligence (AAAI-82), pp. 349-354.
Davis R. ( 1989). Form and content in model-based reasoning. Paper presented at 1989 Workshop on Model Based Reasoning, 11th International Joint Conference on Artificial Intelligence, Detroit, MI.
Prerau D. ( 1987). "Knowledge acquisition in the development of a large expert system". AI Magazine, 8( 2), 43-51.
Prerau D. ( 1990). Developing and managing expert systems: Proven techniques for business and industry. Reading, MA: Addison-Wesley.
Prerau D., Gunderson A., & Levine S. ( 1988). "The PROPHET expert system: Proactive maintenance of telephone company outside plant". Proceedings of the Fourth Annual Artificial Intelligence and Advanced Computer Technology Conference (AI'88), Long Beach, CA.
Prerau D., Gunderson A., Reinke, R., & Adler M. ( 1990). "Maintainability techniques in developing large expert systems". IEEE Expert, 5( 2), pp. 71-80.
Scarl E. ( 1989). Sensor failure and missing data: Further inducements for reasoning with models. Paper presented at 1989 Workshop on Model Based Reasoning, 11th International Joint Conference on Artificial Intelligence, Detroit, MI.
Sticklen J., & Chandrasekaran B. ( 1985). "Use of deep-level reasoning in medical diagnosis". Proceedings of the Expert Systems in Government Symposium, pp. 496-502, McLean, VA.