Towards optimal expert system design

Expert system research has emphasized the development of systems that attempt to perform tasks at the level of experts in the field. There has been little attention paid to determining whether such an emphasis will result in expert systems that provide maximum value to the users of these systems. The authors briefly discuss past research to show that current development tools and methods will not result in systems that maximize value. While the emphasis has been on improving the outputs of these systems, the associated costs have not been carefully considered. As a consequence, the cost of providing the information that is required when using expert systems may be much more than it need be. One way to improve the situation is to produce the same quality outputs at a lower cost. Given an existing rule-base, the costs of acquiring inputs, and the probabilities of each input state, they show how a cost minimizing information acquisition strategy can be determined. Alternative approaches to implementing such a strategy are discussed