Diabetes expert systems: planning for long term use.

Expert systems may play an important role in the future in assisting diabetics to control their disease. However the data available suggest that expert systems are difficult to build and more difficult to maintain over a long period. An examination of the problems in maintaining one expert system, GARVAN-ES1, suggests that the problems arise because experts never report on how they reach a decision, rather they justify why the decision is correct. These justifications vary markedly with the context in which they are required, but in context they are accurate and adequate; the difficulties arise in taking them out of context. It is suggested that expert system building techniques must be able to capture knowledge in context and tools must be available to flexibly change the context in which an expert system knowledge base is examined. Two implementations of such strategies, "ripple down rules" and a "knowledge dictionary", are outlined.