Managing the development and deployment of expert systems

The logic paths in an expert system are nonlinear, allowing coverage of a highly complex problem space in a very concise form. The author maintains that few expert systems have survived commercial application because the traditional methods of development and deployment have failed to compensate for complexity. The author examines the evaluation and validation problem within the context of the life cycle. He offers cautions and suggests a solution for a restricted class of applications. In particular, development should follow a well-known progression from prototype to working system and E&V should be formalized on the basis of scientific method and include formal verification, bench testing and field testing.<<ETX>>