Expert Systems: Matching Techniques to Tasks.

The overall flow of the discussion is in the direction of the evolution of expert systems from numerical programs to highly organized symbolic structures engaged in distinct types of problem-solving and communicating with one another.

Abstract

Abstract : The major line of argument that we will pursue in this paper can be outlined as follows. In Sec. II, we briefly trace the development of the idea of knowledge-based systems in AI. Sec. III is devoted to discussing the increasing need for symbolic content to expert reasoning as the size and demands of the task domain increase; i.e., we will analyze why a complete mathematical model of the situation, even if available, will not meet many of the demands placed on expert reasoning. In Sec. IV, we discuss the several distinct senses and roles that the notion of rules can play and have played in expert systems, and how a failure to keep these separate can cause a great deal of confusion. In Sec. V, we will argue that further organizational constructs, such as concepts and types of problem solving, are needed both to construct more powerful expert systems, and to characterize their capabilities. We will also provide two examples of generic problem-solving types, and show how each type of problem-solving induces an organization of knowledge in the form of a cooperating community of specialists engaged in that problem solving type. The overall flow of the discussion is in the direction of the evolution of expert systems from numerical programs to highly organized symbolic structures engaged in distinct types of problem-solving and communicating with one another.