A gentle introduction to knowledge-based systems in medicine

In summary, AI expert systems, through the example of medical problem solving, have been examined, considering goals, problems addressed, and problems resolved (or not resolved). How these systems attempt to replicate the knowledge and strategies of human experts has been shown, as well as how they determine these factors (in the form of the protocol). Through the example of diagnosis, some of the types of knowledge and approaches such systems must encode have been illustrated. Concrete examples of approaches to encoding were presented using MYCIN and PIP. These systems demonstrate more clearly the complexity of the problem domains under consideration, how that complexity can be dealt with, and the limitations and potential of AI. It is apparent that such systems have unique contributions to make, not only in terms of straightforward usefulness but also in terms of inspectability, which may be extended to a capacity for "explaining." On a more basic level, they are generating a reexamination of what is considered "intelligent" behavior--which may itself lead to future concepts, systems, and tools. Moreover, the fundamental goal of generality in the design of AI systems makes such things as, for example, the hypothetico-deductive model of behavior transferrable across domains, conferring a similar ability for revitalization and reexamination in each one.