Artificial intelligence came to be largely synonymous with "expert systems" during the 1980s (May, 1991, p. 118). Expert systems are rule-based computer programs that enable users to take advantage of a vast array of information in order to make decisions. The system compares the current situation to a list of rules and determines the appropriate solution. Expert systems have been used in legal practices, physicians' offices, education, insurance and a variety of other fields. However, there are drawbacks to expert systems, including the availability of an expert, which make them unsuitable for all applications. Neural networks offer a solution to this problem and are being implemented in a number of different applications and environments. Determining whether a neural network is the appropriate response to an artificial intelligence application is the subject of this research.
Expert systems require the assistance of an expert in order to be implemented. In addition to having access to an expert, the individual (or individuals, since more than one expert is often used) must be able to explain how they arrive at a particular decision. It is not enough for an expert to tell the interviewer what the answer to a problem is; the expert must be able to describe the reasoning behind the solution. In this way, rules can be developed and coded into the expert system for later extraction. This is a time-consuming process that can have many iterations before a successful group of rules is arrived at, and the typical development time for an expert system is 12 to 18 months (Caudill, 1991, p. 116).
Neural networks, on the other hand, attempt to mimic the processing of the human brain, although on a much smaller scale. This eliminates the need for consulting an expert, and development time is greatly reduced (to weeks or months) (Caudill, 1991, p. 116). Neural networks do not rely on a set of rules that are based on individual int...