The practical rationalization of expert systems

The practical advance of Expert Systems requires a demystification, necessary, if unsettling, to those looking for the magic solution. The paper will illustrate potential strategies for relating Expert Systems to control. It also illustrates the systematic advances possible when attention is limited to small designs and to application classes with usable analysis structure. This scale and structure allows full design rationalization. The concept of the small, structured Expert System has already been used in a controlled development of the EXACT Pattern Recognition adaptation design. This design is based on a systematic rule structure and a formal strategy of experimental design validation. It illustrates one approach to design rationalization. For the theoretician, inference (Bayes´ Rule or Resolution Principle) algorithms can be examined formally. The Expert System rules can be devised rigorously, incompleteness being the consequence not of heuristic rules, but of an incomplete rule set, or an incompletely rationalized structure. This theory provides another basis for design rationalization. Perhaps specially designed combinations of these formal elements will serve engineering design better than informal general tools. It should be understood that pure Expert Systems and Shells are inherently special-purpose systems incapable of general computation. In most colorless terms, the Expert System can be thought of as systematically managing large collections of IF statements (rules), allowing experimental evolution of the resulting solution and automatic positioning of each individual rule. As computer science structures, Expert System structures can be rationalized in relation to other conventional software structures. Finally, from a pure engineering perspective, the rules and their roles can also be individually classified and analyzed in their relation to an application. As long as rules are simple things like IF statements, their structure in a particular appli- ation may be the most powerful basis for design rationalization.