The contributing editors were recently asked about “inference engines” and, after blushing about their ignorance, have determined that the term refers to a computer programming technique used in “expert systems.”
The so—called “expert systems” are computer systems that employ unique programming techniques to model expert decisions. The systems imitate human reasoning processes by asking for data about a problem or situation and then making deductions based on the information given.. A system designed to evaluate the likelihood of heart disease might ask about the patient's weight, smoking and drinking habits, stress levels, etc. The information is compared with a data base of accumulated information and then expressed as conclusions. The system is capable of displaying the rules used to reach the conclusions.
There are two components of an expert system: a “knowledge base” and an “inference engine.” The knowledge base is a store of accumulated information on a particular subject. The knowledge is expressed in a series of rules, usually in an “IF … THEN … form—in other words, deductive reasoning. These rules are entered into the system by human experts. The first (IF) part of the rule is a series of conditions: for example, the symptoms of an illness. The second (THEN) part of the rule will be a conclusion that can be deduced if the condition in the first part is met. The “inference engine” component of the system is the software necessary to carry out the comparison of data and rules and manipulate any probability facilities the system uses. The two parts of the expert system are separate so that knowledge bases and inference engines can be recombined depending on the requirements of a particular situation.
Expert systems are constructed by a team of specialists drawn from a subject area and from computer science. To date, the fields in which the most work has been done in the development of expert systems are medicine and petroleum exploration. However, there are examples of expert systems in fields as diverse as tax advice and the identification of organic compounds by analysis of mass spectrograms. While some systems are still quite crude, the best of them perform with a reliability level equal to that of human experts in a field. Most of the existing systems require quite large computers. Some 75 percent are mounted on Digital Equipment Corporation Model 10 and 20 computers. The company has been a leader in expert system research among hardware manufacturers. IBM has recently become more involved in the field. The so-called “fifth-generation” computers being developed in Japan and the U.S. are seen as being important components in the development of very complex expert systems.
The leading work in expert systems is going on in university artificial intelligence studies departments such as those at Stanford, MIT, and Carnegie Mellon universities. The Lister Hill Center for Biomedical Communications, the research and development arm of the National Library of Medicine, is also an active participant in research in this area. In the commercial sector, the Stanford Research Institute, Cognitive Systems Inc., Artificial Intelligence Corp., and Tecknowledge are leaders.
