Expert Systems

An expert system is a computer program that aims at capturing and using a human expert's knowledge, making it available to non-experts. The focus of the program is usually decision-making, diagnosis, or prediction. The significance of expert systems lies in their application to domains where expertise is at risk of getting lost (e.g., old-fashioned machinery) or is scarce (e.g., mineral prospecting), or to situations in which experts are distant (e.g., medical aid on an oil platform).

The computing power of an expert system derives from artificial intelligence (AI). The first generation of expert systems relied upon so-called “production rules,” rules for actions to take when certain conditions were met. Alan Newell and Herbert Simon were the first to use production systems in the intelligent solving of logic problems (1956) and in simulating human intelligence (1972). Their developments were based on the idea that human beings use heuristics—something like rules of thumb—to solve problems, rather than deterministic rules. In the early 1970s, Edward A. Feigenbaum started a project called “Heuristic Programming,” which aimed at the practical application of heuristic ideas in the realm of AI.

Two main issues related to the development and use of expert systems are knowledge acquisition and Expert Systems [Page 188]knowledge engineering. Knowledge acquisition refers to the process by which a non-expert in the domain, a knowledge engineer, works in order to capture knowledge from the expert(s). This process includes collecting documents and interviewing experts, as well as observing them during their work. The knowledge-acquisition process can be time-consuming, and has to be intertwined with the work of actually getting the system to function in a similar way as the expert.

Knowledge engineering is the process by which the acquired knowledge is formulated in terms that a computer system can understand. Nowadays, there are several support systems available (so-called “expert system shells”) to help the knowledge engineer to model the expert's knowledge.

Designing the user interface is an important issue in the development of an expert system. The system has to ask the user questions about a given problem, use the answers to work out a problem solution, and present this solution to the user. Early in the development of expert systems, it was found that the presentation of a solution was often insufficient; the user also needed an explanation as to why the presented solution was reasonable. Deriving explanations from a rule-based system was found to be very difficult. A model-based system, by contrast, enabled its user to get explanations, constructed on the basis of the model.

However the knowledge is represented, the dialogue with the user has to be understandable. Therefore, a great deal of work is invested in the creation of natural language interfaces, which may achieve a “natural” dialogue with the user.

Expert systems have been viewed optimistically as well as critically. Optimists have been validated by the multitude of problems that have been found to lend themselves to AI solutions, from process planning in industry to individual health planning. At the same time, the fear of expert systems abounds. Can we really trust expert-system solutions? Who should be blamed should the system be wrong? These questions led to the requirement that expert-system performance be evaluated against the performance of human experts.

...