Many expert systems are rule-based, in the sense that they depend on a set of rules derived from human experts, or from extensive data bases. Effective methods of working through these rules and selecting an appropriate outcome depend on the initial information about a problem provided by the user, and the way in which the supply of additional information is prompted by the expert system in order to help the system to make the necessary decision. Forward and backward chaining strategies may use Bayesian inference methods in a process of obtaining evidence and modifying probabilities in ways that are computationally very elegant.

Expert or knowledge based systems may also be frame based, in the sense that the user is led through a series of 'frames' of information that focus closer and closer on the critical point at issue. Such systems often use hypertext methods that enable the user to highlight words or phrases that influence the direction of the search for information. Frame-based expert systems are like books that can be read in many different ways in the search for quite specific items of information.

There now exist a great many 'shells' which can be used as a basis for the construction of an expert system. The principal problem lies in getting the expertise out of the head of an expert - or even of oneself - in an appropriate form. Many of the methods suggested earlier in this Series can be useful in formatting information as a basis for the generation of appropriate rules or frames. In addition, genetic alogorithms can be used to extract rules from data bases in a form which makes them particularly appropriate for the development of an expert system.

Finally, computer programming languages such as PROLOG are especially useful in the development of expert system approaches to the management of information. These languages solve problems by applying logical deduction to the available facts and rules, in contrast to the more usual method of programming through a series of instructions. The resulting system represents a top-down description of the problem, with the knowledge base and the data incorporated into the body of the program. Indeed, a whole research project can be represented by a PROLOG program, with the main objective expressed as the first statement of the program, supported by the necessary evidence for the justification of that statement.