Expert systems (covers)

Covers are focused on rather narrow class of tasks, though wider, than that program on the basis of which the cover is created. So, the author of the EMYCIN system Van Melle (van Melle) emphasized with one of the first that covers are not universal architecture for problem solving at all. The EMYCIN system developed by it is intended for such problems of medical diagnosing with large volumes of data which give in to the solution using deductive approach in the assumption that space of diagnostic categories permanently.

At the same time, it is impossible to trust too recommendations of a possibility of use of a cover for the solution of specific problems. Today there is no so clear idea of classification of the tasks solved by expert systems that it was possible to provide precisely to what class it is necessary to carry a specific system.

Scope of covers

The majority of commercial products of this type is suitable only for problem areas with small search space. Here usually use a method of exhaustive search with creation of the return inference chain and limited abilities to manage process. But some modern covers can be applied to the solution of a wide range of tasks as support a set of functions of representation of knowledge and management, including also modeling of a direct chain of a logical output, the procedure, transmission of messages, etc.

Simplicity of the knowledge representation languages applied to the majority of covers is, on the one hand, advantage, and with another – a lack of such systems. So, the formalism of production rules used in some expert systems complicates separation of different types of knowledge – heuristic, managing, knowledge of the expected values of entities. Capability of a system to differentiate types of knowledge is one of the main conditions of ensuring its "transparency" for the user

Methods of formation of judgments

Often, as the main method of formation of judgments in a cover of expert system, use creation of the return inference chain. At the same time sets of rules meta - and object level are selected. It is as a result very difficult to create an explanation, exhaustive and clear for the user. Those decisions which are made at a stage of programming of rules, in particular, concerning an order and the number of expressions in an antecedent part can affect strikingly a way of search in a decision space in the course of system operation.

Other problem concerns functionality of systems which are based on rules generally, so, and all covers in which production rules are used as the main knowledge representation language. The lion share of the knowledge base is knowledge of standard cases, of those which often meets in a problem area. Experts easily solve standard problems and are capable to classify them in terms of ideal prototypes even at insufficient completeness of data. In a type of their abstractness, impracticablly to provide such knowledge in expert system only by means of rules of the condition action form. Here more difficult formalism which will nullify one of the main advantages of use of production rules as a fixed asset of decision making is necessary.

Uncertainty processing

The following problem of covers concerns the uncertainty processing mechanism. Some covers already include the certain formal mechanism of work with uncertainty, for example, based on use of coefficients of confidence. However in covers also such mechanisms which will not be approved with outputs of probability theory are used and have the properties which are hard giving in to the analysis. The specific method of processing of uncertainty at the solution of a specific objective in this data domain can give pragmatical justification in relation to the scheme of processing of coefficients of confidence. Extend this device to other scopes, having built in it a cover it is inexpedient.

In comparison with the first developments modern covers more flexible, at least, that without special work can be integrated into the majority of the operating environments available in the software market and are equipped with rather developed means of the user interface.

Multifunction environments of creation of covers

Multifunction software environments allow the experienced programmer to experiment at the solution of new classes of problems, selecting suitable combinations of the different methods provided in the available modular set. As there is no only universal language of representation of knowledge for any expert system, it is necessary to integrate several different schemes of representation, especially at a stage of creation of a prototype. The full theory of such hybrid systems does not exist, but experiments with different schemes of representation and a logical output showed that everyone has the shortcomings. Therefore the greatest effect brings consolidation of different techniques so that the dignity of some compensated shortcomings of others.

The structured objects, for example frames, were practical means for storage and manipulation of descriptions of objects of a problem area, but use of such knowledge requires inclusion in the program of fragments of a program code which then are difficult for analyzing. The main idea of the first attempts of data together of the styles based on rules and frames consisted in integrating capability to represent the objects characteristic of frames, with opportunities to connect conditions and actions using production rules.

One of the first multifunction environments of artificial intelligence is LOOPS in which within uniform architecture of message exchange four principles of programming were integrated:

• procedure oriented programming. The LISP language in which active component are procedures, and passive – data.
• the programming focused on rules. The principle is similar previous, but a role of procedures is played by the rules "condition action". The set of rules is connected with managing components, necessary for the conflict resolution in the simplest form.
• object-oriented programming. The structured objects have dual structure: they have properties and procedures, and data, and ghost effects are usually localized within an object.
• the programming focused on data. The data access and their updating starts certain procedures, and does not matter why the component of data is changed. It can be result of a ghost effect or action of other procedures

Within the basic object-oriented principle of creation it is possible to combine the environment modules maintaining different programming styles. Usually conditions in production rules and logical phrases contact values of slots of the structured objects, and rules modify values of these slots. Such style of consolidation of paradigms is implemented in the CLIPS language now.

Connected subjects

• Expert systems
• Expert systems (Architecture)
• Development of expert systems
• Expert systems (representation of knowledge)

Links

• MGSU CAD in Construction department of MGSU
• Martelk