Expert Systems

An artificial intelligence system built on the basis of in-depth specialized knowledge about a certain subject area (obtained from experts in this field) is called an expert system. Expert systems are one of the few types of artificial intelligence systems that are widely used and have found practical application. There are expert systems in military affairs, geology, engineering, computer science, space technology, mathematics, medicine, meteorology, industry, agriculture, management, physics, chemistry, electronics, law, etc. And only that expert systems remain very complex, expensive, and most importantly, highly specialized programs, restrains their even wider distribution.

Features of expert systems:

• competence - in a specific subject area the expert system should reach the same level as human specialists; at the same time, it should use the same heuristic methods, also reflect the subject area deeply and widely;

• symbolic reasoning -   the knowledge on which the expert system is based represents the concepts of the real world in symbolic form, the reasoning also occurs in the form of transformation of symbol sets;

• depth - the examination should solve serious, non-trivial tasks, characterized by the complexity of the knowledge that the expert system uses, or an abundance of information; this does not allow using the full search of options as a method for solving a problem and forces to resort to heuristic, creative, informal methods;

• self-awareness - the expert system should include a mechanism for explaining how it comes to solving the problem.

Expert systems are created to solve various kinds of problems, but they have a similar structure (Fig. 8); The main types of their activities can be grouped into categories listed in Table. 2

Fig. 8. Diagram of a generalized expert system

Expert systems that perform interpretation, as a rule, use information from sensors to describe the situation. For example, it may be an interpretation of the readings of measuring instruments at a chemical plant to determine the state of the process. Interpreting systems deal not with clear symbolic representations of a problem situation, but directly with real data. They encounter difficulties that other types of systems do not have, because they have to process information that is “noisy”, insufficient, incomplete, unreliable or erroneous. They need special methods for recording the characteristics of continuous streams of data, signals or images and methods for their symbolic representation.

Table 2. Typical categories of methods for applying expert systems

Interpretive expert systems can handle various types of data. For example, the system of scene analysis and speech recognition, using natural information (in one case, visual images, in the other - sound signals), analyzes their characteristics and understands their meaning. Interpretation in the field of chemistry uses X-ray diffraction, spectral analysis or nuclear magnetic resonance data to derive the chemical structure of substances. The geological interpretation system in geology uses logging probing - measuring the conductivity of rocks in and around boreholes to determine subsurface geological structures. Medical interpretive systems, based on the testimony of the tracking systems (for example, temperature, pulse, blood pressure), establish the diagnosis or severity of the disease. In the military, interpretive systems, receiving data from radar, radio communications and sonar devices, assess the situation and identify targets.

Expert systems that make a forecast determine the likely consequences of a given situation. Examples are the prediction of crop damage from a certain type of harmful insects, estimating the demand for oil on the world market, and predicting the location of the next armed conflict based on intelligence data. Prediction systems sometimes use simulation modeling, i.e. programs that reflect the causal relationships in the real world to generate situations or scenarios that may occur with certain input data. Possible situations together with knowledge of the processes generating these situations form the prerequisites for forecasting. Artificial intelligence specialists have so far developed relatively few predictive systems, perhaps because it is very difficult to interact with imitation models and create them.

Expert systems perform diagnostics using situation descriptions, behavioral characteristics, or knowledge of the design of components to determine the likely causes of a malfunctioning diagnosed system. Examples are the determination of the causes of the disease by the symptoms observed in patients; localization of faults in electronic circuits and identification of faulty components in the cooling system of nuclear reactors. Diagnostic systems are often consultants who not only diagnose, but also help in debugging. They can interact with the user to assist with troubleshooting, and then suggest remedial actions. Medicine seems to be a completely natural area for diagnosis, and indeed, more diagnostic systems have been developed in the medical field than in any other single subject area. However, many diagnostic systems are currently being developed for engineering and computer systems applications.

Expert systems that perform design develop object configurations taking into account the set of limitations inherent in the problem. Examples include genetic engineering, the development of VLSI and the synthesis of complex organic molecules.

Expert systems engaged in planning, design actions; they define the complete sequence of actions before they start. Examples are the creation of a plan for applying a sequence of chemical reactions to groups of atoms in order to synthesize complex organic compounds or the creation of an air combat plan with the goal of neutralizing a certain factor of enemy combat capability.

Expert systems performing observation compare actual behavior with the expected behavior of the system. Examples include monitoring the readings of measuring instruments in nuclear reactors in order to detect emergencies or evaluating monitoring data of patients placed in intensive care units. Observing expert systems compare observed behavior with a set of acceptable situations of normal behavior. Observing expert systems, by their very nature, must operate in real time and implement the interpretation of the behavior of the observed object, which depends both on time and context.

Expert systems that perform training, subject to diagnosis, "debugging" and correcting (correcting) the behavior of the student. As examples, we will give students learning how to troubleshoot electrical circuits, teaching naval sailors how to use an engine on a ship, and teaching medical students how to choose antimicrobial therapy. Teaching systems create a model of what the student knows and how he applies this knowledge to solving a problem. Systems diagnose and point out to the learner his mistakes, analyzing the model and building plans for correcting these errors. They correct the behavior of students by fulfilling these plans with the help of direct instructions to the students.