Expert System. Structure of an Expert System

An expert system is a computer information system developed to act like a human expert in a specific area of knowledge. It is an interactive computer-based decision tool that uses both facts and heuristics to solve difficult decision problems based on an expert’s knowledge. Because the knowledge of an expert tends to be domain-specific rather than general, expert systems that represent this knowledge usually reflect the specialized nature of such expertise.

Expert systems provide the means for overcoming the shortcomings of conventional human decision-making processes and conventional software through integrating human expertise and power of computers.

Although a generally accepted view on traditional computer program is summarized as
Traditional computer program = Data + Algorithm, the expert system can be described as

Expert system = Knowledge base + Inference engine

An expert system consists typically of the following major components:

- Knowledge base comprises specific knowledge about the problem domain under consideration. It differs from a database because much knowledge in the knowledge base is represented implicitly. The knowledge is most commonly represented in terms of production rules.

A production rule has the following structure:
- IF conditions
- THEN conclusions

- Knowledge-acquisition interface helps experts to express knowledge in a form that can be incorporated in a knowledge base. Determination of the problem domain and its characteristics, identifying the concepts that are used to describe the objects and their interrelationships; acquiring the knowledge; and representing it through suitable representation technique, such as production rules, implementation, and validation, can be listed as the stages of the knowledge acquisition process.

- Inference engine employs general rules of inference to arrive at logical conclusions according to the knowledge base of the expert system. Two main inference approaches are used by an inference engine to exploit the knowledge base: forward chaining (data driven) and backward chaining (goal-driven reasoning). Forward chaining begins with data input by the user and scans the rules to find those whose antecedent conditions are fulfilled by the data. It then fires those rules and deduces their consequences.

The consequences are added to the knowledge base, and the rules are revisited to observe which new rules may now be fired. This process is repeated until all rules which may be fired have been fired (25). As opposed to forward chaining that is data driven, backward chaining is goal driven because the inference process is guided by the final goal or objective that should be reached rather than by the available information. The process identifies rules that have the goal as a consequence.

- User interface is responsible for the form of communicating with the user. User interface attempts to equip the user with most capabilities of interacting with a human expert. The user interface presents the conclusions and explains the reasoning for justification purposes. Most of them may provide sensitivity analysis and what-if analysis tools to observe the changes that would have occurred if the variables had taken different values.

Figure 2. Structure of an expert system

Figure 2 illustrates the structure of an expert system and the interrelationship between its components. Particularly for the cases in which substantial information and data processing and analysis are required, expert systems derive conclusions at a much faster rate compared with human experts. Furthermore, expert systems are apt to deal with incomplete and uncertain information. However, the knowledge acquired in the expert system depends on the expert, and thus, the conclusions obtained are prone to change with knowledge elicited from a different human expert.