An expert system is a knowledge base decision support system which helps people make decisions. It is an example of an intelligent system, as it produces responses which are very close to those produced by humans.
An expert system normally contains a knowledge repository into which experts provide information. The user inputs problems into this expert system, which it processes through the system repository to generate a solution.
In order to produce an expert system, the knowledge repository or knowledge base must be built by Knowledge Engineers, who extract information from an expert. They then encode the data using facts and rules using correct syntax. The engineer must have communication and technical skills in order to elicit all the information and recognise and fill in any gaps.
The brain of the expert system is the inference engine, a piece of complex software which uses facts and rules in the knowledge base together with current conditions entered by the user to draw an appropriate conclusion.
Depending on the responses of users, the inference engine determines which rule to fire next. Irrelevant rules and facts are discarded progressively, therefore narrowing the process. For a more in-depth look, see components of a legal expert system. Examples of these can be found at JustSys, eGanges and RuleBurst.
The user interface is the point of interaction between the user and the expert system. Whether or not a system is user-friendly will have a direct influence over the user’s reaction to the system, the ease of use and how much training might be required for users. Thus it is crucial that it is effectively designed.
Questions asked of the user must be meaningful, relevant, unambiguous and as simple as possible.
The user interface generally includes:
This can be requested by the user at any time to provide an aid to explain why a solution has been offered or why a particular question has to been asked. It does this by simply listing the fired question, each of which is stored as the session proceeds on a working database/database of facts. Once the session has finished this temporary data is deleted.
The list of rules which have been fired, and the user-entered responses to questions asked by the software, which are stored on the disk. The information is used whenever the user asks for an explanation (see above) of conclusions reached so far.
This allows expert systems to produce an answer in terms of probability, as a human would. Responses such as highly probable, likely, yes etc. are produced to represent numeric probabilities produced by the software. The percentage number is known as the certainty factor. Probabilities are assigned to individual rules and are mathematically combined to give an overall probability for a given solution.
Rule of thumb, which may not be able to be proved, but has a high probability of being correct. This is valuable in expert stems.
Method of determining a solution or goal by starting at the first rule and successfully eliminating irrelevant rules based on user responses. Works best where there are a large number of possible solutions, or one solution is very often the valid one.
The system assumes the most common goal. It works backwards by asking questions of the user to determine that the necessary conditions exist. If the necessary conditions are not satisfied, that goal is discarded and the system starts again assuming the second most common goal.
See Russell Allen’s collection of resources under the Expert Systems Tab.