An expert system is a program running on a computer. Like a human expert, it knows a lot about a subject. People can ask the expert system a question. The expert system will then use a set of rules and give answers to the question. This method of automated reasoning belongs to a field of computer science called artificial intelligence.
Different groups of people can have different kinds of access to an expert system. The people managing the network of computers have different needs than the office worker or secretary.
How expert systems work[change | change source]
Expert systems are made of
- A set of facts, rules and principles
- A set of data that is solved using its expertise
- A user interface
When they are asked a question, they will filter the data with the rules they have. They might give back a result, or ask an additional question.
Categories of expert systems[change | change source]
- Case-based expert systems have a number of cases; each case describes the problem in context, and the solution of the problem, in the specific case. The current problem is matched as closely as possible to a case. The solution of the case found is then applied to the current problem. The main problem of such systems is to define how cases relate to each other, or are similar. An example of where such a system may be used is that of a patient: This patient shows a given number of symptoms, and the expert system could diagnose the patient.
- Rule-based systems do not rely on cases; instead, there are a number of rules. These are expressed in the form IF A THEN B. In most systems, human experts have to write the rules.
- A third approach is to construct systems that are based on decision trees. Such systems are capable of learning, or of extending their knowledge base, using inductive reasoning. When the classification is done, the system uses a path through the tree. At the end, it arrives at a leaf, which indicates the class of solution to the problem. Each bifurcation in the tree is based on an attribute, that is checked. The value of the attribute determines how to continue the path. Ideally, good results are obtained using small trees. The problem is finding good attributes to base the decisions on.