Glossary
What is an expert system in AI?
In short
An expert system is a program that solves tasks in a narrow field by rules written down from specialists, and it can explain how it reached its conclusion.
AI basics · expert system
MMOOZZGG.com editorsChecked: 1 min read
For a child
A doctor keeps a thick notebook of tips in the office: “if the throat hurts and the temperature is high, check this”. Imagine that such a notebook was given to a computer. It asks you questions, looks through the rules and tells you what to do next. It invents nothing by itself; it only follows the advice of experienced adults carefully.
In plain words
Expert systems appeared long before neural networks. They are built like this: a knowledge engineer interviews an experienced specialist and writes down their experience as rules: “if we see this, then the conclusion is that”. The program stores hundreds of such rules and a mechanism that goes through them. The user answers questions, the system combines the rules, gives a recommendation and, on request, shows the chain of reasoning. The downside is that all knowledge is written by hand: the program does not learn from examples and cannot cope with anything that is not in the rules.
For business
Such systems still work in banks, insurance and industry under other names: business rule engines, product configurators, application checking systems. They are valued for predictability and transparency: you can show a regulator the rule by which a customer was refused. The weak spot is maintenance: the rules grow into thousands, they contradict each other, and when the expert leaves, the base has no owner. Today the approaches are often combined: a language model understands a customer’s free-form question, and the rules check the answer and decide what can be promised. A similar task is solved by an AI assistant that looks for the answer in the company’s documents instead of in hand-written rules.
For a specialist
An expert system consists of a knowledge base with facts and if-then rules and an inference engine that applies them by forward chaining (from data to goal) or backward chaining (from goal to data). The classic example is MYCIN from the Stanford Heuristic Programming Project, work on which began in the spring of 1972. The program gave diagnostic and therapy advice for patients with infections and handled uncertainty through certainty factors. MYCIN grew out of the earlier DENDRAL system, and EMYCIN was split off from it: a shell without the medical knowledge base, for building new systems. Moving an expert’s knowledge into a program was called knowledge engineering. The limits of the approach: costly knowledge collection (the knowledge acquisition bottleneck), brittleness at the edges of the field, and no learning from data. The ideas live on in rule engines and neurosymbolic systems.
Example
A boiler maker gives its service engineers a program: the technician enters the error code, the pressure and the age of the equipment, and the program, using the chief engineer’s rules, names the likely fault and the part needed. A newcomer solves on site the problems for which they used to call the office.
Related terms
Questions and answers
Sources
- B. G. Buchanan, E. H. Shortliffe (eds.), Rule-Based Expert Systems: The MYCIN Experiments of the Stanford Heuristic Programming Project, 1984people.dbmi.columbia.edu
- B. G. Buchanan, E. H. Shortliffe, Rule-Based Expert Systems, Preface, 1984impact.dbmi.columbia.edu
- B. G. Buchanan, E. H. Shortliffe, Rule-Based Expert Systems, Chapter 1, 1984people.dbmi.columbia.edu
Where companies use it
Not sure whether your company needs this?
AI solutions map