Key takeaways
- An expert system has three main parts: a knowledge base of rules, an inference engine and an explanation facility.
- MYCIN advised on infections and R1, later called XCON, configured computer orders.
- Expert systems faded because capturing and maintaining expert knowledge as rules was slow and costly.
- Their ideas live on in rules engines, configure-price-quote checks and hybrids that pair language models with rules.
How an expert system works
An expert system works by applying a specialist's rules, stored in a knowledge base, to the facts of a new case through an inference engine. A knowledge engineer interviewed experts and turned their judgement into if-then rules; users then answered the system's questions about a case and received advice.
Many expert systems could also explain themselves. Asked why it wanted a fact, or how it reached a conclusion, the system listed the rules it had used. That made its advice easier to trust and its mistakes easier to trace, a quality later AI methods have found hard to match.
Expert system examples
The best-known expert system examples come from medicine, chemistry and computer manufacturing. Each showed that narrow, rule-rich expertise could be put into software, and each showed the cost of doing so.
- MYCIN, built at Stanford, suggested likely causes and treatments for bacterial infections; it stayed a research system rather than a tool in routine use.
- DENDRAL helped chemists work out the structure of molecules from measurement data.
- R1, later called XCON, checked and completed customer orders for Digital Equipment Corporation computers so that every part fitted together.
Why expert systems faded
Expert systems faded because getting knowledge into them, and keeping it current, cost more than most organisations could sustain. Experts often cannot put their judgement into words, rules multiplied as exceptions surfaced, and every change in the domain meant more interviews and more rules.
They were also brittle at the edges: a case slightly outside the rules got a wrong answer delivered with full confidence. As machine learning made it possible to learn patterns from data instead, attention moved away from hand-built rule sets.
Expert system vs machine learning
An expert system reasons from rules people wrote, while machine learning reasons from patterns a model found in data. The expert system can explain every step but knows nothing outside its rules; a learned model copes with variety but cannot easily say why it answered as it did.
Knowledge-based system is the broader name for the first family: any system that works from an explicit store of knowledge rather than from training alone.
Where expert system ideas live on
Expert system ideas live on wherever software checks a case against written rules. A configure-price-quote tool that will not let a rep sell an add-on without the base licence, or a compliance check before a contract goes out, is a small expert system in everything but name.
The ideas are also back in research. Recent work pairs a language model with a rule-based system, using the model to read messy text and the rules to reach decisions that can be audited, one small proof of concept doing this for medical reports. These are early studies, not a settled method.
Where Pebbles Ai fits
Common questions
Are expert systems still used?
Rarely under that name, but their pattern of written rules applied by an engine runs inside many business tools, from loan checks to product configurators.
What is a knowledge-based system?
It is the broader family expert systems belong to: any system that reasons from an explicit store of knowledge rather than from patterns learned in training. Expert systems are the ones built around one specialist's know-how.
What did a knowledge engineer do?
A knowledge engineer interviewed specialists and turned their judgement into rules a program could apply. Getting tacit expertise into words was the hardest part, and the main reason expert systems were costly to build.
About this page. Written by the Pebbles Ai team for people who run sales, marketing and growth. Last updated .
