A classic example is MYCIN, built at Stanford in the 1970s to suggest treatments for blood infections. Its rules looked roughly like “if the infection is in the blood and the organism has these features, then it is likely to be this bacterium.” The program chained hundreds of rules together and could explain which rules led to its conclusion. Businesses in the 1980s used similar systems to configure computer orders, approve credit and diagnose equipment faults.
Expert systems are a form of symbolic AI. Their knowledge is written by people, not learned from data. That makes them predictable and easy to audit, but slow and costly to build, and brittle when a situation falls outside the rules.
Why it matters
Modern machine learning systems learn patterns from data instead of hand-written rules, which makes them more flexible but harder to explain. Many production AI systems today combine the two: a learned model with explicit business rules around it.
No published Chain of Thought episode discusses expert systems directly yet, so this entry relies on the sources below.
Related terms
Symbolic AI is the broader tradition. Robotic process automation is a modern rule-driven automation approach.
Sources
- Buchanan and Shortliffe: Rule-Based Expert Systems (the MYCIN experiments) — Full text of the 1984 book documenting MYCIN, the Stanford rule-based consultation program for diagnosing and treating infections; checked October 24, 2026.
- Russell and Norvig: Artificial Intelligence: A Modern Approach — Standard AI textbook site; the book covers knowledge-based and rule-based systems and the history of expert systems; checked October 24, 2026.