History

Expert System

An expert system encodes a human expert's domain knowledge as hand-written if-then rules plus an inference engine, the dominant AI approach of the 1970s-80s.

An expert system encodes a human expert's domain knowledge as a large base of hand-written if-then rules, plus an inference engine to chain them together — the dominant AI approach of the 1970s–80s (e.g. MYCIN for medical diagnosis). Expert systems were genuinely useful for narrow domains but didn't generalize or scale: every new domain needed a new expert to interview and a new rule base to maintain by hand — the bottleneck that machine learning later solved by learning patterns from data instead of encoding them manually.

How it works

An expert system separates two components. The knowledge base holds declarative rules — IF fever AND stiff-neck THEN suspect meningitis — and the inference engine applies them. Chaining runs in one of two directions: forward chaining starts from known facts and derives whatever conclusions they support, while backward chaining starts from a hypothesis and searches for rules whose conclusions would establish it, recursively asking the user for facts it's missing. Rules often carry confidence weights so competing conclusions can be ranked, and a conflict-resolution strategy decides which of several matching rules fires first. Because every step is an explicit rule, the system can print the chain that produced an answer — an explainability property deep learning models lack.

When it breaks

  • Knowledge acquisition is the bottleneck. Every rule has to be extracted by interviewing an expert, and experts are poor at articulating judgment they apply unconsciously.
  • Rule bases turn brittle past a few hundred rules. New rules interact with old ones in ways nobody anticipated, and no maintainer can hold the full interaction surface in their head.
  • Coverage is exactly the enumerated cases. Anything outside them yields no answer rather than a degraded one.
  • The approach survives where inputs are structured and decisions must be auditable — tax and compliance engines, underwriting, clinical decision support — and increasingly as guardrails constraining LLM output rather than as a replacement for it.

See also: AI, Machine Learning

Learn more: History & Landscape · Wikipedia: Expert system

Mentioned in

Lessons where this comes up in context.

On this page