Expert systems (1970s–1980s) encoded domain expertise as if-then rules and drew inferences with forward or backward chaining. MYCIN diagnosed blood infections; XCON configured VAX computers. They were commercially successful for a decade, then largely disappeared into the substrate of ordinary software.
- 01State the two-part architecture of an expert system.
- 02Explain why expert systems failed to generalize.
- 03Identify a modern application that is essentially an expert system in disguise.
The architecture
A knowledge base of if-then rules elicited from human experts. An inference engine that chains through rules to reach conclusions. A separation that promised — but rarely delivered — maintainability by non-programmers.
The knowledge-acquisition bottleneck
Getting expertise out of experts and into rules was slow, expensive, and lossy. Every rule was a maintenance liability as the domain evolved. When machine learning could learn from data instead of interviews, expert systems were the first paradigm to be displaced.
- Knowledge base + inference engine.
- Knowledge acquisition was the fatal bottleneck.