Knowledge Engineering

Knowledge engineering is the branch of artificial intelligence that focuses on encoding expert human knowledge into computer systems, typically in the form of rule-based or symbolic representations. The approach was dominant during the 1980s, when expert systems such as MYCIN (medical diagnosis), DENDRAL (chemical analysis), and R1 (computer configuration) demonstrated that narrow domains could be automated by capturing the decision-making rules of human specialists.

The knowledge engineering approach assumes that expertise consists of explicit rules, facts, and heuristics that can be extracted from human experts through structured interviews and formalized into machine-executable logic. This assumption proved brittle in practice. Expert systems performed well within their narrow domains but failed catastrophically at edge cases, struggled with incomplete or conflicting information, and required expensive, ongoing maintenance as domain knowledge evolved. The approach was largely superseded by machine learning methods that learn representations from data rather than encoding them from human expertise — a transition that Rich Sutton later articulated as the bitter lesson.