DENDRAL (Stanford, from the mid-1960s) helped chemists infer molecular structures from mass spectrometry. MYCIN (1970s) suggested antibiotics for bacterial infections with certainty-weighted rules.
The pattern was clear: interview experts, encode heuristics as if–then rules, explain the inference chain. Companies built commercial shells; Japan launched the Fifth Generation project.
Niche success did not scale to common-sense knowledge. The cost of maintaining rule bases and their brittleness on novel cases prepared the ground for learning from data.
