AM
Lenat's program started from basic set-theoretic notions and, guided by heuristics of interestingness, derived concepts such as prime numbers by itself.
Why it matters
An attempt to automate not the solving of a problem but the posing of one, a rare case where the goal was discovery rather than an answer.
AM held around 250 heuristics of the form: if a function often returns the same value, examine it separately. The program rediscovered primes and the Goldbach conjecture, then ran out of steam as its new concepts grew emptier. Lenat was later criticised for results that depended on the vocabulary built in. That limit is what led him to Cyc.