Despite the impressive achievements of current generative AI systems, the dream of Artificial General Intelligence remains far away, notwithstanding the hype offered by various tech CEOs.[1] The reasons are easy to state, if hard to quantify. Human intelligence requires three primary features, none of which have been fully cracked: logic, associative learning, and value sensitivity. I’ll explain each in turn.

Logic was once thought to be the apotheosis of human reasoning and the key to human intelligence.[2] Getting machines to reproduce logical inference was a massive breakthrough in the mid-1950s, with Newell, Simon, and Shaw’s Logic Theorist (1956)[3] and General Problem Solver (1957)[4], which were able to perform logical inferences and even prove some advanced mathematical theorems from the Principia Mathematica. The success reportedly prompted Simon to say to his students, “Over Christmas, Al Newell and I invented a thinking machine.”[5]

Millions upon millions of dollars were subsequently spent in anticipation of “solving” the problem of intelligence. But it didn’t work, not really. Logic-based AI—as useful as it remains today—proved brittle in the face of incomplete or contradictory information; perceptual inputs proved difficult (even impossible!) to capture in logical formulae; and as every schoolkid knows, logic is too hard to realistically be the whole or even the root of human cognition.[6]

This brings us to associative learning. Recognizing the co-occurrence of properties (smoke, fire) or the predictive........

© Psychology Today