The First Winter
An AI winter is a period when the money and the excitement dry up because the technology cannot deliver what was promised. There have been two big ones.
Two walls that caused it
Combinatorial explosion: a search that branches ten ways at each step has ten billion paths at ten steps. Clever pruning helps, but the world is too big to search naively. The frame problem: when something changes, which facts stop being true? A robot that moves a cup has to update half the universe of statements, and knowing which half is genuinely hard. Symbolic systems needed a human to tell them every relevant consequence.
Why the winters mattered
Funding follows promise, and promise outruns capability. The winters were not a failure of intelligence research; they were a failure of forecasting. Each one scattered talent into other fields, only for those ideas to return later.
The quiet decade
Between the winters, the work continued — quietly, in university labs, on problems nobody was funding. Backpropagation was being understood. Convolutional networks were being tested on postal codes. Reinforcement learning was being formalized. Every one of those threads would matter enormously once compute caught up.