The quiet work

During the winters, one thread survived.

A small group of researchers, underfunded and mostly ignored, had stopped asking "what rules does thinking follow?" and started asking something else entirely: what if machines didn't follow rules at all? What if instead of being told the answer, they learned it?

The idea wasn't new. Networks of simple connected units that could learn from experience had been proposed in the 1940s, inspired loosely by how neurons in the brain work. But they had been dismissed as too slow, too limited, and too far from anything useful. A few people kept working on them anyway.