The First AI Winter: 1974 to 1980

The phrase AI winter was coined by researchers borrowing from nuclear winter. Between roughly 1974 and 1980, government and corporate funding for AI dropped sharply in the US, UK, and eventually Japan. Labs closed. Graduate students switched fields. Conferences shrank.

The causes were a cluster of disappointments. Machine translation had flopped after the ALPAC report in 1966. Speech understanding programs missed DARPA's ambitious targets. The promised walking, seeing robots never arrived.

What researchers learned to do differently

• Narrow the problem to a specific domain, like medicine or chemistry
• Work with human experts rather than replacing them
• Publish limits along with successes
• Stop promising general intelligence by the end of the decade

Expert systems emerged directly from these lessons. They delivered real value on narrow tasks and brought some funding back by the early 1980s. But the habits of overpromising were only paused, not cured.

“AI has been a story of bursts of enthusiasm followed by disappointment, each time.”

The big idea: the first winter was not the end of AI but the end of a specific vision of AI. The field that came out of it was humbler, more applied, and quietly better.