AI winter
/ AY-EYE WIN-tur /
An AI winter is a period when excitement about artificial intelligence collapses, funding dries up, and the field falls out of fashion — after a boom in which promises ran far ahead of what the technology could actually deliver. The pattern is human, not technical: bold predictions attract money and headlines; the systems disappoint; disillusionment sets in; investors and governments pull back; researchers even avoid the label 'AI' to get funded at all.
There have been two big ones. The first, roughly in the 1970s, followed the heady early days of symbolic AI: a 1973 UK report (the Lighthill report) and skeptical reviews in the US judged that the grand promises — fluent machine translation, general problem-solvers — had not materialized, and money was cut. The second, in the late 1980s and early 1990s, followed the collapse of the expert-system industry, when those costly, brittle systems failed to pay off and the specialized hardware built for them became obsolete.
The point to carry away is sobering and useful: AI has always moved in hype cycles. A surge of optimism (and investment) is reliably followed by a correction when reality falls short of the marketing. Knowing this history is the best antidote to today's breathless claims. It does not mean current progress is fake — the advances are real — but it is a standing reminder to separate genuine capability from promotional promise, and to be wary whenever 'AI will soon do everything' is in the air.
In 1966 researchers expected machine translation to be largely solved; a US advisory report (ALPAC, 1966) found it slower, costlier, and worse than human translators, and funding was slashed for years. The lesson recurs: the gap between a confident demo and a reliable product is where winters are born.
Overpromise, underdeliver, retrench — the recurring shape of an AI winter.
AI winters were caused by inflated expectations, not by AI being worthless — useful work continued quietly throughout. The history is a reason to read every 'AI will soon do X' claim with healthy skepticism, today included.