AI Winter

An AI winter is a period in which disappointed expectations lead to sharp cuts in funding, commercial interest and public attention for artificial intelligence. The field has gone through two major winters – roughly from the mid-1970s to around 1980, and from the late 1980s into the mid-1990s – and the term returns whenever observers wonder whether today’s boom will end the same way.

The first winter

Early AI researchers promised a great deal. After the 1956 Dartmouth workshop, whose 1955 proposal coined the term “artificial intelligence,” pioneers predicted machines that would soon translate languages, prove theorems and match human intelligence. Progress proved slower. In 1966 a report by ALPAC, a National Academy of Sciences committee advising US funding agencies, concluded that machine translation had not lived up to expectations, and funding for it collapsed. In 1973 the mathematician James Lighthill, in a report for Britain’s Science Research Council, concluded that, outside a few narrow areas, AI had fallen well short of its early promises, which led to deep cuts in British support. In the United States, the Pentagon’s research agency DARPA moved toward mission-oriented projects and reduced funding for open-ended AI research. Marvin Minsky and Seymour Papert’s 1969 book Perceptrons, which set out the limits of simple neural networks, also by many accounts helped turn attention away from that approach for years.

The second winter

In the 1980s, expert systems that encoded specialists’ rules became a business, Japan launched its ambitious Fifth Generation Computer Systems project, and the United States and Britain answered with programs of their own. The boom faltered at the end of the decade. Expert systems proved brittle and costly to maintain, the market for specialized Lisp machines collapsed in the late 1980s as cheaper desktop computers caught up, and the Fifth Generation project ended in 1992 without reaching its goals. At a 1984 panel on “The Dark Ages of AI” at the conference of the American Association for Artificial Intelligence, Drew McDermott, Roger Schank and others had warned that inflated expectations would bring a backlash; the downturn that followed came to be called an AI winter.

Thaw and boom

Research continued through the winters – in statistical learning, neural networks and robotics – and the field returned with machine learning, faster computers and large datasets, culminating in the deep learning breakthroughs after 2012.

Will there be another?

Today’s investment dwarfs the field’s earlier booms, and debates over an AI bubble echo the old warnings; in November 2026 a one-day drop of about 5 percent in the Nasdaq, led by chip stocks, showed how quickly sentiment can turn. The differences are real: AI products now earn revenue and have hundreds of millions of regular users, and the price of a given task keeps falling. Still, a financial correction, disappointment with AI agents or a slowdown in scaling laws could cool enthusiasm, at least for a while.

Why it matters

AI winters show that overpromising has a price. Honest claims about what systems can do, measured by results in real use, protect a field from boom and bust – and keep useful research going when the hype fades.

Written with Claude, made by Anthropic, a company whose business depends on continued investment in AI and which began marketing its stock-market listing in November 2026.

Related

AI Bubble · Deep Learning · Neural Network · Bitter Lesson · Natural Language Processing · Scaling Laws · Techno-Optimism

Sources: James Lighthill: Artificial Intelligence: A General Survey (Science Research Council; Chilton Computing archive) (written 1972, published 1973) · National Research Council (ALPAC): Language and Machines: Computers in Translation and Linguistics (1966) · Drew McDermott et al.: The Dark Ages of AI: A Panel Discussion at AAAI-84 (AI Magazine) (1985) · Andrew Pollack: ‘Fifth Generation’ Became Japan’s Lost Generation (New York Times) (June 5, 1992) · Epoch AI: LLM inference prices have fallen rapidly but unequally across tasks (March 12, 2025) · Stock Analysis: Nasdaq 100 falls as chip stocks slide (November 5, 2026) · Nils J. Nilsson: The Quest for Artificial Intelligence (Cambridge University Press) (2009)

Leave a Reply

Your email address will not be published. Required fields are marked *