The Wikipedia article "AI winter" (https://en.wikipedia.org/wiki/AI_winter, page created Dec 28, 2005; retrieved May 18, 2026) describes the recurring periods of reduced funding and interest in artificial-intelligence research that have followed waves of inflated expectation. It identifies two major winters (~1974–1980 and 1987–2000) and several smaller episodes, and frames the present moment as an AI boom dating from around 2012.
Definition and origin of the term
An "AI winter" is described as a period of reduced funding and interest in AI research, typically following a wave of inflated expectations that disappointment, criticism, funding cuts, and stagnation in turn. The field has experienced several hype cycles, each followed by years, and sometimes decades, of dormancy before renewed interest.
The term first appeared in 1984 at the AAAI annual meeting, in a public debate where Roger Schank and Marvin Minsky — both veterans of the 1970s winter — warned the business community that 1980s enthusiasm had spiraled out of control and would collapse. They described a chain reaction analogous to "nuclear winter": pessimism in the AI community leading to pessimism in the press, then cutbacks in funding, then the end of serious research. Three years later, per the article, the billion-dollar AI industry began to collapse.
The two major winters
The first major winter, ~1974–1980, was triggered by the 1973 Lighthill report (UK Parliament) and DARPA's parallel disillusionment with speech-understanding and other "mission-oriented" programs. The Lighthill report led to the dismantling of AI research in the UK, with only Edinburgh, Essex, and Sussex continuing. DARPA's funding cuts followed the 1969 Mansfield Amendment, which forced mission-oriented funding criteria. Hans Moravec attributes the cuts to a "web of increasing exaggeration" in researcher promises.
The second major winter, ~1987–2000, was triggered by the collapse of the LISP machine market in 1987, as specialized hardware was made obsolete by cheaper general-purpose workstations from Sun, IBM, and Apple; by the cancellation of the Strategic Computing Initiative in 1988; and by widespread abandonment of expert systems in the 1990s as they proved too expensive to maintain and too brittle in deployment. Japan's Fifth Generation Computer project, which had triggered the 1980s arms race, missed its original goals.
Smaller episodes
The article references several earlier or narrower episodes:
- 1966 — Failure of machine translation after the ALPAC report concluded MT was more expensive, less accurate, and slower than human translation (~$20M spent at NRC). The episode is a foundational source for the underestimated word-sense disambiguation and commonsense-knowledge problem.
- 1969 — The critique of perceptrons by Minsky and Papert ended mainstream single-layer neural-network research. Backpropagation, which would have rescued multilayer networks, was still years away. Funding for neural nets was difficult to find through the early 1980s.
- 1971–75 — DARPA's frustration with the Speech Understanding Research program at CMU (HARPY/HEARSAY-II). DARPA cancelled the $3M/year contract in 1974, feeling "duped," though the project's underlying technology (hidden Markov models) became the basis for the $4B/year speech-recognition industry by 2001.
- 1973–74 — DARPA's broader cutbacks to academic AI research.
Counterargument: "there was no winter"
Historian Thomas Haigh argues that 1970s AI funding cuts "did not slow progress." ACM SIGART membership tripled between 1973 and mid-1978 (1,241 to 3,500), growing faster than ACM overall. On this account, the "winter" affected the major laboratories and largest projects but coincided with rapid growth in the global researcher population. Haigh's "researcher headcount kept rising" critique is the principal empirical pushback against framing periodic capital-market corrections as full-field winters.
Recovery and current cycle
The article states that enthusiasm has generally increased since the early 1990s low. Beginning around 2012, it attributes increased funding and investment from research and corporate communities to the current (as of 2026) AI boom. The entry frames the present moment as a boom, implicitly raising the question of whether a third winter is plausible.
Provenance and use
This entry serves as the historical anchor for the AI Winter concept page and for the AI Bubble Debate. The boom-bust pattern it establishes is the historical precedent that current contrarian framings of the 2026 AI valuation and capex environment reference (see AI Chip Mania Sows Seeds of Its Own Destruction (Mackintosh, WSJ, May 16 2026), Circular Financing in AI).
Relationships
- depends-on: AI Bubble Debate — winter pattern is the historical precedent the debate references
- supports: AI Acceleration Paradox — boom-bust framing complements the productivity-paradox claim
- contradicts: (none directly — this is a primary reference, not an argumentative position)
- related: AI Chip Mania Sows Seeds of Its Own Destruction (Mackintosh, WSJ, May 16 2026), AI Economic Primitives, Scaling Laws
Key quotes
"Three years later [after the 1984 AAAI debate], the billion-dollar AI industry began to collapse."
"Many researchers were caught up in a web of increasing exaggeration. Their initial promises to DARPA had been much too optimistic." — Hans Moravec
"The general purpose robot is a mirage." — Title of the 1973 BBC "Controversy" debate following the Lighthill report (Lighthill vs. Donald Michie / John McCarthy / Richard Gregory)