---
title: "AI Winters Explained: Why AI Booms Bust"
slug: ai-winters-explained
description: "What the AI winters were, when and why: the ALPAC and Lighthill reports, the Lisp machine collapse, and whether today's AI boom is different."
date: 2026-10-26
last_reviewed: 2026-10-26
status: published
type: knowledge-base
image: ai-winters-explained
image_alt: "Illustration of a large glass dome with cracks spreading across it, enclosing rows of lit data-center buildings at dusk."
image_caption: "Illustration: a cracking dome over a data-center campus, a picture of boom and bust. It does not show a real facility. AI-generated image."
tags: [ai winter, history, lighthill report, alpac, expert systems, ai investment]
---

# AI Winters Explained: Why AI Booms Bust

**In short:** an AI winter is a period when funding, commercial investment and confidence in artificial intelligence contract sharply after promises go unmet. Historians usually name two: a first in the 1970s, after the ALPAC report (1966) ended most US machine-translation funding and the Lighthill report (1972 to 1973) curtailed UK research, and a second in the late 1980s, when the expert-system and Lisp-machine market collapsed and DARPA cut AI programs. Whether today's boom will end the same way is Contested.

The encyclopedia's full treatment is in [AI winters](../vol02_history/HIST-004_ai_winters.md). Terms used below are defined in the [glossary](/glossary.html).

## Quick reference

| Event | Date | What happened | Evidence grade |
|---|---|---|---|
| ALPAC report | 1966 | US National Academy of Sciences committee found no near prospect of useful machine translation | Established |
| Lighthill report | Dated July 1972, published by the UK Science Research Council | Judged that AI had not delivered the "major impact" promised; UK AI research curtailed | Established |
| "Dark Ages of AI" panel | AAAI-84 conference, 1984 | Leading researchers warned that a downturn ("AI winter") was coming | Established |
| DARPA cuts | From September 1987 | New office director Jacob Schwartz cancelled or did not renew some AI programs | Well Supported |
| Lisp machine and expert-system decline | Late 1980s | Specialized hardware lost out to general workstations and PCs; many AI firms folded | Well Supported |
| A third winter after the current boom | Not yet | Disputed; no consensus on likelihood or timing | Contested |

## What is an AI winter?

The term describes a funding and reputation cycle, not a technical one. Expectations rise, often encouraged by researchers and vendors; sponsors fund ambitious goals; results fall short; money and prestige drain away. Nils Nilsson, in his history *The Quest for Artificial Intelligence* (Cambridge University Press, 2010), puts it this way: in the early 1980s many AI sponsors "had greatly inflated expectations of what AI could do," and the failure to deliver, together with mounting criticism, "combined in the mid- to late 1980s to bring on what came to be called an 'AI winter.'"

The phrase itself is tied to a panel at the 1984 AAAI conference, published in *AI Magazine* in 1985 as "The Dark Ages of AI." Its chair, Drew McDermott, spoke of "deep unease" among researchers that expectations were too high and would "eventually result in disaster." Nilsson quotes the panel using the words "AI winter."

## What did the ALPAC report (1966) conclude?

In April 1964 the US National Academy of Sciences formed the Automatic Language Processing Advisory Committee (ALPAC), chaired by John R. Pierce of Bell Laboratories, to advise the Department of Defense, the CIA and the National Science Foundation on machine translation (MT). Its report, *Language and Machines: Computers in Translation and Linguistics*, was published by the National Academies in 1966.

As quoted by Nilsson, the committee concluded that "there is no immediate or predictable prospect of useful machine translation." The historian John Hutchins, also quoted by Nilsson, called its influence "profound": it "brought a virtual end to MT research in the USA for over a decade," and the field's centre moved to Canada and Europe. Hutchins calls 1967 to 1976 MT's "quiet decade." See [Natural language processing: a history](../vol02_history/HIST-012_natural_language_processing_a_history.md).

ALPAC judged one application, not AI as a whole, but it set a pattern: an outside review measures results against promises and funders act on the gap.

## What did the Lighthill report say, and what did it do?

The UK Science Research Council (SRC) asked Sir James Lighthill, Lucasian Professor of Applied Mathematics at Cambridge, for an outsider's survey of AI. His report, *Artificial Intelligence: A General Survey*, is dated July 1972 and was published by the SRC with replies from AI researchers (usually cited as 1973; Nilsson dates the volume 1972).

Lighthill split the field into three categories: A, "Advanced Automation"; C, computer-based studies of the central nervous system; and B, the "Bridge" between them, mainly "Building Robots." He favoured continued work in A and C but was critical of most basic research in B. His central judgment: "In no part of the field have the discoveries made so far produced the major impact that was then promised." He blamed the "combinatorial explosion": in areas such as theorem proving, the number of possibilities grows so fast that it cancels out "any advantages from increase in computer power." He also cited the US National Academy review of machine translation as a "notorious" disappointment.

Nilsson writes that the report "resulted in a substantial curtailment of AI research in the United Kingdom," including Donald Michie's FREDDY robot work at Edinburgh. The historian Jon Agar (2020) is more careful: the report's criticisms "have been held to be a major cause" of the slowdown, and he notes that the SRC gave "a preliminary steer" to an apparently independent report. The cuts are Well Supported; how much the report caused them, rather than reflected decisions already forming, is debated.

## Why did the second AI winter happen in the late 1980s?

The 1980s boom was built on [expert systems](../vol02_history/HIST-006_expert_systems_dendral_mycin.md): programs that encode specialists' if-then rules. Companies sold expert-system "shells" and Lisp machines, workstations built at MIT to run the LISP language efficiently and sold by Lisp Machines, Inc. and Symbolics. National programs added fuel: Japan's Fifth Generation Computer Systems project (1982), the UK's Alvey programme, Europe's ESPRIT and DARPA's Strategic Computing program.

Three things turned the boom, according to Nilsson's account:

1. **Hardware economics.** The Lisp machine companies "gradually lost out to other providers of workstations." Expert systems could run on "low-cost workstations and personal computers," so many expert-system firms "ceased to exist, were bought by larger companies, or had to reorient."
2. **Government retrenchment.** Jacob Schwartz, sceptical of some AI approaches, became director of DARPA's Information Systems Technology Office in September 1987 and "promptly canceled some AI programs and failed to renew others." Strategic Computing "gradually disappeared from view." Japan's Fifth Generation project left software that, Nilsson notes, "did not find notable applications."
3. **Unmet promises.** In Nilsson's words, assessments of "AI's difficulties and achievements, compared to some of its promises, led to the end of the boom in the late 1980s." Indicators fell: AAAI membership declined through the late 1980s and levelled at 4,000 to 5,000 by 1996.

Research did not stop. Nilsson describes a shift to "more modest and achievable goals," and statistical methods and neural networks kept advancing, leading eventually to the [deep learning breakthrough](../vol02_history/HIST-005_deep_learning_breakthrough.md).

## Is today's AI boom heading for another winter?

This is Contested. The scale of today's boom is documented: Stanford's 2026 AI Index reports that US private AI investment reached 285.9 billion US dollars in 2025 and that organizational adoption reached 88%. Its first takeaway is that "AI capability is not plateauing." See [The contemporary AI era](../vol02_history/HIST-011_the_contemporary_ai_era_2022_onward_living_record.md).

| Argument | Points toward | Grade |
|---|---|---|
| Current systems earn revenue across many industries, unlike 1960s MT or 1980s shells | No winter, or a milder correction | Contested |
| Investment depends on expectations of continued scaling, which may not hold (see [scaling laws](./ai-scaling-laws-explained.md)) | A funding pullback if progress stalls | Hypothetical |
| Reliability problems such as [hallucination](./why-ai-hallucinates.md) limit high-stakes use, echoing expert-system brittleness | Slower adoption than forecast | Contested |
| A financial correction could occur without a research winter, as after the dot-com bust | Market cycle, not field collapse | Hypothetical |

The historical lesson is narrower than "every boom busts": earlier winters followed specific failures against specific promises, made worse by a few concentrated funders. Watch the gap between public claims and measured results; the [Turing test and AI benchmarks](./turing-test-and-ai-benchmarks.md) post covers how that measurement works, and [The uncertainties that matter most](../vol12_forecasts/FCST-UNCERT-001_the_uncertainties_that_matter_most.md) covers what could change the picture.

## Frequently asked questions

**How many AI winters have there been?**
Most histories count two: one in the 1970s and one from the late 1980s into the early 1990s. Some writers also treat the post-ALPAC machine-translation slump of the late 1960s as a separate, narrower winter.

**Who coined the term "AI winter"?**
The term is linked to the "Dark Ages of AI" panel at the 1984 AAAI conference, where leading researchers warned of a coming downturn. No single coiner is confirmed in the sources checked here.

**Did the Lighthill report end AI research in Britain?**
No. It led to substantial cuts, notably at Edinburgh, but research continued at a smaller scale. How far the report caused the cuts is debated by historians.

**Why did Lisp machines fail?**
Cheaper general-purpose workstations and personal computers became able to run AI software, removing the reason to buy specialized hardware.

**Will there be another AI winter?**
No one knows. Investment and adoption were at record levels in 2025, but whether progress continues to justify them is Contested.

## Sources

- Nilsson, N. J.: [The Quest for Artificial Intelligence (Cambridge University Press, 2010; author's web edition, archived)](https://web.archive.org/web/2024/https://ai.stanford.edu/~nilsson/QAI/qai.pdf)
- Lighthill, J.: [Artificial Intelligence: A General Survey (July 1972; copy hosted by the University of Edinburgh)](https://www.aiai.ed.ac.uk/events/lighthill1973/lighthill.pdf)
- Agar, J.: [What is science for? The Lighthill report on artificial intelligence reinterpreted (British Journal for the History of Science, 53(3), 2020)](https://doi.org/10.1017/S0007087420000230)
- National Research Council (ALPAC): [Language and Machines: Computers in Translation and Linguistics (1966)](https://doi.org/10.17226/9547)
- McDermott, D., Waldrop, M. M., Chandrasekaran, B., McDermott, J., Schank, R.: [The Dark Ages of AI: A Panel Discussion at AAAI-84 (AI Magazine 6(3), 1985)](https://ojs.aaai.org/aimagazine/index.php/aimagazine/article/view/494)
- Stanford HAI: [The 2026 AI Index Report](https://hai.stanford.edu/ai-index/2026-ai-index-report)

*Status: draft, reviewed 26 October 2026 against the primary sources above. The ALPAC conclusion and the Hutchins quotation are given as quoted by Nilsson; the report's own text could not be read online. The exact size of the 1970s and 1980s funding cuts were not checked against primary budget records, and investment figures for the current boom change every year.*