Black October, explainedAppsGuidesAI DictionarySmall FixesOn RecordAboutFollow the lab

AI winter

Also called AI winters

An AI winter is a stretch of years when big promises about AI went unmet, so money and interest dried up and research slowed.

The basicsEveryday

In plain words

AI has had seasons. In good years, researchers make bold promises, governments and companies pour in money, and newspapers predict thinking machines. Then the results arrive, and they are smaller than promised.

When that gap gets too wide, funders lose patience. Grants are cut, companies close, and researchers stop putting the word AI on their funding requests. That cold spell is called an AI winter.

Historians usually name two: one in the mid-1970s and a deeper one from the late 1980s into the 1990s. Research never stopped. Quiet work during the cold years, such as neural networks that learned to read handwritten postcodes, laid the ground for the deep learning boom of the 2010s.

Everyday example

The café that promised too much

A new café opens with posters promising the best food in the city. Investors line up and so do customers. But the dishes turn out ordinary, the crowds thin, investors pull out and staff are let go. The place goes quiet. In the back, the cook keeps practising one dish. Years later that dish is truly good, word spreads, and the queues come back, bigger than before.

Where the example breaks: A café rises or falls on one kitchen. AI winters hit a whole field across several countries, and they were driven as much by government budgets and politics as by the quality of the science.

Go deeper

How it actually works

Each winter followed the same pattern: expectations rose faster than results, then funders corrected hard.

  1. The first chill: translation and the Lighthill report

    In 1966 a US advisory committee, ALPAC, found no immediate prospect of useful machine translation, and US funding for it largely ended for about twenty years. In 1973 the Lighthill report, written for the UK Science Research Council, led to what Edinburgh's own history calls a massive loss of confidence in AI among UK academics.

  2. Tighter money in the US

    From 1969 the Mansfield Amendment pushed US military research money toward projects with a clear mission. DARPA, the US defence research agency, judged its speech understanding programme had missed its goals and ended it in the 1970s without the planned follow-on.

  3. The expert system boom and bust

    In the early 1980s companies rushed into expert systems, programs built from hand-written rules, and bought special Lisp machine computers to run them. The rules were costly to maintain and broke outside narrow tasks. In 1987 the market for Lisp machines collapsed, and the second, deeper winter set in.

  4. Where the name comes from

    The term is usually traced to a 1984 panel at the AAAI conference in Austin, Texas. Researchers including Roger Schank warned that expectations about AI were too high and would eventually end in disaster. Three years later, the bust they feared arrived.

  5. The thaw

    Through the 1990s and 2000s, statistical machine learning kept proving useful in practical jobs. In 2012 a deep neural network won the ImageNet image recognition contest by a wide margin, and money and attention flooded back into the field.

Usual dates
About 1974 to 1980, and 1987 into the 1990s. Sources differ
Name traced to
A 1984 panel at the AAAI conference, Austin, Texas
Big triggers
ALPAC report (1966), Lighthill report (1973), Lisp machine collapse (1987)
A dissent
Historian Thomas Haigh argues the 1970s dip was smaller than the legend, since the field kept growing

Every AI boom since has raised the same question: is another winter coming?

Questions people ask

What caused the AI winters?

Mostly a gap between promises and results. Early AI looked brilliant on small puzzles but struggled with messy real problems like translation and speech. Computers were slow and data was scarce. When official reviews judged the field had overpromised, governments and companies cut funding.

How many AI winters have there been?

Two are usually named: one in the mid to late 1970s and a deeper one from the late 1980s into the 1990s. Sources differ on exact dates, and some historians argue the first was milder than often told, because the research community kept growing through the 1970s.

What was the Lighthill report?

A 1973 review of AI research by the mathematician Sir James Lighthill for the UK Science Research Council. It was sharply critical of basic research in areas such as robotics and language. It led to a big loss of confidence in AI in UK universities and is often named as a trigger of the first winter.

Is another AI winter coming?

Nobody knows. Unlike in the 1970s and 1980s, today's AI is built into products many people use every day, which makes a full freeze less likely. A cooling is still possible if costs stay high and promised leaps, such as artificial general intelligence, take longer than expected.

Further reading

Prepared by Mundhra Labs. Reviewed by Aditya Mundhra. Checked October 2026.