AI Winter
The periods when AI's promises outran its results and the money left — twice, and the question of whether the pattern is over is genuinely open.
Reviewed July 13, 2026Stable
When not to use it
- (It's a historical pattern, not a tool.)*
- As a prediction. The pattern rhyming isn't evidence it repeats. Revenue is a real structural difference.
- As reassurance. "This time is different" preceded both winters.
- To conflate market and science. Neural networks were correct throughout the second winter and unfundable anyway.
- As one phenomenon. One was a correct technical critique; one was a hardware market shift plus an engineering wall.
Reach for something else instead
- (Ways to think about it instead.)*
- Hype cycles — the general form, less loaded.
- Paradigm exhaustion — a specific approach hits a wall; the field doesn't.
- Watching revenue, not capability claims — the thing that actually distinguishes the situations.
This entry is part of a longer guide: What is artificial intelligence?
Read more on the blog
- What actually caused the AI wintersThe book that supposedly killed neural networks proved a true theorem and attached a false conjecture. The field remembered the conjecture. Three myths about the AI winters, and what the record shows instead.
- Who invented deep learning, and why it took so longBackpropagation was invented at least four times before it stuck. The ideas behind deep learning were mostly in place by 1990. What was missing was not insight.
- AI vs machine learning vs deep learning: the differenceThese three terms are used interchangeably and are not interchangeable. They are nested: deep learning sits inside machine learning, which sits inside artificial intelligence. Knowing which circle you are in tells you what to expect about data, cost, transparency, and how the system will fail.
Further reading
- Lighthill (1973), Artificial Intelligence: A General Survey — the report that dismantled British AI, and its combinatorial explosion argument was correct.
- Crevier (1993), AI: The Tumultuous History of the Search for Artificial Intelligence — the account written from inside the second winter.
- Russell & Norvig (2020), Artificial Intelligence: A Modern Approach, ch. 1 — the standard sober history.
Primary sources, listed so you can check the claims on this page rather than take them on trust.
Where people go wrong
- Treating Lighthill as a fool. He was right about combinatorial explosion; it did kill symbolic AI.
- Assuming a winter means the technology was fake. Backprop was published during the run-up to the second one.
- Missing that researchers just rebranded. The work continued; the word became radioactive.
- Reading "the pattern is repeating" as evidence. Structural differences — revenue, diffusion — are the actual argument.
At a glance
FieldFoundations
Two of them~1974–80 and ~1987–93
The firstLighthill's combinatorial explosion critique, which was correct
The secondLisp machines obsoleted, expert systems unmaintainable
The mechanismreal breakthrough, extrapolation past evidence, money, wall, exit
The lessona winter is a funding event, not a research event
DifficultyBeginner
Often compared with
Winter vs. refutation — one is the money leaving, the other is the science being wrong. Neural networks were correct throughout the second winter and unfundable anyway.
Where this sits
A destination. 2 concepts lead here, and nothing in the corpus depends on it.
2Levelsteps in
2Needs firstconcepts
0Opens upnothing further
1Areastays here
Learn these firstSymbolic AI
Computed from the prerequisite graph, not assigned. How this works