A stretch of time when hopes for AI collapsed and funding and interest dried up.
An AI winter is a period when expectations for AI ran far ahead of what it could actually do, then collapsed, taking research money and interest with them. People usually count two: one in the mid-1970s and one in the late 1980s.
Around the first, big promises such as machine translation came along slower than hoped, and the 1969 book Perceptrons showed the limits of one-layer neural networks, which cooled that research too. The second came when the boom in expert systems, built from rules typed in by hand, fizzled out. Research never fully stopped, and that work later fed into deep learning.
Think of it like this
You know a game with an amazing trailer that turns out disappointing? Players leave, and you hear nothing about a sequel for ages. AI winters were like that: the technology didn't vanish, the hype cooled first.
For example
Whenever AI hype gets very loud, someone asks: "Is another AI winter coming?"
Common mix-ups
Mix-upAI research stopped completely during the AI winters.
ActuallyWhen funding got hard to win, some researchers carried on under other labels such as "pattern recognition" or "machine learning" instead of "AI". They changed the sign over the door and kept working.
Mix-upThere was one AI winter, with one cause.
ActuallyAI winters are usually counted as two, and even at the same time different fields had different seasons. When expert systems were taking off in the early 1980s, neural networks were still in retreat — but as that boom faded in the late 1980s, backpropagation spread and neural network research began to stir again.
Try it in class
For teachers at the start of a lesson — or for students working on their own.
- Question
- What signs tell you that expectations are running far ahead of what AI can do?
- Activity
- Draw a line on the board from 1950 to today. Each group marks when expectations rose and when they cooled, using this entry and record E0006. Then pick three recent headlines, split each into "what's promised" and "what's already been shown", and mark the ones with a big gap. Records on this site labelled "company-reported" are worth a look too.
Records about this
Source-checked moments from AI history.
Pinned down the limits mathematically, triggering a slump in neural network research that lasted 10 years or more