AI Winter
An AI winter is a period of sharply reduced funding, commercial investment and scientific enthusiasm for artificial intelligence that follows a period of inflated expectations. The term was coined in 1984, at a meeting of the American Association for Artificial Intelligence, when researchers including Roger Schank and Marvin Minsky warned that a business boom built on overpromising would be followed by disillusionment, much as a nuclear winter follows a nuclear exchange. Historians usually identify two real downturns: one beginning in the mid-1970s and a second in the late 1980s extending into the early 1990s.
The first winter grew out of unmet early promises and critical reviews. Optimistic predictions from the 1950s and 1960s, including claims that machines would match human intellect within a generation, collided with the fact that toy demonstrations did not scale to real problems because of combinatorial explosion. In 1966 the ALPAC report in the United States judged machine translation slower, less accurate and costlier than human translation, and funding for it dried up. Minsky and Papert's book "Perceptrons" (1969) proved limits of single-layer networks and dampened enthusiasm for neural approaches. The Mansfield Amendment of 1969 pushed DARPA toward mission-oriented work, and the Lighthill report, commissioned by Britain's Science Research Council and published in 1973, concluded that AI had failed to achieve its grand objectives; it led to deep cuts in British university research around 1974. DARPA likewise curtailed open-ended programs such as speech understanding after disappointment with results.
The second winter was a commercial bust as much as a scientific one. Expert systems had produced real successes, and in the 1980s companies spent heavily on them and on dedicated Lisp machines from vendors such as Symbolics and Lisp Machines Inc. In 1987 the Lisp machine market collapsed as general-purpose desktop workstations from Sun and others became both cheaper and faster. Rule bases proved expensive to maintain and brittle outside their narrow domains, so enthusiasm cooled. Japan's Fifth Generation Computer Systems project, launched in 1982 with an ambitious logic-programming vision, did not deliver its aims by its end in 1992, and DARPA's Strategic Computing Initiative was scaled back in the late 1980s. Many AI companies failed or retreated.
A winter was a contraction, never a complete stop. Work continued, often under different names such as machine learning, pattern recognition, informatics, knowledge-based systems or intelligent agents, because the AI label had become a liability for grant proposals. Backpropagation was popularized by Rumelhart, Hinton and Williams in 1986, support vector machines and probabilistic graphical models matured in the 1990s, and Deep Blue beat Garry Kasparov in 1997. A common misconception is that nothing happened for years; in truth the field changed its methods, moved toward statistics, benchmarks and measurable results, and built the foundation for the deep learning revival that began with AlexNet in 2012.
The lesson is about the gap between claims and capability, and it is hard to apply to the present. Both winters followed a recognizable pattern of demonstrations that impress, promises that extrapolate them, deployments that reveal brittleness, and funders that withdraw. Whether current enthusiasm for large models resembles that pattern is debated: they now bring huge revenue and daily use, which earlier technologies lacked, yet disappointing reliability or cost could still trigger a pullback in specific areas. Anyone claiming to predict a new winter is guessing, but the history offers a good checklist, asking what has actually been delivered, at what cost, under which conditions, and how much depends on a single approach.
Read the AI winter as a lesson in expectation management rather than a verdict on the field: twice, overpromising met brittle technology and funding fell, and each time the work that survived was the work that delivered measurable results and rebuilt itself on firmer foundations.