Short answers, pulled from the story.
An AI winter is a period of reduced funding and interest in artificial intelligence research, typically following a cycle of excessive optimism, overpromising by researchers, and failure to deliver results that match expectations. The pattern has repeated several times since the 1950s, with two major winters occurring approximately 1974-1980 and 1987-2000.
The term first appeared in 1984 at the annual meeting of the American Association of Artificial Intelligence. Researchers Roger Schank and Marvin Minsky introduced it in a public debate, comparing the anticipated collapse to nuclear winter.
Multiple factors converged: the Lighthill report of 1973 criticized AI research in the UK and led to the near-total dismantling of the field there; the Mansfield Amendment of 1969 forced DARPA to cut undirected academic research funding; and the ALPAC report of 1966 had already ended funding for machine translation after roughly 20 million dollars was spent. Researcher Hans Moravec attributed the crisis partly to a cycle of increasingly unrealistic promises made to DARPA.
The collapse of the LISP machine market in 1987 triggered the second major AI winter. General-purpose workstations and desktop computers became powerful enough to run AI programs, making specialized LISP machines unnecessary. An industry worth half a billion dollars was replaced in a single year. The failure of expert systems and Japan's Fifth Generation computer project, which ended in 1992, compounded the downturn.
Funding cuts hit major laboratories hardest, but researcher numbers continued to grow. ACM's SIGART membership doubled between 1969 and 1973, and nearly tripled again by mid-1978 to 3,500, suggesting the broader research community was expanding even during the supposed darkest years of the first winter. Historian Thomas Haigh has argued on this basis that there was no broadly based AI winter in the 1970s.
Interest began recovering in the early 2010s, with a turning point in 2012 when the deep learning network AlexNet won the ImageNet Large Scale Visual Recognition Challenge with half the error rate of its nearest competitor. The late 2022 release of ChatGPT, which reached over 100 million users by January 2023, further accelerated the current boom. Total AI investment reached 50 billion dollars in 2022.