Textbook of AI

An AI winter is a period of significantly reduced funding, public interest and academic respectability for artificial-intelligence research. The phrase, coined by Roger Schank and Marvin Minsky during a 1984 AAAI panel, was a deliberate echo of the nuclear winter then in public discourse. Two AI winters are usually distinguished:

The first AI winter (~1974–1980) followed the 1966 ALPAC report's effect on US machine-translation funding and the 1973 Lighthill report's effect on UK AI funding. Combinatorial explosion in symbolic search, the failure of micro-worlds to scale, and over-claims by AI proponents combined to produce a sharp pullback by funders.

The second AI winter (~1987–1993) followed the collapse of the Lisp machine market (Symbolics, LMI), the recognition that expert systems were expensive to maintain and brittle outside their narrow domains, and the Japanese Fifth Generation Computer Systems Project's failure to deliver on its 1981 ambitions. Many AI companies failed; many AI departments contracted; the term AI became sufficiently tainted that researchers preferred to describe their work as "machine learning", "knowledge-based systems" or "informatics".

Some commentators identify smaller intra-decade slowdowns (a 2004–2008 lull, for example) as further AI winters; others reserve the term for the two major episodes. With the post-2012 deep learning revolution and the post-2020 LLM boom, AI funding and attention reached unprecedented levels, but discussions of a possible third AI winter resurface periodically when models fail to meet inflated expectations.