The DARPA cuts
Era 2 · AI is born (1943–1979) › The first winter
- Lineage
- Symbols and logic
- Kind
- Institutional
Who
DARPA, Carnegie Mellon University
What happened
DARPA scales back open-ended funding for academic AI research and requires every project to have a military use. In 1976 it ends its speech understanding program without a successor, even though Carnegie Mellon's HARPY system had met the goals set for it.
Why it mattered
The era of generous funding with no specific project ends in the United States. It is the date the story of the first AI winter usually starts from, a story some historians dispute.
The full story
In 1971 a study group chaired by Allen Newell proposed a five-year speech understanding program to the agency: systems that would understand spoken sentences on a limited topic, with a vocabulary of about a thousand words. Several groups took part, among them Carnegie Mellon and the firm BBN.
The program ended in 1976. Carnegie Mellon's HARPY understood sentences with a vocabulary of 1,011 words and a 5 percent semantic error rate, below the 10 percent required. It was the only system that met the specifications, and it introduced beam search, a technique speech recognition used for decades. Even so, DARPA did not continue the work: the results were far from any military use, and according to the most widely told version, the agency felt misled.
The end of the program came on top of the rule change brought by the Mansfield Amendment and pressure from Congress for the agency, renamed DARPA with a D for Defense, to justify every project by its military use. The big AI labs, used to stable and open-ended funding, had to submit specific projects with deadlines and applications.
1974 is usually taken as the start of a first AI winter that would last until 1980. The historian Thomas Haigh disputes that account: the data Haigh examines show no general collapse. The ACM's AI group nearly tripled its membership over those years, and the field kept growing outside the labs most dependent on DARPA. The phrase "AI winter" is also later: it first appeared in 1984, in a debate at the annual meeting of the AAAI.
Part of a period
- The first AI winter (1974–1980)
Sources
Scholarship and context
- Review of the ARPA Speech Understanding Project. The Journal of the Acoustical Society of America, 62(6), pp. 1345-1366 (opens in a new tab). Dennis H. Klatt, 1977
- A Historical Perspective of Speech Recognition. Communications of the ACM, 57(1), pp. 94-103 (opens in a new tab). Xuedong Huang, James Baker, and Raj Reddy, 2014
- There Was No ‘First AI Winter’. Communications of the ACM, vol. 66, no. 12, pp. 35-39 (opens in a new tab). Thomas Haigh, 2023
Further reading
How to cite
“The DARPA cuts.” The History of AI, revised November 13, 2026. https://thehistoryofai.com/en/milestone/darpa-cuts-1974
This entry was last reviewed on November 13, 2026.