ARPA / DARPA

Contribution to AI

Money handed over in large blocks, to institutions rather than to proposals, is the instrument that built academic computing in America. From 1962 the Information Processing Techniques Office backed a small number of sites — MIT, Stanford, Carnegie Mellon, SRI — with contracts renewed on the strength of the people rather than a itemised plan of work, which let laboratories hire, buy machines and change direction without asking permission. Time-sharing was funded as a priority in its own right, so the researchers who wanted to build thinking machines had interactive computers to think on a decade before anyone else did. The network that linked those same contractor sites was, in its first years, largely a network of AI groups. The agency also demonstrated how quickly a field can be starved. After the Mansfield Amendment pushed spending towards mission-relevant work, the Speech Understanding Research programme of 1971 to 1976 ended with cancellation even though Carnegie Mellon's Harpy had met the thousand-word target, and the withdrawal of support across AI in the mid-1970s is the substance of the first winter. Strategic Computing spent roughly a billion dollars between 1983 and 1993 and delivered hardware rather than machine intelligence. The later Grand Challenges swapped contracts for prizes: nothing finished the 2004 desert course, five vehicles finished in 2005, and the teams involved became the staffing pool for the entire driverless-car industry.

Drafted with AI and edited by hand (claude-opus-5, reviewed 2026-08).