On October 28, 1983, the US Defense Advanced Research Projects Agency published “Strategic Computing: New-Generation Computing Technology,” a plan to spend roughly a billion dollars over a decade building what it called “a new generation of machine intelligence technology.” The program was DARPA’s answer to Japan’s Fifth Generation project, and it explicitly tied basic AI research - expert systems, natural language, vision, speech - to concrete military payoffs.
The plan was organized around three demonstration applications, each chosen to pull the underlying technology forward. An Autonomous Land Vehicle would navigate roads and terrain on its own. A Pilot’s Associate would act as an expert co-pilot, advising a fighter pilot in real time. And a naval Battle Management System would help commanders fuse data and plan under pressure. Beneath these sat a “technology base” of AI software, advanced chip design, and parallel computer architectures - the same architectures that companies like Thinking Machines were building.
Strategic Computing ran from 1983 to about 1993 and spent on the order of a billion dollars. It funded real advances in microelectronics, parallel hardware, and computer vision, and it kept American AI labs employed through lean years. But its headline AI goals - a tank that drives itself into combat, software that reasons like a veteran pilot - proved far harder than the expert-system optimism of 1983 assumed, and the program never delivered the integrated machine intelligence it promised.
Why business readers should care: Strategic Computing is a textbook example of a large, top-down technology bet justified by spectacular end-goals. It produced durable infrastructure even though its flagship demonstrations fell short - a reminder that ambitious programs are often best judged by the capabilities they leave behind, not the promises in their launch document.