The United States’ autonomous reconnaissance vehicles, AI pilots, and intelligent combat systems it is developing today were already written into a decade-long plan as early as 1983.

Back then, the plan was drafted by DARPA—the Defense Advanced Research Projects Agency under the U.S. Department of Defense—which specifically funds high-risk, frontier technologies that are difficult to realize in the short term.
In this 110-page “Strategic Computing” plan, DARPA prepared to build three systems:
A reconnaissance vehicle capable of autonomously traversing complex terrain;
An AI co-pilot trained by pilots, able to identify threats, plan missions, and provide tactical advice;
A naval combat management system that can assess enemy intentions and simulate the course of a battle.
The supporting technologies include machine vision, speech recognition, natural language, expert systems, parallel computing, and microelectronics.
If you swap these terms with today’s phrasing, they roughly correspond to multimodal systems, Agents, autonomous driving, and AI chips. Reading the entire document, it really does feel like it was written just last year.
It’s just that the computing power in 1983 couldn’t keep up with this roadmap.
The plan required the unmanned vehicle to reach 10 kilometers per hour on public roads by 1985, and to increase that to 80 kilometers per hour by 1990. But when 1985 arrived, it had driven autonomously for only 1 kilometer, with an average speed of just 3 kilometers per hour.
The ten-year timeline wasn’t met, yet the technologies in the plan gradually came to fruition over the following forty years.
At the time, the U.S. was defending against Japan’s “fifth-generation computers” moving first. Today, the list of competitors has changed to Chinese models, chips, and robots.
AI plans that were not completed during the Cold War are now being checked off one by one in a new round of technological competition.