Strategic Computing Initiative

• 1. Introduction

  ›

  The Strategic Computing Initiative set itself one of the most ambitious goals in the history of technology: a machine that could run ten billion instructions per second, and see, hear, speak, and think like a human. That was not a science fiction premise. It was the official target of a United States Department of Defense program that ran from 1983 to 1993 and cost one billion dollars. The question the program left behind is not simply whether it succeeded or failed. The deeper question is what happens when a government decides to engineer human-level machine intelligence on a deadline, and who gets to decide when the clock runs out.

• 2. The Sputnik Moment

  ›

  Japan's fifth generation computer project was the spark. The Japanese government had set aside billions for research into computing and artificial intelligence, and Washington read that investment the way it had once read the Soviet satellite launch in 1957: as a direct challenge to American technological leadership. The British government responded with its own program, known as Alvey. A consortium of American companies launched the Microelectronics and Computer Technology Corporation. The United States government launched the Strategic Computing Initiative. Each of these programs was, in its way, a defensive posture dressed up as a scientific ambition. The SCI's designers looked explicitly to the Apollo moon program as a structural model, envisioning different companies and academic groups building separate subsystems that would eventually be woven into a single integrated whole. Alex Roland and Philip Shiman described it as boasting not mere tactics or strategy, but "grand strategy, a master plan for an entire campaign."

• 3. Architecture of a Dream

  ›

  The Defense Advanced Research Projects Agency funded the initiative, and its Information Processing Technology Office directed it. Robert Kahn, who led IPTO in the program's early years, gave the project its founding momentum. Clint Kelly managed the SC Initiative for three years and developed many of its specific application programs, including the Autonomous Land Vehicle. By 1985, the program had spent $100 million and supported 92 separate projects spread across 60 institutions. Half of those were in industry; the other half were in universities and government laboratories. The degree of machine integration the program was chasing, according to Roland and Shiman, would "rival that achieved by the human brain, the most complex instrument known to man."

• 4. The Reckoning

  ›

  By the late 1980s, the horizon had not come any closer. Insiders pointed to failures of integration, organization, and communication as the structural problems eating away at the program. When Jack Schwarz took over the leadership of IPTO in 1987, he made a sharp break. He cut funding to artificial intelligence research, the software component of the initiative, in a manner that the writer Pamela McCorduck described as "deeply and brutally" and as "eviscerating" the program. Schwarz believed that DARPA's money should follow only the most promising technologies. He put it in plain terms: DARPA should "surf", not "dog paddle", and he did not believe AI was "the next wave."

• 5. Concrete Wins

  ›

  The Autonomous Land Vehicle program did not deliver a thinking machine, but it did something arguably more durable. Working alongside the Navlab project at Carnegie Melnie University, it built the scientific and technical foundation for autonomous navigation. The methods it pioneered, combining video cameras with laser scanners and inertial navigation units, form the basis of almost all commercial driverless car development today. The ALV became Demo I in a lineage that included the Demo II and III programs, as well as Perceptor and the DARPA Grand Challenge. On the software side, a program called DART, the Dynamic Analysis and Replanning Tool, used artificial intelligence techniques to handle military logistics. Introduced in 1991, DART proved its value during Desert Storm. By 1995, the savings it generated for the Department of Defense were calculated to equal all the money DARPA had channeled into AI research over the previous thirty years combined.

• 6. What Came Next

  ›

  The Strategic Computing Initiative gave way in the 1990s to the Accelerated Strategic Computing Initiative, and then to the Advanced Simulation and Computing Program. Neither successor carried artificial general intelligence as a goal. Both turned instead toward supercomputing for large-scale simulation, including atomic bomb simulations. The SCI of the 1980s is a distinct program from the 2015 National Strategic Computing Initiative, and the two are unrelated despite sharing similar names. The DART program's payoff, calculated in 1995 as offsetting thirty years of prior DARPA AI investment, stands as the clearest single measure of what the initiative actually returned for its billion-dollar outlay.