Happy Birthday, Sputnik! (Thanks for the Internet)
By Gary Anthes
September 24, 2007 12:00 PM ET
But around 2000, Kleinrock and other top-shelf technology researchers say, the agency, now called the Defense Advanced Research Projects Agency (DARPA), began to focus more on pragmatic, military objectives. A new administration was in power in Washington, and then 9/11 changed priorities everywhere. Observers say DARPA shifted much of its funding from long-range to shorter-term research, from universities to military contractors, and from unclassified work to secret programs.
Of government funding for IT, Kleinrock says, "our researchers are now being channeled into small science, small and incremental goals, short-term focus and small funding levels." The result, critics say, is that DARPA is much less likely today to spawn the kinds of revolutionary advances in IT that came from Licklider and his successors.
DARPA officials declined to be interviewed for this story. But Jan Walker, a spokesperson for DARPA Director Anthony Tether, said, "Dr. Tether ... does not agree. DARPA has not pulled back from long-term, high-risk, high-payoff research in IT or turned more to short-term projects." (See sidebar, DARPA's Response.)
A Shot in the Rear
David Farber, now a professor of computer science and public policy at Carnegie Mellon, was a young researcher at AT&T Bell Laboratories when Sputnik went up.
"We people in technology had a firm belief that we were leaders in science, and suddenly we got trumped," he recalls. "That was deeply disturbing. The Russians were considerably better than we thought they were, so what other fields were they good in?"
Farber says U.S. university science programs back then were weak and out of date, but higher education soon got a "shot in the rear end" via Eisenhower's ARPA. "It provided a jolt of funding," he says. "There's nothing to move academics like funding."
Farber says U.S. universities are no longer weak in science, but they are again suffering from lack of funds for long-range research.
"In the early years, ARPA was willing to fund things like artificial intelligence -- take five years and see what happens," he says. "Nobody cared whether you delivered something in six months. It was, 'Go and put forth your best effort and see if you can budge the field.' Now that's changed. It's more driven by, 'What did you do for us this year?'"
DARPA's budget calls for it to spend $414 million this year on information, communications and computing technologies, plus $483 million more on electronics, including things such as semiconductors. From 2001 to 2004, the percentage going to universities has shrunk from 39% to 21%, according the Senate Armed Services Committee. The beneficiaries have been defense contractors.
Meanwhile, funding from the National Science Foundation (NSF) for computer science and engineering -- most of it for universities -- has increased from $478 million in 2001 to $709 million this year, up 48%. But the NSF tends to fund smaller, more-focused efforts. And because contract awards are based on peer review, bidders on NSF jobs are inhibited from taking the kinds of chances that Licklider would have favored.
"At NSF, people look at your proposal and assign a grade, and if you are an outlier, chances are you won't get funded," says Victor Zue, who directs MIT's 900-person Computer Science and Artificial Intelligence Laboratory, the direct descendent of MIT's Project MAC, which was started with a $2 million ARPA grant in 1963.
"At DARPA, at least in the old days, they tended to fund people, and the program managers had tremendous latitude to say, 'I'm just going to bet on this.' At NSF, you don't bet on something."
"We are confident that anyone who attended DARPATech [in Aug. 2007] and heard the speeches given by DARPA's [managers] clearly understands that DARPA continues to be interested in high-risk, high-payoff research," says DARPA spokesperson Jan Walker.
Walker offers the following projects as examples of DARPA's current research efforts:
- Computing systems able to assimilate knowledge by being immersed in a situation
- Universal [language] translation
- Realistic agent-based societal simulation environments
- Networks that design themselves and collaborate with application services to jointly optimize performance
- Self-forming information infrastructures that automatically organize services and applications
- Routing protocols that allow computers to choose the best path for traffic, and new methods for route discovery for wide area networks
- Devices to interconnect an optically switched backbone with metropolitan-level IP networks
- Photonic communications in a microprocessor having a theoretical maximum performance of 10 TFLOPS (trillion floating-point operations per second)