DARPA and the Decline of US Computer Science Research

The Defense Advanced Research Projects Agency (DARPA) has been something of a godfather to the computer science field. Begun in 1958 in response to Russia’s launch of Sputnik, it immediately took an interest in computer networking and underwrote the research that led to the creation of ARPANET, the Internet’s predecessor, in 1969.

The agency’s support of computer science didn’t stop there. Its funds bolstered artificial intelligence research in the 1980s, and embedded software in the 1990s. Indeed, many of the foundational technologies of the computer industry got started in DARPA-funded projects. “They invested large sums of money in high risk, high payoff projects,” recalls Leonard Kleinrock, professor of computer science at UCLA, inventor of packet-switching technology, and overseer of the first ARPANET computer link in 1969.

But changes in DARPA’s priorities have led to a serious decrease in funding for basic computer science research. In recent years, the agency has focused more on bankrolling projects with shorter completion times and concrete, applied goals, in place of the high-risk, long-term payoff projects that made DARPA so successful in its early years.

Although DARPA’s funding of computer science research has remained stable, money granted to universities has fallen sharply. The New York Times reported that DARPA’s funding of university-based research dropped from $214 million to $123 million between 2001 and 2004.

Edward Lee, chair of the electrical engineering division at the University of California at Berkeley, notes that in its early years, DARPA’s investments in software technology “created a rather thriving research community. That funding tended to draw people into an area, and a lot of (young) people shifted their work.” Then DARPA’s focus shifted, “and now they’re all scrambling for money. One consequence is they’re shifting their attention to NSF, which is getting deluged with proposals.”

The National Science Foundation (NSF), the other traditional source of funding of basic computer science research, has indeed found itself swamped with applications but with no significant increase in its research budget to offset the increased funding for computer-science research. In 2002 (the most recent year for which data is available), the agency was responsible for 86% of the federal support of IT at universities, according to Deborah Crawford, deputy assistant director of NSF’s Computer and Information Science and Engineering program. In 1992, that number was 56%. In 2000, NSF received about 3000 computer science proposals with a success rate of 30%. In 2004, the number of applications had risen to 6500 and the success rate had dropped to 16%.

Growing Importance of Computer Science Research

Crawford credits other factors with the increase in applications to NSF’s computer science division. “The growth in our proposals can’t be accounted for by just [DARPA’s cuts]. I think what you see is the growing importance of computer science research in a host of applications, and so there are more opportunities for basic research, and that results in a larger number of proposals.”

Another factor may be the dot-com crash. In the late 1990s boom, companies had plenty of cash to pour into basic research, but when times got tight, many companies pared back these long-term efforts to focus on survival. Lucent, AT&T, and others are shadows of their former selves. IBM remains strong, but even its labs do not enjoy the level of funding that they had a few years ago.

That trend is reflected in Ph.D. jobs. According to the Computing Research Association’s Taulbee Survey, throughout the 1990s, about half of new PhDs took industry jobs, with the rest going to academia. But in 2000-2001, that changed. Industry jobs nosedived to about 30% in 2002-2003, while the percentage going to academia rose to 70%. As a result, more young faculty are competing for meager funds. DARPA’s flagging support of basic computer science research exacerbates the problem.

And the pressure isn’t only financial. DARPA has required an increasing percentage of its projects to be classified and sometimes places restrictions on citizenship of its funding recipients, making it more difficult for investigators to support the majority of CS graduate students who are not U.S. citizens or permanent residents. According to a 2004 survey by the Computing Research Association, slightly more than 50% of computer science graduate students are foreign. “[Citizenship requirements are] all of a sudden [preventing] a lot of the strongest students from getting financial aid, some of whom might not be well suited to teach”–and hence to earn their living as teaching assistants–“for language or cultural reasons,” says William Aspray, professor of informatics at Indiana University. “My guess is that this going to be somewhat harmful to the research lives of these departments, because faculty members rely heavily on a cadre of graduate students to get projects done.”

“If they want to classify research, that’s fine,” says Kleinrock. “But it doesn’t open it up to the kind of talent that we want to draw to this country.”

Big Impact

In the late 1990s, the dot-com bubble made it easier for faculty to get industry support, and NSF could give small awards to single investigators, confident that they could use them to leverage additional support. NSF now tends to give out “more enabling awards so that an investigator is getting funded at a level that allows him or her to do all of the proposed work,” says Crawford.

Still, as Lee points out, these smaller, enabling awards aren’t the only way of getting a young career started. Multi-investigator awards and larger projects can be good for young faculty, Lee says. “They divert a lot of funding to young faculty and incubate their work.”

Getting By

Forging relationships may represent the best chance for early career scientists to thrive in today’s uncertain funding environment, but this is hardly an ideal approach. With less money to spare, funding agencies like NSF tend to get more conservative and hand out dollars to projects they feel are most likely to succeed, according to Spafford. A senior researcher’s track record is more likely to be the factor that convinces a reviewer.

Yet helping a senior scientist get a grant may not be the best way to earn a reputation as an experienced, independent scientist, especially in a field like computer science that has such a strong lone-ranger character and history. “Computer science tends to be a very solitary field. With collaborative work, it’s not always clear how much can be attributed to a junior colleague. In some environments that’s a danger, in others it’s not a problem,” says Spafford.

Another approach is to help ensure the success of your own grant proposals, and a good way to do that–apart from writing very strong proposals–is to get involved in the review process. “NSF sometimes has trouble finding qualified reviewers,” says Spafford. This is especially true in times like these when the ratio of proposals to funded scientists is low. “By looking at a number of proposals, some successful and some not, you can learn a lot about how to structure your own proposals,” he says.

Asprey agrees. “You know so much more about how to craft your own proposals by the time you’re finished.” Extending the trip to Washington by a day or two to visit program managers at various agencies can also pay off. “Find out what programs available, what programs are coming up, and float some ideas by them to see what they think,” says Spafford.

These strategies and others could help steer you through difficult times until times turn better, as they generally do sooner or later. Lee sees the current funding crisis as a product of the political climate, which could change with a new administration. And even at DARPA, the news isn’t entirely bad. Some areas of basic science research continue to see solid funding, including cybersecurity and nanotechnology, as well as imaging and ad hoc networking, according to Kleinrock and Spafford. But that narrower focus troubles some computer science researchers, and NSF is not likely to be in a position to pick up the slack. Crawford projects that the agency’s computer science budget will increase by 1%-2% annually over the next few years.

Rumblings in Congress indicate that lawmakers may be paying attention to complaints about waning funds. At a 12 May hearing, the House Science Committee heard testimony about DARPA’s changes in computer science funding. Agency director Anthony Tether defended DARPA’s funding practices, maintaining that the reductions in computer science funding were caused by an increased focus on multidisciplinary work. Representative Sherman Boehlert (R-NY), who presided over the hearing, expressed concern over the changes, insisting that basic computer science research is the key to long-term information technology superiority.

So while there may be some hope for increases in federal funds, nothing is imminent. Lee is ambivalent, accusing Tether of playing with numbers to support his case. “I was not encouraged,” he says. But Congress’ interest was “a positive sign.”

Ask Dr Bridget for Advice

NIH Panel Weighs New Grant for Young Investigators