As part of this report from the digital frontier, BUSINESS WEEK set out to discover how leading computer-science researchers rate the work being done at the world's university and industrial laboratories.
During May, we conducted an informal poll by E-mail. Exchanges with a dozen researchers and several heads of computer-science departments helped us assemble a list of leading lab candidates. Neither that list nor the researchers who voted were selected scientifically, so the results cannot be considered statistically valid. They represent only the collective opinion of the respondents.
GAINED GROUND. Still, the results attracted considerable interest from those who were polled. Randy H. Katz, head of the Computer Science Div. of the University of California at Berkeley, obviously scrutinized the initial results, because he later thanked us for showing that Berkeley has gained ground on the traditional leading labs: Stanford University, Massachusetts Institute of Technology, and Carnegie Mellon. And several researchers said they took time to vote just to see the results of our final "X-Lab" tabulation, summarized on this page. The X-Lab section was where we asked researchers to name the lab where they would most like to work.
The poll was conducted in two rounds. Each time, we asked researchers to rank labs in two groups--the top 5 and the top 10. To gauge the spread of information-technology research, first-round respondents were asked to designate the top 5 and top 10 both currently and in the late 1970s. The first ballot went to about 200 researchers. More than 80 sent it back. Many suggested additional labs to be included on the second-round ballot.
GALLIC PRIDE. Many participants expressed strong feelings about the poll during the first round. Michael P. Fourman, head of Informatics at the University of Edinburgh, thanked us for including Edinburgh in the artificial-intelligence section, but he added: "I think you underestimate us!" Fourman went on to make such a convincing case for his operation that we dispatched a reporter to Scotland to profile it.
When the write-in names were added to the second-round ballot, we also dropped several first-round labs--those receiving the fewest votes. That prompted protests when the second-round ballots went out. A researcher from one university that had been dropped fired back a warning that, because certain well-regarded names were missing, the poll was dangerously flawed. A French researcher sent a message railing against English speakers' domination of technology in particular and language in general. Among other things, he criticized us for not sending a French version to French researchers. He had taken the trouble to analyze the labs on the ballot, and he determined that the list of labs was "86% d'anglophones et 13% de non-anglophones." With such a sampling, he lamented, the results were a foregone conclusion.
Most complaints related to the categories on the ballot. We grouped laboratories under four broad areas: (1) computer science; (2) telecommunications, networking, and groupware; (3) artificial intelligence, robotics, speech recognition, data mining, and interfaces; and (4) biologically inspired areas such as artificial life, genetic algorithms, and evolutionary programming. Several researchers threw up their hands--judging from their notes. "Just wanted you to know that I'm not participating," said one, "because I couldn't begin to rank the labs under such mishmash headings."
Second-round ballots went to nearly 800 universities and corporate labs. Roughly 200 second-round ballots were returned, and they were overwhelmingly from researchers at U.S. labs, so the results are highly colored by an American perspective. Only two dozen replies came from Europe, and even fewer arrived from Asia. Researchers in Europe and Asia largely agreed that U.S. labs dominate the top five spots. But most foreign voters included at least one overseas lab among their top five selections--and even more non-U.S. labs made their top-10 picks. These labs, however, were pushed much lower in the standings, even within the four categories above, by the heavy U.S. turnout.
While the results may not fairly represent opinions outside the U.S., we believe the findings do approximate the consensus in the U.S. Detailed results, including breakdowns within the four groupings, are available in the four related tables with this story.