Research Revolution

Deep Technical Research, Too
Not surprisingly, many of those peo-ple want to flex their intellects in areas deeper than consumer software. Microsoft's Live Labs includes Jennifer Chayes, a theoretical mathematician, and database big thinker Jim Gray.

Google researchers are working with NASA's Ames Research Center in Silicon Valley on an eclectic mix of projects, including distributed programming and supercomputing. One Google researcher runs a genetic algorithm on a NASA supercomputer that, at 51.9 teraflops, is the world's fourth fastest. All of Google's software today is highly distributed, breaking jobs into pieces that run on thousands of servers. But the NASA deal could show the company whether future computing problems could benefit from a different approach, says Norvig, who worked at Ames before landing at Google four years ago.

Google also has ambitions to become an application service provider to scientists, renting out its massive server farm for professors to run jobs on. So far, concerns about data security, privacy, and whether the company can spare any cycles have held that up, Norvig says. Professors constantly send in requests to run programs on Google's machines, he says, "and the best we can say is, 'Send us a grad student for the summer.' We'd like to do better."

Sun also is turning to Web research to improve its products. The company is funding Berkeley's Reliable, Adaptive, and Distributed Systems lab to apply a branch of AI called machine learning, which uses algorithms that improve with experience, to design Internet server farms that perform well under tremendous floods of traffic and memory bottlenecks. One goal is to design software that can interact with the millions of Sun systems and software components at customer sites, diagnose their problems, and keep them in tune. Another is to incubate a Web company that could build its business on Sparc and Solaris. "AI in some sense is going to creep back up on us," Sun CTO Greg Papadopoulos says. "That's one of the quiet, big-bet things we're doing."

Big bets are the currency of the Web, and they're what could make Google research into the next IBM Watson lab or Xerox PARC. That's if Google will have it. "I don't think we want to be like that," Norvig says. "Those are certainly icons and important places, but they were also known for not being as productive as they could given the caliber of people they had. ... We're still operating as a startup company and trying to keep the culture that way."

If Google and others can pull it off, they just might reinvent both computer science research and the Web.

--With Thomas Claburn