Metadata: Speaking Of IBM

IBM's T.J. Watson Research Center in Yorktown Heights, N.Y., was built in 1961 as the ultimate bastion of pure corporate research. Its architect, Eero Saarinen, laid it out as a perfect 1,090-foot glass and fieldstone quarter-circle, a gentle arc that never quite seems to end when you walk its main corridor. But it is an incomplete shape, meant to suggest there is always unfinished business to pursue.

Visiting there is a healthy reminder--amid Detroit's woes and hand-wringing about Asian economic dominance--that a U.S. institution (populated, it must be said, by a lot of emigrés) can compete vigorously in innovation and the sciences.

IBM (nyse: IBM - news - people ) runs the biggest private research operation in the world--with an estimated $6 billion budget, 3,000 people and eight labs in all--and, year after year, accumulates far more patents than its rivals. The Watson Center remains IBM's locus of work on speech recognition, nanophotonics, high-performance computing, IT services research and the social Web. (See: "IBM's Infinite Research Problem.")

While Google (nasdaq: GOOG - news - people ) scales back its beloved plan to let all employees tinker on side projects with 20% of their time, IBM will continue to commit one-third of its research to what its head of worldwide operations, Mark Deane, calls "exploratory" work.

That work includes radical ideas like "cognitive computing." In November, Dharmendra Modha of IBM's Almaden Research Center in San José, Calif., won a $5 million grant from the Defense Department to lead a group of 10 IBM scientists and seven outside researchers to come up with a brain-like computer that can deal with ambiguity, continuously learn and make split-second decisions based on constantly changing data.

Despite all the fancy supercomputers and smartphones we now have, computers still compute using the same model John Von Neumann wrote up 63 years ago, with data and memory mapped together in the same address space. It's a brittle architecture for handling the swarms of data flooding market trading floors, the Internet and machines laden with actuators and sensors. That is where computing needs to go.

Some of the work will stem from IBM's decades of experience in artificial intelligence, which also informs its work on speech recognition and machine translation. Millions of mobile phones and cars on the market now use voice commands to make information requests from servers on the Web. Honda (nyse: HMC - news - people ), Acura and General Motors' (nyse: GM - news - people )Onstar system all use IBM software for voice-activated controls, as does Vlingo, a voice-search start-up that powers Yahoo!'s (nasdaq: YHOO - news - people )mobile search on the BlackBerry and iPhone.