One answer is what IBM calls “deep computing.” The concept comes from Deep Blue, whose 1997 rematch victory over Kasparov resulted not only from massive computing power, but from dramatically improved chess-playing algorithms as well. It’s that combination of power and approach, applied to vast amounts of data, that defines deep computing.
For IBM, the challenge goes far beyond fun and games. So this May, the company launched its Deep Computing Institute, an umbrella organization that pulls together 100-odd researchers at the Watson, Haifa, Tokyo and Almaden labs. By focusing these distributed talents in computing theory, statistics, computational biology, financial mathematics and data mining-expertise previously turned largely on scientific challenges-IBM wants to solve complex business problems.
One ambitious project under way this spring aims to combine Big Blue’s supercomputer weather-prediction capabilities with energy and financial modeling to help utilities meet power demands more efficiently-even to the point of buying and selling excess capacity on spot markets.
It’s a deep computing problem if ever there was one. IBM isn’t known for weather forecasting. But Deep Thunder-an SP computer outfitted with 3-D graphics and powerful modeling algorithms-can often predict weather in localized areas with greater precision than anything available through government or commercial channels. The system debuted at the 1996 Olympics in Atlanta, where it aided the scheduling of sailing and other weather-dependent events-and helped save the closing ceremonies by predicting that a powerful storm would stay 10 miles from Olympic stadium.
Because weather is central to determining energy demand, even conventional weather forecasts can help utilities use their generators more efficiently. In a program with a midwestern utility that ended last year, IBM researcher Samer Takriti helped develop proprietary algorithms he says can trim 3 percent to 5 percent off generating costs-enough to save the typical utility upwards of $40 million annually. Add in Deep Thunder’s capabilities, Takriti reasons, and the model could get even better; he’s now working with Thunder researchers to do just that. At the same time, recent industry deregulation is spawning volatile energy markets-complete with futures, options and hedging strategies. So Takriti hopes to capitalize on IBM’s years of modeling supply-and-demand dynamics and currency trading to create a system that accurately forecasts energy prices. The company believes the fusion of weather prediction and financial modeling could be applied to agricultural industries, property insurance and other fields.
Even as IBM redirects deep-computing resources from science to business, another vein of research focuses on mining the business of science. An especially hot area is computational biology, which involves analyzing huge databases of biological information to unearth patterns that might point the way toward a new drug or a better crop. IBM dived into this blossoming field in the early 1990s and now has some two dozen researchers assigned to it, with scores more in related areas. One key hire was Barry Robson, who before joining IBM helped launch several computational-biology ventures and conceived a computer system instrumental in creating a test for mad cow disease. Now strategic adviser to IBM’s Computational Biology Center, Robson says that through clinical drug trials and decades of genetics research, scientists have amassed stockpiles of biological data that today’s powerful computers and sophisticated algorithms can finally begin to decipher. “My passion is to really see it become an everyday applied discipline,” he adds.
IBM’s biggest publicly announced effort-an agricultural genetics program launched with Monsanto in January 1998-revolves around Teiresias, an algorithm that can scour vast protein or gene sequences to find repeated patterns that might code for similar functions in different molecules. Monsanto hopes Teiresias, named after a blind seer in Greek mythology, will hunt through its proprietary databases and speed the identity of genes responsible for improved yields, higher nutritional content or pest resistance.
The initial deal concluded this spring. Under a new agreement that runs through 1999, IBM will also use the algorithm to look for patterns in the far larger public databases maintained by the National Institutes of Health and other government organizations. Beyond the discovery of new genes, the company hopes Teiresias will turn up unsuspected commonalities across protein families that will enable scientists to design drugs to attack a wide variety of ailments.