The Day of Brain-Like Computers Is Dawning
The science-fiction fantasy of a computer that works like a living brain is beginning to emerge as a reality, with applications, major government funding, market research and two publicly-traded companies attesting to the validity of the concept.
Such computers are based on a technology called neural networks because they are patterned after the interconnected nerve cells, or neurons, in the brain.
“We are trying to build a brain, and we are trying to build it out of brain-like parts,” said Bernard Widrow, a professor at Stanford University and a pioneer in the neural network field.
“It might take a thousand years to build a whole brain,” he cautioned. But he added that “in the next decade some useful applications will result.”
In fact, although most neural network applications are still in the developmental stage, at least four are currently being used. The most sophisticated is a system that evaluates mortgage loans and determines whether or not they should be approved.
“Overall (the system) is found to be much more consistent in its classifications . . . than are human underwriters,” according to Nestor Inc., the application’s developer.
Neural networks are a type of artificial intelligence. The most common artificial intelligence approach, expert systems, are programmed with a set of rules and make determinations by applying those rules.
But the system breaks down when a situation occurs for which there is no pre-learned rule, the kinds of situations living brains encounter all the time.
A brain’s billions of nerve cells are all interconnected and change with experience. With all the nerve cells working on the same problem together, even a slug can perform a simple task, like recognizing an object instantly, that would require a standard computer to go through millions of calculations.
Neural networks try to simulate those interconnections electronically so that they can learn by experience instead of being programmed. Researchers say their greatest potential at this point is object recognition. Applications have been developed to verify check signatures and identify heart disease by reading electrocardiograms.
Expects Hefty Growth
Owen Carroll, a heart researcher at the State University of New York at Stonybrook, said he believes a neural network-based heartbeat classification system would make such analysis much more widely available. “At present, a diagnosis by a highly trained specialist can cost $150 and up. We’re trying to bring this cost down to $40.”
Neural network consultant Tom Schwartz of Schwartz Associates estimates that sales of neural network modeling tools, the building blocks for applications, will grow 50% a year for the next five years, to $150 million by 1992.
“The neural network industry leaped into being over the past twelve months,” Schwartz said in a recent report. “As of June, 1987, there were less than 12 units of modeling software in the hands of users. By July, 1988, the installed base jumped to almost 10,000 units.”
Neural networks got an important seal of approval last month, and the largest infusion of research and development funds, from the Defense Advanced Research Projects Agency.
DARPA, traditionally a leader within the federal government in financing computer research, recommends that the agency spend $400 million over the next eight years on neural network technology.
It would be one of the largest research initiatives undertaken by DARPA. “More important, it will probably attract like amount of funding from the other armed service branches,” said Edward Rosenfeld, editor of the newsletter Intelligence, which follows the neural network field.
The technology is also attracting the interest of the Japanese. Japan’s Ministry of International Trade and Industry received funding requests last month for a major effort to develop neural network technologies. Rosenfeld said that although Japan lags behind the United States and Europe in this area, indications are that neural networks will be a major part of its Sixth Generation Human Frontier Project, a consortium of government and industry to develop advanced technologies.
In the United States, neural network research has started moving out of university laboratories and into corporate research departments. Consultants and analysts said about 250 companies worldwide have significant neural network projects.
Several large companies are investing in the technology, including GTE Corp., Perkin-Elmer and Ford Motor Co., but there are also about 35 to 40 start-ups in the United States. Only two, Nestor of Providence, R.I., and Excalibur Technologies Inc. of Albuquerque, N.M., are publicly traded.
“It’s not an industry yet, it’s a technology,” said Gerald Michalski, analyst with the market research firm NewScience Associates.
But the scientists involved in the technology are convinced that it will dramatically change the world of computing in the next two decades.
“I see this technology as being as important to the future as the telephone or the transistor,” said Leon Cooper, founder of Nestor and a Nobel prize winner for physics.