Finishing First with the Fifth

By Philip Elmer-DeWltt

The race is on to build fast, "intelligent" machines

In a large, sunny room on the 21st floor of a Tokyo skyscraper, Electronics Engineer Kazuhiro Fuchi, 47, has assembled a group of three dozen computer scientists who spend each workday in 1983 thinking

about 1993. Buttressed by a

$450 million commitment from the Japanese government, Fuchi and his colleagues hunch

over terminal screens and pore over mounds of studies for up to 16 hours each day. Their mission: to advance Japan one giant step in the world of computers, creating new ultrahigh-speed machines dedicated to that branch of computer science known as artificial intelligence (AI).

Until now, the pioneering work in computers was done almost exclusively by a select group of European and American scientists who shared a loosely defined mandate: to make dumb machines act as if they had human intelligence. Over the past 25 years, the AI laboratories of such institutions as M.I.T., Stanford, Carnegie-Mellon and Scotland's University of Edinburgh have introduced word processing, video games, time sharing, robot control and advanced missile-guidance systems. Lately, AI research has concentrated on building systems that can mimic the brain work of skilled experts in such fields as oil exploration, battlefield command and computer design itself. Now Japan has made it a national goal to take its place within ten years among the world leaders in the emerging knowledge industry. "We no longer need chase the more developed countries," a consortium of Japanese computer manufacturers and scholars declared in 1981. "We can fulfill our duty as an economic power [by] investing in this leading field." At the heart of their effort are the machines known as fifth-generation computers.

Traditionally, the stages in computer development have been determined by advances in the technology of the inner components of the machines. The first generation relied on vacuum tubes, the second on transistors, the third on integrated circuits and the fourth, just emerging, relies on very-large-scale integrated circuits (VLSI), with chips so compact they must be designed by another computer. The fifth generation will arrange large numbers of VLSI chips in parallel, clearing the way for dramatic breakthroughs in speed and power. More important, these computers will use their new potency not only to process mathematical data, like most computers today, but also to perform human-like reasoning: finding patterns, making assumptions, drawing inferences, reaching conclusions. Using a language called PROLOG (programming in logic), the new machines will hold intricate catalogues of knowledge that will contain the rules of thumb and mental shortcuts humans use to solve knotty problems.

Fuchi, at the center of the Japanese effort, has long chafed under what he has perceived as a lack of first-rate computer research in his country. Twenty years ago, in a fit of pique at inflexible practices in Japan's Electrotechnical Laboratory, he walked off the job for two weeks. Such rash assertions of personality are rare in Japan, and Fuchi's individualism has captured the admiration of many of his peers. Stanford Computer Science Professor Edward Feigenbaum calls Fuchi "a type almost unheard of in the East, one of those who, by force of will, can make something happen out of nothing."

Among Fuchi's fifth-generation research targets for the 1990s:

>A speech-activated typewriter, with a vocabulary of 10,000 words and the capacity to handle the voice patterns of hundreds of human speakers.

> An optical scanner that can distinguish among 100,000 pictures, sufficient to read Japanese written characters.

> An automatic translating machine with a vocabulary of 100,000 words that can translate Japanese text into other major languages with 90% accuracy (the remaining 10% to be supplied by humans).

Even in the volatile computer business, this is a revolutionary program. A machine that could achieve any one of Fuchi's goals would achieve instant market success. A family of Japanese computers capable of all these tasks would threaten U.S. dominance in its most promising technology. Stanford's Feigenbaum observes, "The U.S. is squandering its lead at the rate of one day per day."

There are signs in the U.S. that the alarm has been heard. Since Japan launched its fifth-generation project two years ago, dozens of U.S. firms, from Westinghouse to Atari, have started AI departments. Early this year the Pentagon's Defense Advanced Research Projects Agency announced that it expected to spend up to $95 million a year on "new generation" computers for military applications. IBM, which has traditionally taken a hands-off attitude toward such "blue sky" efforts, is said to have committed a 25-man team to building a fifth-generation machine. Says Raj Reddy, director of Carnegie-Mellon's Robotics Institute: "The Japanese may have awakened a sleeping giant."

For Japan, there is more at stake than market shares of the next generation of computer hardware. Having already begun the transformation from an industrial economy to one based on gathering and disseminating information, the Japanese now hope to overcome age-old barriers of language and culture with intelligent machines: automatic translation devices coupled with systems that can dispense medical advice, prepare tax returns, design buildings, repair complex electronic equipment. In their vision of the future, fifth-generation computers will be delivery vehicles for the ultimate export products: the know-how and native intelligence of the Japanese people. --By Philip Elmer-DeWltt. Reported by Thomas Levenson/Tokyo

With reporting by Thomas Levenson This file is automatically generated by a robot program, so viewer discretion is required.