The advertisements for a current book on Japanese computer research are ominous. “The next generation of computers will think and reason,” according to the promotions for The Fifth Generation. “And they will speak Japanese.” The tone reflects the growing international recognition that the first nation to perfect machines that mimic human thought will make all existing computers obsolete. With the world computer-based information processing industry worth roughly $175 billion annually, the winner may also hold the balance of global economic power.
The reason for the current alarm among computer scientists, particularly in the United States, is the 1981 announcement by the Japanese government that a consortium of major computer companies and public research laboratories will spend $450 million over 10 years to build a new so-called fifth generation (5G) computer—a machine that can manipulate concepts faster than current computers can process numbers. Such machines would be far more complex than today’s enormous high-speed filing systems; theoretically they will be capable of independent thought. In The Fifth Generation,
coauthor and Stanford University professor of computer science Edward Feigenbaum jingoistically urges the United States to match the Japanese effort. Evoking the Japanese attack on Pearl Harbor in the Second World War,
he warns, “At dawn we slept.”
The Japanese 5G project, headed by University of Tokyo professor of electrical engineering Tohru Moto-oka, is specifically aimed at building machines that can think and communicate as well as—or better than—man. The science used in developing such a machine has been dubbed artificial intelligence (Aí). After tapping the best Aí minds in the West throughout the 1970s, the Japanese have set astonishing goals for their proposed 5G computers.The machines will be designed to:
• obey written or spoken commands in several languages and provide translations on demand
• store and retrieve vast libraries of information under a complete range of cross references
• reason logically and draw inferences;
• write computer programs without human supervision
• solve hundreds of interconnected problems simultaneously rather than consecutively.
Applications of 5G technology remain theoretical. But researchers speculate that the machines might one day use the principles of biotechnology to create new life forms. And the Pentagon hopes that they can make battlefield decisions in future high-tech warfare.
In order to break through to the 5G level, researchers believe that they must overcome the limitations of current computers that can only process information one piece at a time. They theorize that by building more powerful computers or groups of computers and programming them to “think” about several things at once, they can more closely approximate the workings of the human brain.
The success of 5G and similar projects still depends on giant leaps forward in semiconductor technology. The four generations preceding the 5G represent the various stages of computer hardware evolution, starting with vacuum tubes used in the first computer, the Electronic Numerical Integrator and Calculator (ENIAC), developed at the University of Pennsylvania in 1946. Transistors arrived in 1947. In the late 1950s scientists developed a revolutionary process that allowed transistors to be compressed onto one tiny silicon “chip” to create an integrated circuit. The fourth-generation machines currently available are made up of largescale integrated circuits with a maximum of 500,000 transistors on a chip three centimetres square. But the 5G project aims at very large-scale integrated circuits with 10 million transistors per chip, which would increase computing speed 10,000fold.
The determination of Japan’s ministry of international trade and industry to have its 5G project succeed has roused other nations from their torpor. In Britain a department of industry committee recommended last year that $570 million be spent in the next five years on advanced computer research. In the United States, Aí has been largely funded by the Advanced Research Products Agency at the Pentagon, which recently rated 5G computers and AI as key elements in maintaining the West’s superior military technology. But now 16 U.S. electronics companies, ranging from Control Data Corp. to Uni vac, have braved stiff antitrust laws to form their own consortium—the Microelectronics and Computer Technology Corp.—with the aim of pooling innovative research and development resources. In Canada, on the other hand, a highly praised computer translation project at the University of Montreal had its funding cut off by the Treasury Board in 1981. This year all the federal government could muster was a Janu-
ary workshop on artificial intelligence sponsored by the Science Council.
Canada’s early effort was not the only Aí research to bear fruit, however. Feigenbaum has been a pioneer in so-called “expert systems” programs. These relatively simple Aí programs capitalize on human expert knowledge in specific fields. One successful example is PROSPECTOR, developed by SRI International, an industrial laboratory allied with Stanford University and Vancouver geologist Victor Hollister. PROSPECTOR questions a geologist exploring a spe-
cific site about the composition and configuration of rock samples. The questions become more and more detailed until the program delivers an opinion on whether the mineral sought might be present. A similar diagnostic program for doctors has been developed at the University of Pittsburgh. And Alan Mackworth, director of the laboratory for computational vision at the University of British Columbia, is working on MAPSEE, a program that interprets hand-drawn sketch maps for use with remote-sensing satellite data.
Mackworth and other Canadian AI investigators like Zenon Pylyshyn, director of the multidisciplinary centre for cognitive science at London’s University of Western Ontario, carry out AI research that is similar to work in the United States. But they believe the Canadian government has been slow to climb onto the Aí bandwagon. “We are graduating very competent people who go off to the United States,” says Pylyshyn. “Aí has great technological potential, but Canada could end up a Third World country in this respect.” However, Gilles Julien, executive director of the federally funded Natural Sciences and Engineering Research Council (NSERC), says, “There are not enough applications from the universities in the area of 5G computers.” For their part, the universities claim that NSERC has discouraged applications because Aí does not fit its regular funding categories.
Nevertheless, the University of Montreal computer translation project, set up in 1965, was a world first. Its METEO program, which electronically translates English weather reports into French, still operates in Dorval, outside Montreal, and attracted several teams of Japanese researchers who visited the project throughout the late 1970s. Although the university’s research project was cancelled because it was not cost effective, it, or a similar project, may be revived as the demand for AI increases, according to Gregg Gauld of the federal government’s Translation Bureau.
Although increasing veils of secrecy make evaluation of the research difficult, AI professionals believe that the Japanese 5G I has a good chance of achieving I its stated goal of a machine that u thinks like a human. “Nothing the brain does seems particularly mysterious or impossible to duplicate,” says Marvin Minsky, a leading U.S. Aí researcher at the Massachusetts Institute of Technology and a professor of electrical engineering and computer science. In fact, he goes even further. Minsky and other scientists firmly believe in machines that will be vastly superior in intelligence to humans. “If we are lucky, they might decide to keep us as pets,” says Minsky. It is a joke that HAL, the murderous rebel computer in Stanley Kubrick’s 2001: A Space Odyssey, would have enjoyed.
The story you want is part of the Maclean’s Archives. To access it, log in here or sign up for your free 30-day trial.
Experience anything and everything Maclean's has ever published — over 3,500 issues and 150,000 articles, images and advertisements — since 1905. Browse on your own, or explore our curated collections and timely recommendations.WATCH THIS VIDEO for highlights of everything the Maclean's Archives has to offer.