Japan edges closer to building the first 'thinking' computer

A scene in a Japanese bank circa 1990:

You walk up to a young teller and say: ''I want to withdraw 50,000 yen from my account.'' Within seconds the money is in your hand - no checkbook, cash card , or identification needed.

The teller pays up simply on recognition of face and voice because ''he'' or ''she'' is actually a robot - a glorified computer and television camera. And, if you've got enough money in your account, the robot might even be programmed to offer you a deep respectful bow, jokes Yoshiya Shinagawa, a professor of medicine specializing in studies of differences between computers and the human brain.

The gap between human and machine intelligence, he says, is definitely narrowing, as Japan gives top priority to the next ''fifth generation'' of computers: machines that can ''think'' for themselves. But these high-powered machines now on Japanese drawing boards - including the robot bank teller - also may help to lay to rest Japan's reputation as a technological copycat.

The Japanese computer industry has grown powerful by originally examining the machines of the powerful International Business Machines corporation and producing decent copies at a lower price. But the Japanese are determined to chart a major new technological territory of their own.

The Ministry of International Trade and Industry (MITI) last year launched a 10-year program to create the exotic computers that the Japanese say will bear as much resemblance to existing machines as the supersonic Concorde does to the Wright brothers' wood-and-glue flying contraption.

To do so the Japanese have to overcome what has long been their basic weakness. They are great at producing the machines themselves but poor at what makes them function: the ''software'' programs.

Until now, Japan has not been able to match the original ideas and intuitive brilliance of British and American computer programmers, for example. One reason is the traditional Japanese distrust of flashy individualistic thinking.

One way the government is trying to foster originality is by having individual companies both compete and cooperate with each other, with the support of government research institutes. In the past few months this push has paid off handsomely, with new developments and improvements being announced almost daily.

Since April, for example, the government electro-technical laboratory and several private firms have been racing to beat IBM in increasing the processing speed of computers.

By early July, Nippon Electric Company (NEC) had the lead with almost 10 trillionths of a second (compared to IBM's reported best of about 14 trillionths of a second) to perform a basic binary operation.

On July 6, meanwhile, Fujitsu announced a new scientific computer with the world's fastest processing speed - 500 million floating point arithmetic operations per second (flops, for short), compared to a top of 400 million for the Cray Company of the United States. A floating point operation requires at least several thousand elementary binary operations. NEC says it will shortly come out with an 800 million flops system.

Fujitsu's machine is designed for complex calculations involving nuclear power plant operations, the processing of data from meteorological and earth-resources exploration satellites, and the structural analysis of high-rise buildings and long bridges, company officials say. Fujitsu is also using its high-speed capabilities for a new laser printer that can deliver up to 21,200 lines of print per minute.

The problem of computers talking to each other is being tackled by Hitachi. It has come up with a system of infrared rays to link distant machines for data transfer rather than using traditional cables.

Such developments in speed and applications, say industry analysts, are vital steppingstones along the road to developing the super computer. The fact that the Japanese firms have been in the forefront of those developments is giving them confidence they can make the MITI 1990 deadline.

Sozaburo Okamatsu, one of the key government figures in the 10- year budget research and development program, says the Japanese machine as now planned will be able to hear and talk - first in a special computer language, but ultimately in human language.

By 1991 the Japanese government wants to have a working prototype that could perform such functions as automatically translating languages, converting speech into the printed word, and making decisions. Modern computers can process vast amounts of data far faster than the human brain, but they are almost incapable of using the information to form reasoned judgments. Estimates of the project's cost range from $500 million to $1 billion.