Japan lags Silicon Valley in conventional internet services but could become a global leader in the deep-learning technology of artificial intelligence, a leading AI expert has said.
Yutaka Matsuo, 41, Project Associate Professor at the University of Tokyo, believes machine learning will revolutionize sectors such as image recognition.
“In deep learning, Japan still has a great chance to compete,” Matsuo said in an interview.
He singled out Sony Corp., Ricoh Co., Olympus Corp. and Canon Inc. as companies that show promise.
“Changes happening now are in image processing technology, and Japan has been good at this,” he said.
With Japan traditionally strong in robotics, Japanese companies should be able to take a lead in self-learning robots and machines in fields including farming, construction and nursing care, Matsuo said.
A graduate of the University of Tokyo, Matsuo has worked in AI research at institutions such as Stanford University.
He is now one of Japan’s leading AI researchers and has served on numerous government advisory panels involved with the sector and robotics.
Deep learning, which allows computers to recognize patterns and automatically classify images, is what triggered the current, third AI boom, he said.
Matsuo says the first was from 1956 and into the 1960s, when scientists coined the phrase.
The second boom was in the 1980s, when an early version was rolled to industry, although it was not particularly successful.
In conventional image recognition, people had to teach computers how to distinguish dogs from wolves, say, by supplying their characteristics — like defining those with folded ears as dogs and erect ears as wolves. But computers would then mistake dogs with wolf-like ears because they cannot make judgments outside the supplied definition.
“Let’s say there was a person with a brain disease. The healthy eyes send visual information to the brain, but it cannot recognize what it’s seeing. … That has been the conventional computer,” Matsuo said.
But with deep-learning technology, computers’ image recognition ability has dramatically improved and even exceeded that of humans, he said.
“Today, the reversal is happening in the ability to recognize images,” he said.
In February 2015, an AI system developed by Microsoft outperformed humans in an image-recognition test. It recorded a 4.9 percent error rate, surpassing the 5.1 percent rate of humans.
“It’s revolutionary,” said Matsuo, who believes computers will become able to perform many of the tasks currently done by humans.
For example, an image-recognition security camera can distinguish strangers from familiar faces and sound an alarm, which means security guards no longer need to watch banks of monitors, he said.
These cameras can also detect signs of possible incidents that are difficult for humans to notice — like drunken drivers or people inclined to start fights in the street.
“There are many jobs that human beings are doing only because computers haven’t had the ability,” Matsuo said, likening the advent of image-recognition to the transistor revolution.
“Before transistors were invented, people had to rely on huge vacuum tubes to amplify electronic signals. But the simple, small device allowed people to create compact computer devices, such as PCs and smartphones,” he said.
As AI technology develops, some people fear robotics will take over people’s jobs. Research by the World Economic Forum in January estimated that over 5 million jobs will be lost by 2020 as a result of progress in advanced genetics, AI, robotics and other technologies.
But Matsuo sees the question not about losing jobs to robots and AI but about redefining people’s tasks at work. He says the change would allow people to focus on more creative, communicative tasks rather than the repetitive, menial labor that computers are useful for.
“Before automated ticket gates were introduced at railway stations, employees had to stand and punch tickets by hand. But their jobs today are to guide people and to do maintenance on those machines,” he said.
“The important thing is to continue updating their skills to keep up with changes in their tasks rather than to be satisfied with skills and knowledge they acquired in the first 20 years of their lives.”
Matsuo regularly addresses audiences of businesses and academics. He was interviewed before delivering a lecture at auto parts maker Nifco Inc., the parent company of The Japan Times.