Japan has a long of history of friendliness towards robots, and it maintains a global lead in R&D on artificial intelligence. This infatuation, though, could have serious drawbacks.
FORTUNE — Noriko Arai is not the first woman to spotlight the hazard that boys and their toys can present, and she won’t be the last. Her concerns may well be merited. Japan, and possibly all industrialized societies, says this mathematician, is about to reap the unintended consequences of a mostly, but not exclusively, Japanese male obsession with creating artificial intelligent life.
She claims Japan obtained a lead in the field after the U.S. pulled back on research efforts in the 1980s. “The Japanese couldn’t give up the dream of AI because of this persistent dream of humanoid robots,” says Arai, who works as a professor of mathematics at Japan’s National Institute of Informatics (NII). “There are lots of intelligent youngsters interested in AI here, and AI scientists, unlike in Europe or the U.S., are highly esteemed.”
Japan has a long of history of friendliness towards robots. Yes, the Japanese cultural imagination has yielded its fair share of ruthless Terminator-style sci-fi robots, but many of them are human-like, friendly helpers like the comic character Doraemon. Japan also maintains a global lead in R&D on robotics. “Japan has successfully maintained its high patents rate and achieved rapid growth in the areas of speech recognition,” says a report published recently on the subject by the British-Embassy-in-Japan’s chief science officer.
Arai devoted a year to study the economic impact of artificial intelligence. She published a book in Japan in 2010 on AI’s effects on office jobs titled simply “How Computers Can Take Our Jobs.” “I wanted to know if we can get more jobs from AI [and] how much white-collar jobs will be replaced by machines in the next 10 to 20 years. If it is just 10% of such jobs, it will be a catastrophe,” hinting that she expects the economic impact of AI on employment to be much larger than that. Arai also considered the ethical consequences of having machines and algorithms replace humans in the workplace — as they have on the trading floor — without really understanding how they work.
As market flash crashes have shown, Arai says AI machine learning is beyond human understanding and reasoning. “As with flash crashes, we will always have 20% incorrectness, which will harm our society. We can’t understand machines and they don’t understand us,” she says. It is foolish to believe such machines will think like humans, she adds. Ultimately, we can’t work out why they come up with correct answers or wrong answers.
Just because we program a machine using logic, that doesn’t mean a machine will always follow logical rules to reach their outcomes. “In the case of AI machines, they are not like factory robots whose actions can be precisely explained by physics and science. This is a serious ethical problem,” she points out.
Arai launched a project within NII to see just how far AI and machine learning could go in the near future. The aim is to build a robot that could pass the notoriously hard Tokyo University entrance exam. Known now as the Todai Robot project — Tokyo University goes by the name “Todai” in Japanese — Arai kicked off the first experiments two years ago, integrating the latest AI know-how to crack seemingly insurmountable problems. Unlike the general entrance exam in Japan, the rigorous Todai exam includes written essays. Her task has been eased, she says, by the respect AI receives in Japan.
Over the past few decades, American AI researchers have pursued specialized projects involving finance, military applications, and speech recognition. Scientists in Japan, on the other hand, have been more interested in solving general problems such as how to get a humanoid robot to make tea or solve a crossword puzzle. “It’s a sort of mad scientist dream … pursued mostly by many in Japan and some Europeans,” Arai says.
With its laissez-faire attitude to robots and a dire need to fill jobs without resorting to immigration as the working population ages, Japan seems tailor-made for AI. Nissan’s autonomous self-drive vehicle has just been granted a drivers license in Japan while the U.S. is still wondering how to legislate for the self-driving Google GOOG car. Japan may soon see its first robot taxis and automated trucks.
Having software that can pass Japanese university entrance exams will show just what AI is capable of doing and should serve as a wake-up-call, says Arai. “Such problems may be raised by the media but politicians and economists don’t want to think seriously about it. They don’t anticipate the qualitative risks AI may cause. They tend to think AI may create more jobs.”
Arai claims Japan will develop software that will easily pass the less difficult standard Japanese university entrance exam by 2016. This multiple-choice test, taken by half a million students every year, is given by university exam boards all over Japan. The threshold to pass this test is 60%. Students must attain at least an 80% score on this test to qualify to take the Todai exam.
IBM Japan IBM is also trying out its supercomputer on Todai test questions, but NII is leading the charge so far. “Our AI now gets 50% correct, which is good enough to pass. Last year, we got four out of six questions right. In 10 years, it will compete well with Todai 19-year-old students,” says Arai.