|

Will AI become a threat to future white-collar workers?

by

Staff Writer

Can a robot get into the University of Tokyo, Japan’s most prestigious university, by 2021?

Some people may say “yes,” thinking that since robots can now beat professional chess players and shogi (Japanese chess) players, it shouldn’t be so difficult to realize this in the not-so-distant future. But it is not that simple.

To answer this question, the Tokyo-based National Institute of Informatics (NII), Japan’s only general academic research institute for informatics, launched a 10-year project in 2011. Collaborating with various companies and institutions, the project aims to attain a high score on the National Center Test for University Admissions, a standardized, multiple-choice test taken by more than half a million high school graduates every year, by 2016, and achieve the level to enter the University of Tokyo in 2021.

But its ultimate goal is not simply creating an artificial intelligence (AI) to answer questions on university entrance exams. It is to raise the accuracy of AI techniques that have been accumulated to date and to deepen the understanding of what it means to be human, according to Noriko Arai, director of the project at NII.

“Back in 2010, I wrote the book ‘How computers can take our jobs’ and predicted that 50 percent of white-collar jobs will be lost due to computers in 30 years. But not many Japanese at the time took my prediction seriously enough,” Arai said in a recent interview with The Japan Times. “Since university entrance exams serve as a kind of gateway to becoming a white-collar worker, I thought people would pay more attention if AI outperforms real students on the tests.”

Japan has led the world in the area of robot technology with Honda Motor Co.’s ASIMO, a humanoid robot that can walk and dance and Softbank Corp.’s Pepper, which communicates with people by analyzing facial expressions and voice patterns, to name a couple.

But Arai, a professor of mathematics, said it is wrong to think these robots can act independently as ASIMO’s dancing ability and Pepper’s communication skills have been programed in advance to respond to different situations.

“To develop AI with interdisciplinary knowledge that can answer various questions, we should have a common benchmark that can measure the ability of AI,” she said. “By having AI solve university exam questions, which require different tasks, we will be able to know clearly how much their ability has developed so far.”

The business sector is actively involved in this project.

Companies in the education industry such as textbook publisher Tokyo Shoseki Co., Benesse Corp. and cram school operator Sapix Yozemi Group participate in the project as data suppliers. Software firm Cybernet Systems Co., and various research institutes like Fujitsu Laboratories and NTT Communications Science Laboratories, are also involved with the project.

Arai said as the national standardized tests cover eight subjects — two social studies, two sciences, Japanese, English and two mathematics, each subject provides those companies opportunities to explore technologies that can provide ideas for future businesses and products.

For example, education firms like Benesse and cram schools may be able to speed up the process of grading their tests if the process can be automated and screened by AI.

Meanwhile, mobile phone companies hope to take away technology that can help improve voice-agent apps, such as Shabette Concier or Siri, which answer verbal questions using information culled from various sources, including Wikipedia.

According to Arai, a fill-in-the-blank question on an English test can be useful for such research.

For example, if a person in a conversation says, “It’s unbelievable that we graduated from school 10 years ago,” what should the proper reply be? The right answer may be: “True. I feel like it was just yesterday,” but it is often difficult for AI to fill in the blank with such a sophisticated answer.

After taking mock exams in 2013 and 2014, the AI achieved a score on par with the average Japanese high school graduate. The results mean that AI would have about an 80 percent chance of getting into about 80 percent of Japan’s private universities.

Arai is confident AI can reach a higher score, perhaps eventually outperforming more than two-thirds of students who take the test, but admitted attaining a score good enough to be accepted by the University of Tokyo would be difficult.

Similar projects to use university entrance exams as a benchmark are also underway outside Japan.

In the U.S., the Allen Institute for Artificial Intelligence also started Project Aristo, which has a system that acquires and stores vast amounts of knowledge in data form, then applies this knowledge to answer a variety of science questions from standardized exams.

This year, according to Arai, the Chinese government also launched a similar three-year project jointly with iFlyTek Co., a company that specializes in voice-recognition systems.

Thanks to these research projects, AI may outperform humans in many areas in the near future. But it does not necessarily mean white-collar workers have a dim future with no hope to remain in the market.

Arai cited bank loans as an example. A decision to provide an auto or housing loan to a consumer does not require difficult judgment and can be processed by computers using big data, but providing loans to venture firms is a different story, she said.

“If a small venture company has developed an innovative product that does not exist in the market, bankers must judge how valuable it is and how society will accept it and whether it will make money. Big data can’t provide the answers to such questions, and such decisions will remain in the hands of people,” she said.

Likewise, the job of pharmacists can be partly done by AI, but people are needed for those aspects requiring high levels of knowledge and judgement.

Arai warns that world leaders must foresee such a future society and form national strategies accordingly.

“I think global leaders such as those who gather at Davos or OECD meetings should draw up a future scenario for their countries and for the world and discuss what needs to be done to minimize damage to society, to raise productivity and to redistribute benefits gained from increased productivity,” she said.

In the future, Arai predicts workers will be divided roughly into two categories based on what they do. The first one will be low-income earners who assist and fix operations done by not-yet-perfect AI, while the second group will be high-income earners who make the important decisions that computers cannot make.

“Those who work above AI and make high-level decisions will be the future white-collar workers, but they need extremely high-levels of education. They have to acquire abilities to make difficult judgments and possess excellent communication skills,” she said, adding that the current Japanese education system that emphasizes memorization must change.

“We must find out what skills can outpace AI and seriously discuss how we can acquire such skills before it’s too late,” she said.


Download the PDF of this  ‘Summer Davos’ Special