The Yomiuri Shimbun This is the fourth installment of an interview series with leading intellectual figures around the world. The following is excerpted from an interview with Prof. Noriko Arai at the National Institute of Informatics, an expert on artificial intelligence.
I think artificial intelligence (AI) will never be able to surpass human intelligence.
Even if a computer that can go beyond the calculating power of the entire human brain emerges in the future, it will be unable to do anything unless human beings provide it with a framework for what to do. AI will not be able to surpass human beings.
However, it is certain that advancing AI as well as its surrounding technologies will cover considerable portions of what human beings have done so far and change society significantly.
We’ve been working on the Todai Robot project since 2011 under the provocative theme “Can a robot (AI) get into the University of Tokyo?” After about five years, the AI-equipped robot has reached the level of being able to pass entrance exams for prestigious [Japanese] private universities. The framework we gave it was to find more correct answers on the entrance exams.
However, the real aim of the project was not to enable the robot to pass the entrance exam for the University of Tokyo, but to determine “what AI can do” and “to what extent we can entrust [what we have done so far] to robots.” That will cause us to again consider the serious question “What should human beings do?” And we wanted to pose the social issue of what the future of human beings will be as a result of the advance of AI technologies.
In terms of the ability to handle information within a framework, AI has already surpassed human beings. It was a symbolic event when AI beat a top-level professional Go player within the framework of the game.
For example, when asked a true-false question, in which the ability to judge whether something is true is a key factor, the Todai Robot can give an answer with a high degree of accuracy by searching a huge amount of data. But when questions involve factors such as a sense of values, it suddenly becomes difficult for the robot to reach a correct answer.
In the politics and economy portion of a practice exam, the Todai Robot could correctly answer only one out of 10 questions testing depth of understanding of democracy.
For that reason, in terms of the challenge of arriving at an answer to a question like “Where do human beings find their values?” there is a very high hurdle to AI being able to complete the task in place of human beings. Nursing care and child-rearing will be very difficult for AI.
In addition, the issue of responsibility appears when we consider to what extent we can entrust what we have done so far to AI.
Even if medical support AI appears that can detect diseases that are highly likely based on inspecting huge amounts of data and cases, human doctors have to decide in the end whether to perform an operation. In the same way, we cannot totally entrust to AI the operation of automobiles that could cause fatalities in cases of traffic accidents.
This is because AI can’t take responsibility.
Many things will become convenient with the use of AI. Even if the realization of full autonomous driving is impossible, there will be significant progress in the practical use of technologies that help drivers prevent accidents.
Although agriculture has largely depended on accumulated experience, AI will be able to control agriculture based on big data that link harvest estimates with weather and other conditions.
On the other hand, AI will cause situations that we can’t sit back and do nothing about.
Work that AI can perform better than human beings will naturally be performed by AI.
For example, work such as accounting audits and the examination of bank loans have a strong image of being done by the elite. But much of this work involves analyzing data within a framework, a task AI is well-suited to perform. Product sales work will be replaced by AI that can determine the proper time to renew products based on data on consumers’ duration of use. In such a scenario, almost half of white-collar workers would lose their jobs to AI.
I think, at that time, work done by human beings will be divided into two categories — that which AI cannot do because it requires high creativity and assuming heavy responsibility, and that which costs employers less to entrust to human beings than to AI.
How many people in Japan produce creative work that AI can’t handle, or have the ability to perform creative jobs?
Whether the children responsible for the next generations will use AI or be used by it depends on education. An education that places importance on the ability to memorize will only produce children incapable of competing with AI.
What is important is an education that nurtures the ability to think about things and understand their meaning, and to express that thinking. This is considered the most difficult ability for AI to obtain.
However, I have no choice but to worry about the current situation.
In the development of the Todai Robot project, the robot was found to be very poor at reading and understanding relatively long sentences while taking the context into account.
But many junior high and high school students were also found to be inferior to the Todai Robot in reading comprehension ability.
It may be that they excessively depend for their communication and knowledge acquisition on social media like Twitter, which is used to exchange very short information and ideas without thinking much.
An increasing number of families have no newspapers or books at home, and many children are constantly glued to smartphone screens and seldom read logically organized sentences.
Also in the world of adults, public opinion and politics have already begun to operate based on fragmentary and emotional information. The trend may accelerate further.
Through AI research, we are confronted with an ironic question — “Have human beings deteriorated?”
This interview was conducted by Yomiuri Shimbun Staff Writer Akira Fuyuki.
Arai, a doctor of science, took up her current position in 2006 after serving as a visiting researcher at the Institute for Advanced Study at Princeton University and other posts. She directs a project aimed at enabling an AI-equipped robot named Torobo-kun to solve problems on university entrance exams. She is 54.Speech