Todai Robot Project


NII Today No.46 Jun2013 > That's Collaboration 2

The Younger Generation Discusses

Their Hopes for the Todai Robot Project

What does the next generation of researchers, which has not experienced the past setbacks in the AI field, think of the Todai Robot Project? We talked with Yasuhiro Matsumura, who passed the Tokyo University entrance exam last year, and Hiroyoshi Komatsu, a high school student who is participating in the Todai Robot Project, about what interests them in the project, and their hopes for its future, from their respective perspectives as an impartial observer and someone directly involved in the project.

Would a robot pass the
mathematics section
but struggle with the Japanese

Arai  I heard the other day from an acquaintance at Tohoku University that when they asked engineering department students what project they'd like to take part in the future, several answered "Todai Robot Project". A lot of people are interested in the project, not just students majoring in natural language processing or mathematics. Today, I'd like to get the opinions of the younger generation on this project. How did you two first hear about the Todai Robot Project? 
Matsumura I'm part of a newspaper club at University of Tokyo, and I found out about it in a news release but, to be honest, I wasn't particularly interested when I first heard about it. I read an interview with you, Professor Arai, in a mathematics journal in 2011, and I found your approach of putting science and mathematics questions into words to be very close to my own interests. It made a big impression. Then this spring, when I interviewed you for the University of Tokyo Newspaper, I heard more about the project, and realized how interesting it was. 
Komatsu When the project first began, I saw it covered on the news, both on TV and in newspapers, as well as NHK Special, and was amazed that this kind of world existed. I'm in 11th grade now, but I started programming when I was in 7th grade. In 9th grade I grew interested in artificial intelligence research and began working on natural language processing, so I wanted to work on the Todai Robot Project, if possible. Actually, after about a month of studying natural language processing, I wanted to join the Association for Natural Language Processing, but my parents objected (laughs). But I kept studying on my own, joining the next year, which led up to my involvement in Todai Robot Project.
Arai The computing power, data, modules, and other things needed for programming are now available in open source form, so anyone who wants can get started, right? So it's not unusual for a high school student such as yourself to participate. Not in this modern age of the web.
—— Matsumura, you passed the University of Tokyo entrance exam just last year. Do you think a robot could pass the test?
Matsumura I think so. I'm not sure about Japanese, but the mathematics section isn't that hard if you understand the patterns.
—— As a member of the project, what do you think, Komatsu?
Komatsu One of the reasons I got involved in the project was that I hoped it might boost my own test grades, but I have to say that it hasn't particularly helped me with my own studies (laugh). I think that Todai Robot Project's ability to reach its objectives on the 2016 National Center Test for University Admissions will vary depending on the subject. It's hard to come up with approaches for solving questions on sections like the Japanese section.
Arai There are a lot of challenges for each section, but I think that believing the objectives are possible is important. I think that the reason so many researchers think that it might be possible this time is because the conditions that would make passing possible are gradually falling into place.

The allure of artificial
intelligence research is
its insights into the human mind

—— What aspects of Todai Robot Project are you interested in?
Matsumura Artificial intelligence research consists of quantitative linguification of the question, "what does it mean for a person to think?", which I think is interesting.
Arai The question of how much of humanity's mental activities can be replicated mechanically has been discussed from a philosophical standpoint since the days of Descartes and Hobbes. During the early 20th century, when Turing developed the logic on which computers are based, this question became divorced from its philosophical underpinning, and a framework was created for considering it as an engineering problem. Artificial intelligence research made many advances, but in the 1980s began faltering. At the time, computing power was still low, there was little data, and the environment itself was not yet robust. The issue of artificial intelligence was broken down into smaller segments, but I think every 30 years these need to brought back together, and used to accomplish whatever is possible with the technologies of the time, leading into the next age of AI research. In this project, I hope to integrate and consider these technologies, and identify what can and cannot be accomplished. What cannot be accomplished can be set aside for 30 years down the line.
Komatasu What's interesting about artificial intelligence is that creating it gives you a greater understanding of people. Even for problems which are extremely easy for people, we still don't understand the thought processes involved. I'm also studying neuroscience and cognitive science, and I hope that by working to create artificial intelligence, I can also shine some light on those fields. Even if I fail, I can use what I have learned to keep moving forward.
Matsumura Plus, if the Todai Robot Project reaches its center test objectives, it will prove that the capacity of artificial intelligence is improving. I expect that this can be applied to a wide range of fields, producing results that benefit society.

A challenge perfectly suited for
Japan - using small data to
improve accuracy

Matsumura I'm not sure whether I want to focus on mathematics or physics in the future, but in the case of physics, modeling is used to provide an approximate answer to a question, while mathematics provides complete and immediate answers to stated questions. I find that very alluring.
Arai That's one of the cool things about mathematics. I call the Todai Robot Project a "risk hedge project". When trying to provide machines with artificial intelligence that corresponds to the learning abilities of humans, one can only improve accuracy in logarithmic relation to the amount of data, so big data is required for accuracy improvements. However, it's difficult for academic institutions inside Japan to collect the big data they need. That's why there needs to be research into how to improve accuracy using small data. Even if we collect all the entrance exam data for the last twenty years, it only constitutes a small amount of data. The approach of making accurate inferences based on small data goes against the prevailing worldwide research current, but it's extremely important. And cool. I hope that young people get to experience that.
Matsumura Artificial intelligence research won't result in the perfect replication of the human mind within the next 100 years, but I wonder what AI will be like a century from now, and whether it will have exceeded human intelligence.
Arai Artificial intelligence research won't result in the perfect replication of the human mind within the next 100 years, but I wonder what AI will be like a century from now, and whether it will have exceeded human intelligence.
Komatasu Yes. Right now I'm thinking exactly how to approach the research. I'd like to make programming improvements that are not subject-specific, like information retrieval improvements.
Arai Let's take on the challenge together.
(Written by Yuko Sakurai)

Hiroyoshi Komatsu (left)

Tokai University Bosei Senior High School

Yasuhiro Matsumura (right)

Natural Sciences I, The University of Tokyo

Noriko Arai

Director, Research Center for Community
Professor, Information and Society Research
Division, NII
"Todai Robot Project" Project Director