ロボットは東大に入れるか。Todai Robot Project

日本語|English

NII Today No.46 Jun2013 > That's Collaboration 1

What Impact Will the Todai Robot Project

Grand Challenge Have on Society?

The aim of artificial intelligence is to reproduce human thought processes and intelligence on computers. AI technologies have made remarkable progress since the earliest days of computing, after several turning points and major challenges. The Todai Robot Project, which began in 2011, is also drawing attention as a major new challenge which will further advance artificial intelligence technology. We talked with researchers on the frontline of AI about its potential, and their hopes for the project.

Tetsunari Inamura (left)

Associate Professor, Informatics Principle Research Division, NII
Associate Professor, Department of Informatics,School of Multidisciplinary
Sciences, The Graduate University for Advanced Studies
"Todai Robot Project" Sub-Project Director

Kohichi Takeda (center)

IBM Distinguished Engineer
Senior Manager, Knowledge Infrastructure
IBM Research - Tokyo

Toyoaki Nishida (right)

Professor
Department of Intelligence Science and Technology
Graduate School of Informatics
Kyoto University


The fundamental question of

"what is intelligence?"

Inamura Artificial intelligence is often thought of as being a young field, but it actually has quite a long history.
Nishida The word "artificial intelligence" was first used at the Dartmouth Conference*1 in 1956. The world's first computer, ENIAC, was developed in 1946, so it would be fair to say that full-fledged research into artificial intelligence began right after computers were invented. Artificial intelligence research since then has undergone a paradigm shift roughly every 10 years. While no artificial intelligences which surpass total human intelligence have been created yet, various elemental technologies, like data mining technology*2 , audio and image recognition technology, natural language processing technology, and information retrieval technology, are already in use in society.
Inamura With regards to creating human-like results — which also applies to robots, in the sense of humanoid robots — what do you think about the relationship between artificial intelligence and robots?
Nishida There are two extremes regarding the embodiment. One holds that it is produced by the brain alone, and the other that it is fundamentally dependent on embodiment. I believe the latter. I believe that embodiment is essential to an intuitive understanding of diagrams or spatial relationships, and hence artificial intelligence and robotics share a significant amount of common thoughts, principles and designs.
Inamura I've specialized in robotic engineering since I was a student, so I'm in charge of the physical challenges presented by Todai Robot Project. These physical challenges are typical examples of hurdles that cannot be overcome by cerebral thought alone. Handling real-world phenomena, like throwing a ball, or breaking something by applying too much force, requires the knowledge and direct understanding of physical laws that people absorb through their own physical experiences. Right now, we are using simulators, but in the future I hope that robots will be able to be active in the real world, just like humans, and acquire knowledge through their experiences. However, the engineering hurdles to achieving the functions of the human body are very high, so in the future the issue of how to tie it together to artificial intelligence, in the sense of cerebral functions, will be a major challenge. 
Nishida We also have to think deeply about the fundamental issue of what is necessary to create humanlike artificial intelligences and robots. For humans, in addition to intelligence, mind and emotion also play important roles. Creating remarkably sophisticated artificial intelligences and robots requires these types of elements to be included.
Takeda The Watson project, which I took part in, did not include these human elements. We focused exclusively on the ability to compute answers to given questions. For example, we didn't use voice recognition, just voice synthesis when it read back the answers. It is not a system which would listen to a question and respond like a human, either. On the other hand, it does not panic when behind and make incorrect answers. In that sense, Watson is not replacing humans, and instead, it is a system for supporting human intellectual activities by specializing on using a large amount of data to generate potential answers and hypotheses to provide evidence-based answers to questions.



Entering the fields of

embodiment and heart

Inamura Just as IBM positioned Watson as a grand challenge, NII has positioned Todai Robot Project as a grand challenge. The objective is to pass the Tokyo University entrance exam, but even if it achieves that, it will be just the first step towards true intelligence. For example, consider this elementary school science problem. "Butter is applied to several locations on a metal rod, and beans are stuck to the butter. If one end of the rod is heated with a candle, in what order will the beans fall off?" Even if the Todai Robot Project were able to pass the Tokyo University entrance exam, it wouldn't be able to answer this question. That's because it cannot be turned into symbols or an equation. However, for a human, even an elementary school student, this can be solved based on just experiential knowledge and common sense. So, how should this type of human knowledge be handled? Right now, we are trying to use advanced programming, but, ideally, in the long term it would be possible to acquire by artificial intelligence on its own. In other words, we will have to move up from Tokyo University to elementary school. That will be really difficult, but it's an area into which robotics and artificial intelligence needs to move. I believe that the Todai Robot Project will be extremely significant as a first step.
Takeda In the latest IBM 5 in 5 – a list of innovations that have the potential to change the way people work, live and interact during the next five years, we discusses a list of innovations to mimic the five human senses. IBM researchers have begun exploring on advances which will help computers to incorporate human senses in their own ways. It will not be easy to convert biological information such as taste or smell into current machine learning and statistical reasoning. However, if it can be done, it might be possible to mimic human-like thinking and intelligence even more. What's more, if physical sensations play a role in the formation of human common sense, as well as affecting feelings and emotions, they would be vital elements in the world of information technology.
Nishida The first-stage objective, scoring highly on the National Center Test for University Admissions, could  be achieved to some extent using a Watson-like approach of focusing intently on the ability to handle questions and responses based on large amounts of data. However, answering questions such as Japanese questions on the second-stage exam will require the inclusion of elements such as feelings and emotions. Generating written answers for the second-stage exam requires enough written expressive ability to impress the reader. It would be a great accomplishment if we were able to achieve that. I think what makes the Todai Robot Project such an important and groundbreaking challenge is that it delves deep into emotional understanding and expression as well.
Takeda The very fact that we can take on those challenges shows the tremendous potential of the field.
Nishida As an artificial intelligence researcher, I would love to have a conversation with an intelligence that could pass the second-stage exam.. I think that that by practically searching for answers to challenges in which human intelligence is reflected, we can penetrate into the heart of human intelligence itself. It would certainly increase human creativity and contribute to greater societal progress. I'm looking forward to the future challenges posed by the Todai Robot Project.
(Written by Akiko Seki)

*1: Dartmouth Conference
Artificial intelligence research conference held in 1956. Its official name is "The Dartmouth Summer Research Project on Artificial Intelligence", and was the first place where the term "artificial intelligence" was used.
*2: Data mining technology
Technology for identifying information which is useful for people from a large volume of unorganized data.