Event Report: NEC iEXPO KANSAI 2017
NEC Future Creation Forum Related Session
A future in which humans connect with AI
Part 2. Panel Discussion
Dr. Michio Kaku (Professor of Theoretical Physics, City College of New York)
Dr. Yutaka Matsuo (Project Associate Professor, Graduate School of Engineering, University of Tokyo)
Katsumi Emura (Chief Technology Officer, NEC Corporation)
Chiaki Hayashi (Representative Director, Loftwork Inc.)
Optical technology driving major advances in robotics industry
Responding to Dr. Michio Kaku’s special session, Dr. Matsuo, Project Associate Professor at the Graduate School of Engineering, University of Tokyo, agreed with Kaku’s prediction that the robotics industry would surpass the automobile industry in size.
"There was an article saying that with advances in deep learning, robotics will grow into an industry 35 times the size of Amazon. Over the past few decades, robots developed a reputation for not being very good at pattern recognition, but I think that deep learning has the potential to change that dramatically. With deeply structured networks, you can approximate very complex functions with combinations of simple functions. This means that extremely advanced pattern recognition is now possible. Robots were once not very good at gripping and lifting things up, but we've developed robots that can do such tasks very well now. I think robotics has tremendous potential as an industry too, picking out garbage, cleaning toilets and so on.
The deep learning that makes this kind of task possible is optical technology. The combination of image sensors and a deep learning mechanism should give robots eyes that can see things even better than humans.”
Just as Andrew Parker, in his book 'In The Blink Of An Eye: How Vision Sparked The Big Bang Of Evolution', argues that it was the development of vision that caused the Cambrian explosion of diverse life forms, Dr. Matsuo predicts that the same kind of explosion in diversity will occur in the world of robots and machines.
So in what specific fields will robots most make their presence? In his talk, Dr. Kaku suggested that AI and robots would have the greatest impact in fields involving dirty, dull and dangerous work. Competition is heating up worldwide to develop robots capable of decommissioning nuclear power plants, working in disaster sites, or propelling themselves through space to explore Mars and some other planet.
Matsuo suggests that robots will also make their presence felt in more familiar areas. "It should be easy to create so-called “deep doors” by equipping automatic doors with vision. Automatic doors open when pressure is applied or when an infrared sensor detects someone approaching, but they don't open until you come quite close, and they remain open for some time after you've passed through. That's because they can't see you approach, or that you’ve gone through. If you equip them with an image sensor and a mechanism for deep learning, such doors should be able to open immediately when someone approaches, and close as soon as you’ve gone through. They could learn to open and close slowly to match the slow walking pace of an elderly woman. I think we'll also see the emergence of automatic doors optimized through deep learning for air-conditioning and safety management."
How smart can robots become with AI?
“The intelligence level of robots now are only about that of a cockroach, however, we will be seeing robots with the same level as rabbits or rats emerging in the near future,” says Dr. Kaku. “We may be able to equip robots with the same intelligence as dogs and cats by the latter half of this century, and monkeys by the end of the century. Monkeys know that they are monkeys. A dog thinks that you are a dog. Dogs obey their owners because they regard them as the pack leader. Robots equipped with a fair level of intelligence are sure to emerge in the future, but at the moment, they are nowhere near that level. They are unable to understand things that are obvious to us, such as the fact that water is wet, or that mothers are older than their children. They are also not very good at communicating with people. They can't gossip, or argue persuasively, or talk about love, so they're unlikely to supersede people in fields that require such skills.”
Dr. Kaku says that he's often asked when AI will be able to predict stock prices. “A friend of mine is using deep learning to train an artificial intelligence in using thousands of data points, but there are so many factors in the stock market that a simple algorithm is unable to make predictions. I think that human beings should make their own judgments based on an overall consideration of psychology, history, the issues of management, legal issues, and such with the assistance of the many algorithms and learning programs.”
How about NEC? We asked Katsumi Emura, the company’s Chief Technology Officer, about the specific fields in which it is applying AI. “We’ve made quite a lot of progress in solving problems with definite goals using data that contains no lies. For example, we've reached quite a high level where using data from sensors located in factories to prevent malfunctions and improve production efficiency is concerned. However, achieving our mission of raising labor productivity requires that we think more deeply about the division of roles between people and AI, and how we use AI. There are limits to what AI can do, so we need to really understand and assess the outcomes of deploying AI. We're currently combining knowledge from various industrial fields to find answers to this issue.”
Knowing the limits of data to utilize big data
Research on the implementation of AI sometimes involves analyzing, assessing, and matching big data. Take, for example, the question of which company and what kind of work a person is suited to, or what kind of person would make the ideal marriage partner. Is AI capable of making such judgments?
Dr. Matsuo replied “School or university entrance exams are usually focused on about five subjects. In other words, they look at only a limited few of the many capabilities of human beings. This might have been unavoidable in times before big data became available, but now we have big data, I think we should be able to carry out more detailed, accurate matching where job-seeking or finding marriage partners and such like is concerned. The problem is that, in the view of some, using data to carry out matching will detract from overall happiness. If you're happy within your immediate surroundings, then maybe you don't need to depend on big data-based matching.”
Dr. Kaku points to the importance of knowing the limits of data. “They say in the United States that there are three types of lies— ordinary lies, damn lies, and data (laughs). People lie, and when combined with data, lies come to look very plausible. Say, for example, if you use data matching to find a marriage partner, the resulting partner may be far from your ideal. They say that women are better at judging character than men. Why might that be so? It's because men lie. We have to take into account that there are limitations to data (laughs). ”
“I think data has great power,” says moderator Chiaki Hayashi. “If you're told that the data shows this or that, you're likely to be persuaded. Because of the gap between what data shows and what people feel, you need to use data discerningly. I have no doubt that AI may be particularly good at certain things, but I think that it would be wise, when deciding something, not to forget human capabilities.”
“I agree. The words of Ryojun Shionuma, the chief priest of Jigenji Temple and a fellow member of the NEC Future Creation Forum, reminded me that human beings have tremendous capabilities,” says Emura. “A lot of us are aware of the progress being made in AI, but we'd be wise to think also about the kind of intellectual assets that we ourselves possess, and to do our best to develop them. I think this perhaps sums up the conclusion of today's panel discussion.”