Professor Junichi Takeno, one of the leading experts in the study of the nature of consciousness and intelligence, explains why he and his team are trying to create a computer analogue of human intelligence, and how this research can help biologists uncover the most important mysteries of human evolution.
Takeno, a professor at Meisy University in Tokyo, and his colleagues have been working for more than 10 years to create machines that have some forms of consciousness and intelligence and work on the same principles as the human brain. In 2005, his team created a robot that can see itself in a mirror - one of the first tests of self-awareness that only humans, chimpanzees, jays and macaques can pass today.
In the years that followed, Takeno and his associates created a range of robots capable of recognizing and displaying various emotions, imitating the actions of humans and other robots, and making decisions on their own. All this, according to the scientist himself, brings us closer to creating a full-fledged intelligent machine, indistinguishable in the way of thinking from a person.
Last week, Takeno spoke about the latest achievements in this area, speaking at the international conference BICA-2017, which was held for the first time in Moscow with the active support and participation of NRNU MEPhI. He announced the creation of an artificial intelligence system that perceives the famous "Rubin's vase", a dual picture that simultaneously depicts a pair of dark human faces and a light vase, in much the same way as a person.
Junichi, many of your colleagues believe that a full-fledged analogue of the human mind and a self-aware machine cannot be created, since the mind has a quantum nature. Is it really?
- This is a very difficult question, but it seems to me that human consciousness can be represented as a program. This program, of course, is very unorthodox, it has a very unusual structure - it was written by evolution on the "genetic code" for many millions of years and it can work only on such "hardware" that we do not fully understand yet.
In order for us to write a program that works in this way, my laboratory colleagues and I are now actively studying the nature of human consciousness, using publications that have been prepared by our fellow neurophysiologists, philosophers, psychologists and other scientists. These works gave us a lot of food for thought and allowed us to make significant progress in writing this program.
For example, the Russian Nobel laureate Pavlov once wrote that “consciousness is a function of the brain, that is, matter organized in a special way,” and he confirmed this statement by experimenting on dogs.
Promotional video:
On the other hand, the nature of human consciousness is often said to remain largely unexplored. We are just trying to solve this riddle, moving from the reverse side - we study how our brain and our consciousness work, gradually creating its semblance. Many people do not want to do this in principle, but it seems to me that they are wrong.
Junichi Takeno and his students in Russia
Man today is the most intelligent creature on Earth, and therefore it is extremely difficult for us to understand how our own intellect works. But you need to study it, because if intelligent robots and machines suddenly appear among us, then a very dangerous situation will arise.
Without understanding how our mind works, how our own genetic program works, we will not be able to predict the actions of such machines and protect ourselves from them. In addition, from a purely philosophical point of view, a person wants to know himself, and therefore such research will always have meaning and value, no matter what skeptics say.
Over the past 10-15 years, you have created many robots that have often made it to the front pages of newspapers and television news reports. Why do most people like robots, but are afraid of "ethereal" intelligent machines?
- It seems to me that this attitude is due to the fact that a person always needs a good companion, a life partner. For example, in the film "Bicentennial Man", based on the story of Isaac Asimov, the robot sympathizes with people and wants to become like them. It is easy and pleasant for people to communicate with him, and such a scenario is one of the ideals that humanity aspires to. We all want to create such a "friend" robot.
There is another side to this attitude. A person always strives for absolute control over things and other people, and a "material" robot with a clear frame and body, we can control at least at the psychological level.
If the machine acquires a mind and goes beyond the body, then this possibility disappears for us, because, as I said above, we do not yet understand how human consciousness or its future computer analogue works, how far it will go beyond its boundaries … In addition, a person perceives mind and consciousness as something natural, and it will be incredibly difficult for us to destroy this intelligent form of life from a psychological point of view.
For the same reason, the research of human consciousness itself is perceived negatively by many people, since it can give scientists or politicians with such knowledge control over your thoughts and mind.
You have created a program capable of reacting to Rubin's vase and other ambiguous pictures in much the same way as a person. Can we talk about whether it is an analogue of human consciousness?
- Now we are somewhere in the middle, so to speak. In the past, we have already created a kind of analogue of a single human thought, the MoNAD system, an analogue of a recursive neural network from the human brain. This time we used it to explain why a person thinks he sees two different things at the same time, people's faces and a vase, and how associative memory works.
Of course, you need to understand that all such explanations are still hypotheses, but these studies help us gradually uncover the mystery of the brain, studying one relatively small problem after another.
For example, in the past, our main task was to create a system that could, like all sentient beings with consciousness, spontaneously recognize themselves in the mirror. Having reached this milestone, we began to tackle the next problem by developing a machine that could recognize and generate emotions.
Given this progress, when will you and your colleagues reach the point where we can talk about a full-fledged intelligent machine?
- What we already have, in principle, already works in some respect, like the human brain. On the other hand, we are well aware that we only know a few basic things about how consciousness and mind function. Both are extremely difficult things, our colleagues are constantly discovering some new qualities and properties of the brain that have yet to be studied.
In the future, it seems to me, the same thing will happen - we will discover new processes and properties of the brain, which we are not aware of now. Therefore, the creation of a full-fledged analogue of the human mind and consciousness, most likely, will take forever.
The same is actually typical of science - every time scientists think that "physics" or "biology" is over, they discover something new that radically changes our understanding of the world.
For example, the first mechanical watches appeared at the end of the Middle Ages, and over the next six hundred years, their design was improved and reworked many times over. The same goes for studying the human mind. The main thing is that we pay maximum attention to safety when conducting such research.
Will the creation of such systems require fundamentally new computers, or are modern chips, in principle, suitable for solving this problem?
- I belong to the school of so-called connectivists, which postulates that all intelligent systems consist of many interconnected simple elements. We believe that already now nothing prevents us from creating a network of artificial neural networks that could acquire consciousness and intelligence.
On the other hand, the creation of such a network will be quite problematic in practice, since modern computers are poorly suited for the parallel execution of many simple calculations. We need devices that can carry out hundreds and thousands of simple operations simultaneously, which will make the operation of neural networks as fast and efficient as possible.