Interview with brain scientist Li Zhaoping and his wife: Ideas perfusion into the brain can be achievedDate: 2018-06-21 14:50
Academician Peter Dayan of the Royal Society and Professor Li Zhaoping of the University of London College of Computational Neuroscience at the closing ceremony of the Shanghai Forum 2018.
On May 28, the closing ceremony of the Shanghai Forum 2018 was held at Fudan University. Peter Dayan, an academician of the Royal Society and a professor of computational neuroscience at the University College London, who attended the conference and gave a keynote speech, accepted an exclusive interview with surging journalists. During the interview, the two brain scientists gave professional answers to the research direction of artificial intelligence, the realization of brain-computer integration, and the risks faced by artificial intelligence.
Peter Dayan is a member of the Royal Society of England and one of the winners of the 2017 Gretlundbeck European Brain Research Award (abbreviated as Brain Award). In 2018, he was elected as an academician of the Royal Society.
Li Zhaoping is a professor of computational neuroscience at the Department of Computer, University College London, and a visiting professor at the State Key Laboratory of Cognitive Neuroscience and Learning at Beijing Normal University.
Surging News: Humans are living organisms with emotions and creative abilities, so should artificial intelligence be more integrated with human-related technologies such as genetic technology, neurons, and brain science?
Li Zhaoping: I am not familiar with genetic technology. I do brain science. I may answer this question a bit subjectively. Brain science is science, and artificial intelligence is engineering technology. The best way to solve AI is to use scientific methods. Of course, if you do n’t understand science, engineering technology can also be used to make useful AI. Just like making a carriage, you can do it without understanding the principles of mechanics. Make a very good carriage. In addition, useful AI doesn't matter even if it doesn't have emotions. You don't have emotions between strangers, but it doesn't affect the communication between them, as long as the person is straightforward to express correctly.
Peter Dayan: There are a positive factor and a negative factor to solve artificial intelligence using genetic methods. The negative factor is that the genes are too complicated. For example, there are 70 genes related to height, and 1,000 genes related to a neurotic disease. Although, you know the genetic root of the disease, but as many as 1,000 genes are related, there is no way for people to adjust the genes to cure the disease. The positive factor is that some things are born to be learned through genetic inheritance. For example, when a lion comes, he should run away with this knowledge. After birth, people know without learning. Through this understanding to understand, many important decisions made by people can be understood at the genetic level, which is a positive impact. In addition, human emotions are actually closely related to genes.
Surging News: The American drama "west world" that is being broadcast is a science fiction film about artificial intelligence. The play describes the scene where the idea is input into the brain through the machine. The life of the controlled person is the script written by the input person. Will it happen?
Peter Dayan: First of all, this has already happened in part. Through social tools like Facebook, people have been manipulated by others. This has not yet used genetic technology. In addition, there are many people who are addicted to certain things. Addiction is actually the control of the reward and punishment system of the human brain, and the controlled people always want to do gambling. In one experiment, an electrode was inserted into the brain of a mouse, and the mouse was often allowed to do the same thing by electric shock. It always liked to do this, but it was actually controlled by man. Through this example we can learn how the brain controls our behavior. So technically speaking, it is possible to control people's ideas. Although using this technique can also cure the disease, we can make some Parkinson's patients less trembling by electric shock. This is a matter of ethics, but technical mind control is achievable.
Surging News: To what extent do we now understand the brain? Can current research determine the direction of artificial intelligence research?
Li Zhaoping: As I said before, AI artificial intelligence is actually an engineering problem, and brain science is a scientific problem. If you want to use engineering technology to realize artificial intelligence, you do n’t necessarily have to create the appearance of a person. . At a higher level, we need to have an intelligent subject, just like the biological subject. For example, human intelligence is an example, and artificial intelligence is another example. After we understand the science of intelligence, we can make engineering intelligence. The understanding of intelligence in science can be studied using humans as an example, or you can use something you have created yourself, but this is intelligent science. With intelligent science, you can create an engineering thing.
Surging News: Can it be said that what we use now, the intelligence in the sense of engineering is still a shallow technology.
Li Zhaoping: It cannot be said that it is shallow, and engineering and science cannot be higher than others. That is to say, the mechanical department and the physics department, who is higher and who is lower? The engineering department can make things, and the physics department knows the principle of motion. I think this can't be compared. But if you know the principle, you can make things better, but you do n’t necessarily have to know the principle to make an airplane.
Surging News: What is intuition, sometimes intuition is very effective, how to explain this quick way of thinking, can artificial intelligence also have intuition?
Peter Dayan: Many decisions are very difficult to make. The reason is that the information is too complicated and too much to determine which solution is good. So there will be a way of thinking about approaching, people will take the short cut, and machines will also take the short cut. But the human shortcut and the mechanical shortcut are not necessarily the same. There are restrictions on people taking short cuts. People only eat so little food, so little energy, so big brain capacity, and machines can be made large, and they can have enough electricity to support it. . If you have intelligent science and know what the principle is, you can know how people should copy short cuts and help us decide whether to buy a house or not. At the same time, it can also decide how to copy the short cut.
Surging News: I understand that human beings are living organisms. The decision-making process should be a chemical or biological reaction, and machine thinking is a physically repeatable and controllable process. Get it through?
Peter Dayan: There is a functionalism in philosophy, as long as the function can be achieved, it does not matter what hardware is implemented, just like a semiconductor, it was a vacuum tube at first, and later became a triode. The hardware of the two is different, but the principle is the same. Therefore, it cannot be said that the two will not work because the hardware is different. How to fly? Whether you fly with an airplane wing or with feathers, as long as you can fly, it doesn't matter if their hardware is different.
Surging News: I guess that organisms must have emotions. The creation of emotions and the thinking of computers should be different, but no matter how computers evolve, there will be no emotions in how they evolve?
Peter Dayan: I want to ask you a question, what is emotion?
Surging news: Emotions should play a role in people's thinking or decision-making, such as love and hate.
Li Zhaoping: In the final analysis, how excited is emotion, and is the judgment of the value of something. Putting the two together is emotion. The things that are combined in this way can be put on the machine. Whether the machine uses the power saving mode can be a question of how excited it is. If you want the machine to have values, it is a matter of choice. For example, should we turn on WeChat or turn on the camera? Who is the first and who is the second? This is a question of values. Emotions can be a choice, and emotions can be an understanding. Reflected in people is that I love you more than him, I love WeChat more than the camera, in the final analysis is actually a choice.
Surging news: I am a little surprised that emotions can be simulated by machines. Two identical computers running the same program will get the same results, but the twins have the same biological composition and highly similar growth history, but generally speaking, the twins have similar appearances, but the differences in personality and talents are very different. Big. Why is there such a big difference between the same person and the same machine?
Peter Dayan: The twin experience you said is the same as your definition. In fact, they have different experiences and will do different things. The two machines are exactly the same, but if they have different situations, they learn different things. In the end, the two machines are different. In fact, this is the same principle as the twins. At the beginning, the same twins have a clear difference at the end, because there will definitely be different experiences in this process. The twins hug the father a little more, and the mother the twins hug a little more, So they can't have absolutely the same growth process. If the twins experience can be absolutely the same, I think they will be the same in the end.
Surging News: Can the human brain nerve electrical signals be simulated by a computer?
Li Zhaoping: It is possible in principle. The actual operation depends on how detailed the simulation is. If the simulation is very detailed, it is not possible, because it is impossible to simulate carefully at the atomic and proton level. For example, you can build a house exactly the same, but every detail and every atom inside is the same? That's impossible, so it depends on how subtle it is.
Surging News: The brain-computer combination that Musk and Chen Tianqiao are doing, can this brain be connected to physical electrical signals?
Peter Dayan: Of course, the technology is there, but this is still a very rough thing, and the emergence of this technology is not to say that it is definitely feasible. The problem is whether this brain-computer communication is what people want. Communication is not effective communication. Or you can ask another question, what kind of brain-computer interface is what humans want? Because the interface is the communication between humans and machines, what kind of communication is what you want. For example, cochlear implants, where electrodes are inserted into the cochlea to stimulate cochlear neurons, can make people hear sounds. This requires the human brain to learn new signals, and it can be heard by learning to adapt to these signals. This is a very good example of brain-computer integration, but the question is whether such an example can be generalized elsewhere.
Surging News: What are the risks in the development of artificial intelligence?
Peter Dayan: Artificial intelligence may cause me troubles in the future. There are two kinds of troubles. One trouble is that artificial intelligence is smarter than us and controls human beings. The other trouble is that we rely on artificial intelligence to help us do everything. If we rely on artificial intelligence to operate nuclear power plants, but artificial intelligence is very stupid, they explode nuclear power plants. I think the second kind of trouble is more likely. We rely too much on dumb artificial intelligence to do important things. Some decisions are difficult for humans to make, because there is too much information, and it is impossible to make a fully thoughtful and scientific decision. At this time, it is necessary to approach the road, but if fantasy lets robots make important decisions for us, robots It is also necessary to take a short walk, but the situation of the machine is the same as that of the person, so we make bad decisions, and the machine may not make good decisions.
Surging news: Many people say that artificial intelligence cannot create and innovate, so will artificial intelligence combine with neuron and brain science and other biological related technologies in the future, will it make innovative creations?
Peter Dayan: Of course, there is no doubt that this will happen. Just like playing Go, Li Shishi's chess proves that the machine has the ability to innovate, which is more powerful than people. This creativity is very powerful, so it can already do it.
Surging News: Playing chess is the result of robots running on programs written by humans. This result is a result that people can think of but have not found?
Li Zhaoping: No, if people can imagine what it does, Li Shishi can also think of it. You said that what people did not think of is creation. They played chess that no one had thought of before they could defeat Li Shishi. The chess played by Alpha GO is something that no one thought of. You can't say that it is done by a program that a person enters. This is learned by itself. It is a thing that has been learned and created by the learning method entered by a person.
Surging News: Those things made by Alpha GO are still one of the human methods. Robots just find these methods. Robots will experiment on their own, like finding a new DNA structure like a biologist?
Peter Dayan: Alpha GO's chess, if it is played by one person, you will think this is an innovation, in fact, this is already an innovation. In the case of Go, the machine can make things that people did not expect. The same is true for finding DNA. How to find DNA is regular. We have to experiment and find many facts. This is the same as playing Go. The machine can also go to see what structure this reaction is and what structure it is, and then find New steps. If a person can do this experiment, this scientist is very powerful. This is innovation. Since Alpha GO can innovate better than people, so long as the machine knows the rules of doing things, experiment like this, experiment like that, and follow the rules. New choices can be made, which is innovation.
Surging News: Is it possible to discover the rules? After all, playing chess still follows the rules of Go.
Peter Dayan: If there is a field where you can do completely open learning, let's assume that a robot is going to race and run. The faster it runs, the better. Everyone runs fast. The robot will even think of a new way to collide When the line is on, the robot falls down, relying on a very long height, can hit the line faster, this is newer than you think, it can break the rules, the robot will definitely have these innovations.
Surging News: Many Chinese people like to read "Dream of Red Mansions", but unfortunately they did not write well enough after the forty times. Will the robot continue to write?
Peter Dayan: Why don't other writers continue to write the following content?
Surging News: Someone has continued the writing, but compared with the original, the writing style, character fate, poetry, etc. are very different from the original. So can you make consistent creations in artistic style?
Peter Dayan: In principle, the robot should be able to do it, because now the computer can compose music, such as learning a lot of Bach's music scores, and then the robot will make new music, which is the same style as Bach. So I think that what is feasible in principle can continue to expand the scope of application, and I don't think there are any obstacles. Although the machine translation is still not satisfactory, I think it should eventually reach a higher level. We already know that artificial intelligence can achieve what we expect in those small mechanisms, although the big aspects have not yet been achieved. For example, robots can make Bach-style music scores. This is just a small success. But I think it should be possible. That is to say, you have successfully walked a certain distance on a road, and I haven't seen any obstacles in front of me that can stop me. Although I haven't reached the end yet, in principle, I can reach the end.
ã€Introduction to the interviewee】
Peter Dayan is a member of the Royal Society of England and the winner of the 2017 Brain Award. He studied mathematics at the University of Cambridge. After graduation, he studied with Professor David Willshaw at the University of Edinburgh and obtained a Ph.D. Since then, he has followed Professor Terry Sejnowski of the Salk Institute for Biology and Professor Geoff Hinton of the University of Toronto to complete two postdoctoral degrees. After serving as an assistant professor at MIT for three years, Dayan facilitated the establishment of the Gatsby Computational Neuroscience Center at University College London in 1998. This year he made an academic visit at the Uber Artificial Intelligence Laboratory. Peter Dayan won the David E. Rumelhart Award in 2012 and was one of the 2017 Gretelundbeck European Brain Research Awards (Brain Award). In 2018, he was elected as an academician of the Royal Society.
Dayan's research field is mainly the mathematical calculation models of neural networks, especially the models related to expression, learning and decision-making. In his recent work, he focused on the various behavioral mechanisms used by humans and other animals to maximize rewards and minimize penalties. These behavior mechanisms are obviously similar to the popular algorithms in artificial intelligence. This correlation provides us with a rich way to understand the daily behavior of humans and animals, and also helps us understand the behavior of people when they suffer from neurological dysfunction and mental illness.
Li Zhaoping is a professor of computational neuroscience at the Department of Computer, University College London, and a visiting professor at the State Key Laboratory of Cognitive Neuroscience and Learning at Beijing Normal University. Li Zhaoping graduated from the Department of Physics of Fudan University in 1984 and received a Ph.D. in Physics from the California Institute of Technology in 1989. She has worked as a postdoctoral fellow at the Fermi National Laboratory in the United States, the Advanced Research Institute in Princeton, and Rockefeller University, taught at the Hong Kong University of Science and Technology, and made academic visits in other academic institutions. In 1998, she co-founded the Gatsby Centre for Computational Neuroscience at University College London, and now she is a professor of computational neuroscience in the computer department of the university. Her scientific research over the years has involved physics, neurobiology, etc., but mainly focused on vision, smell, and nonlinear neurodynamics. From the late 1990s to the early 2000s, she theoretically proposed that the primary visual cortex of the brain establishes a significant map of visual space To guide visual attention automatically, a lot of experimental work in the field is currently experimenting with this theory. In 2014, Oxford University Press published her "Understanding Vision: Theory, Models, and Data".
Computer Backpack
Computer Backpack
Mianyang Crossing Cross-Border E-Commerce Co., Ltd. , https://www.wycypets.com