• 2020-11-28 Saturday

What I am talking about today is "Follow Simon’s Path” because Simon himself has a strong pioneering spirit. At the beginning, I promised Professor Yixin ZHONG that my speech would be finished in a quarter of an hour, but he said no. He insists that I should talk more. Now I am talking as much as possible, hoping to complete this task within the allotted time.


I'm a bioinformatics, but I've developed some insights into AI from a bioinformatics perspective, and I'd like to share them with you. As a pioneer of artificial intelligence, professor Simon makes a significant contribution to its development, which makes the way a machine act just like the intelligence on which a person acts, so the goal of AI is to achieve human-level intelligence. I think its emergence can be traced back to Turing's paper titled "Computational Machines and Intelligence" , in which Turing used his Turing machine as a computing machine to explore the possibility of human intelligence.


Although artificial intelligence only highlights that the way in which the machine reacts can resemble human intelligence, even so, the development of artificial intelligence is very bumpy, and whether the machine has intelligence has always been a controversial subject. Whenever an artificial intelligence expert achieves a breakthrough, he or she will be doubted. For example, someone said that if AI can prove a mathematical theorem, then they will think AI are intelligent. As a result, artificial intelligence experts have realized it and proved all the mathematical theorems in Russell’s book, but some people say that it is not smart. Later, some people say that if people make AI play chess, AI will be smart. As a result, AI play chess well, but some people still think it is not smart. Some people say that if AI can look for minerals and treat diseases like experts, then AI is intelligent, and this has also been achieved, but some people are still not convinced that AI is intelligent. Therefore, this controversy has been continuous for decades, and the history of artificial intelligence has had great ups and downs, including ambitions and dreams, but also failure and sadness. Therefore, it can be said that it is a tough journey. The failure of Japan's smart computer project in 1992, for example, was one of the big downs.


Nowadays, due to the success of deep learning and the analysis needs of big data, the society realizes that intelligence is an inevitable trend of social development, so artificial intelligence has once again received great attention and is forming a boom. But we must not forget the lessons of history. If we do not find the right path to intelligence, even though we may have invested a lot of money and manpower, we still have risks to experience another cliff-like fall. Who knows if it's a cliff after this peak. The development history of artificial intelligence for several decades gives us what enlightenment to study artificial intelligence? I feel that only by drawing lessons from the past can we truly open up the future.


What does history tell us? Artificial intelligence has been debated since Turing's paper, mainly over whether machines can understand, whether they are conscious, and whether they have feelings. For example, the Ludwig Wittgenstein once said, "intelligence can not be separated from emotional power and will power, " as did Roger Penrose, "artificial intelligence experts often avoid consciousness to analyze intelligence and thinking, but intelligence is a matter of consciousness, if there is no consciousness, there's no real wisdom. ". In the face of such questions, even as artificial intelligence experts took reasoning and machine learning to extremes, the machine stunned the world by beating the Garry Kasparov and the nine member Lee Se-dol of go, but it did not answer these question literally. In the final analysis, it is well established that machines do not have consciousness, and it is not known how consciousness will arise.


So can machines really have human intelligence? This dispute has not been resolved for a long time, and I think the breakthrough came in 1994. In his 1994 book The Astonishing Hypothesis : The Scientific Search for the Soul, Francis Crick argued that if the mechanism by which consciousness is produced in the brain was understood, the mystery of consciousness would disappear and it would be possible to simulate consciousness. In this way, a machine could have consciousness or free will. Therefore, this question makes us turn to the question of intelligent mechanism, which is a significant turning point. So it's a turning point, I think, for us to break through the dilemma of artificial intelligence, from asking whether we have intelligent performance to asking whether we have intelligent mechanism.


Let's take another look at what the failure of intelligent computers has taught us. We should recall the history. In October 1981, Japan held a seminar on the fifth generation of computers, the so-called intelligent computers, in Tokyo. Professor Motokunda of the University of Tokyo came up with the "idea of the fifth generation of computers" .Then Japan formulated a 10-year plan to develop the five-generation aircraft, with a total budget of US $430 million, which was a huge investment. Later, the Institute of Next Generation Computer Technology, headed by Yibo YUAN, struggled hard for 10 years to achieve this goal. They hardly ever returned home during the 10 years, and worked hard in this field with a near-death struggle, however, the project failed in 1992 due to the failure to achieve key technical targets, human-computer interaction in natural language, automatic program upgrades and other important goals.

What lessons do people think from this failure? People began to think about it. When we study intelligence, we should not only simulate the intelligence in function, but also simulate in mechanism, and we should also study the mechanism of intelligence generation. In addition, we should not only use top-down reasoning, but also learn from the bottom up and combine with each other, which is an important thinking. At the same time, people begin to pay more attention to the world's only agent "brain", hoping to deeply understand the structure of the brain, the connection of neurons, the cognitive process of the brain, and even want to simulate the connection of brain nerves to generate intelligence. So studying the mechanism and the brain is a profound reflection of this failure. But the question is, can machines and organisms have the same mechanism? If the machine is this mechanism and biology is another mechanism, there will be no results in this study.

 

I study biology and life phenomena from the perspective of information. As we all know, intelligence is a performance in the process of biological evolution, so the mechanism of biological evolution is consistent with that of intelligence. In other words, the mechanism of intelligence is the mechanism of biological evolution. If this is true, bioinformatics will bring us new inspiration.

 

Bioinformatics started from the human genome project, and genome analysis has made me further understand biology and its mechanism. At least I used to think that man is the soul of all things, which means so many characteristics of human beings, such as thinking, emotion, wisdom and so on, are not available to other creatures. But when I did genome analysis, I knew that at the genomic level, people and animals can be no different from microbes and viruses. Therefore, from the perspective of evolution, at the gene level, the operating mechanisms of all kinds of organisms are interlinked, but in the process of evolution, there are sooner or later. All kinds of human characteristics can be produced by other species through evolution.

 

For example, some people say that human beings can use tools, other animals can't. This conclusion is wrong because we found that some animals can use tools. Some people say that human beings have language, and other animals will not have it. Now it is found that some animals already have words and sentences. Some people say that human beings have feelings, others will not. Now it is found that animals also have feelings. Humans can count, and now dogs can count. In the future, you can find out more about some of the characteristics that human beings have, and other creatures can also have them. In principle, they are the same evolutionary process, and there is no gap between them. From the point of view of those who engage in systems, the mechanism plays a fundamental role in the performance of a system. I don't mean that the same mechanism must have the same performance. I just said that the same mechanism may produce the same performance. However, I also want to say one thing that I think is important here. Wiener said that there is a very wonderful thing in the world, that is, random and aimless machines can learn to explore their own purposes. Wiener said that this is Ashibe's brilliant thought. He said that I believe that this is not only a great contribution to contemporary philosophy, but also will lead to the high development of automation technology. It is a wonderful thing in the world that aimless random machines can produce ends.

 

Why? I did a little thinking. First, evolution, which changes its structure to adapt to the outside world, is, I think, one of the most important mechanisms in the world. How does evolution produce a purpose from something without a purpose? I have some assumptions, which are probably wrong. We know that the earliest biological reaction is from perception, so it is a kind of perception reactor. When I feel threatened by the outside world, I react accordingly. When the temperature is high, I escape. This is a kind of perceptual reaction. As long as this condition appears, I will react immediately. So through the repeated learning of the perceptual response, I feel that I must make such a reaction when the condition appears. When the conditioned reflex appears, the storage and memory will appear. Conditioning is the result of learning. If there is a situation between two reflexes, which one should I use? This is a new problem. So it is likely that after continuous learning, I prefer the conditioned reflex environment here. I adopt this conditioned reflex, and I prefer that conditioned reflex environment. I use that conditioned reflex. This requires analyzing and processing this environment, and then making a decision on what kind of conditioned reflex. In other words, the further learning of conditioned reflex needs to process perception and make decision on processing.

 

If we get to this point, the system will make decisions. Decision is a kind of independent thing. This decision-making system is independent of the objective environment, but it produces something such as will, desire and purpose. So evolution, learning, and then through advanced evolution and learning evolution, produce purposefulness, which is actually a key. This idea makes me have a new understanding of evolution, especially the evolution of learning, which gives me a very deep impression.


Now back to my thinking about artificial intelligence. I am also very interested in the long-term dispute of artificial intelligence. I have thought about this problem, and I can't get an answer. But one day, I saw BCI (Human-Computer Interaction), the interaction between human brain and machine brain and I learned that human brain and machine brain can understand and exchange with each other. I immediately realized that the reason why genes have the same mechanism is that genes can be exchanged. After exchange, genes can also understand each other. If I move this idea to human brain and machine brain, will there be the same mechanism at this level? In what level? The level of interaction between human brain and machine brain is the level of information processing, not at the level of chemistry, not at the level of physics, not at the level of energy, but at the level of knowledge processing. If human brain and machine brain can be exchanged, it will be the same as gene. So I figured out an idea: can human brain and machine brain exchange locally? Then I put forward an assumption, which is, if I replace a part of the machine brain with a part of the human brain, can it work normally? Thus, I put forward a new Turing test, which is like this: If there are two people, one is normal people, and one part of one's brain is replaced by machines and chips. Can you distinguish these two people? Can you tell two people apart by asking him questions? If you can't tell the difference, we can say that the operating mechanism of the chip is the same as that of your brain, because they are completely integrated. And I called it Turing Hypothesis.


Now back to my thinking about artificial intelligence. I am also very interested in the long-term dispute of artificial intelligence. I have thought about this problem, and I can't get an answer. But one day, I saw BCI (Human-Computer Interaction), the interaction between human brain and machine brain and I learned that human brain and machine brain can understand and exchange with each other. I immediately realized that the reason why genes have the same mechanism is that genes can be exchanged. After exchange, genes can also understand each other. If I move this idea to human brain and machine brain, will there be the same mechanism at this level? In what level? The level of interaction between human brain and machine brain is the level of information processing, not at the level of chemistry, not at the level of physics, not at the level of energy, but at the level of knowledge processing. If human brain and machine brain can be exchanged, it will be the same as gene. So I figured out an idea: can human brain and machine brain exchange locally? Then I put forward an assumption, which is, if I replace a part of the machine brain with a part of the human brain, can it work normally? Thus, I put forward a new Turing test, which is like this: If there are two people, one is normal people, and one part of one's brain is replaced by machines and chips. Can you distinguish these two people? Can you tell two people apart by asking him questions? If you can't tell the difference, we can say that the operating mechanism of the chip is the same as that of your brain, because they are completely integrated. And I called it Turing Hypothesis. What is the significance of it? It can prove whether the operation mechanism of the machine can be the same as that of the human brain at the level of information processing. Of course, this has yet to be proved, but I think part of it has been proved. The retina is a part of the brain, now it has been replaced by a chip. It can work with a chip, and it has been realized. So the fact has proved that the operation mechanism of the chip is the same as that of the human brain on the level of information processing. What I'm talking about here is the same mechanism at the level of information processing.


Because of the same mechanism at the level of information processing, computers can reason like human beings. Reasoning is a kind of logical thing, which is at the level of information processing. Therefore, it can prove mathematical theorems, play chess, and even surpass the chess king. However, it is only within the category of reasoning. Why didn't computers produce consciousness for a long time? Here I would like to emphasize that a unique mechanism different from other organisms is the function of automatically adapting to external changes and optimizing its own structure, which is evolution. Therefore, biological function is not only information processing, but also evolution. It is precisely because the computer has no evolutionary function, so the computer is unconscious no matter what it does.

If the computer or the machine brain, like the human brain, not only has the same sense of information processing, but also has the evolutionary function, then I believe that the machine brain may have all the human brain has. As for intelligence, the so-called evolution refers to the learning of learning. However, the learning of this learning is different from that of software. Its structure also changes. This is a very important point. Moreover, the structural change records the learning results and improves the learning method. Moreover, its storage and calculation are integrated, which is difficult for computers to do at present.


And this is probably a new topic to study the evolution model of computer learning, and it is a subject worthy of our attention. If the computer has learning and evolution or learning and evolution model, then it can continuously upgrade, and all kinds of problems of human intelligence are possible. So if the computer has the function of learning and evolution, it will produce emotion, even will.

Here we may get some important results:


First, from the point of view of information processing, computers and human brains are consistent, so Turing machines can be used as a universal model of computers and human brains. Second, because learning and evolution are the key to intelligence, our foundation is to study the model of learning and the evolutionary model of learning, which is worthy of our attention. Third, the human brain has undergone a long evolutionary process, and its performance is much better than that of the computer in many aspects, so many aspects of the human brain can be studied by computer and telecommunication experts. The research of brain science and cognitive science is very important for computer science.


Moreover, the human brain changes the neural connection through DNA changes in the evolutionary process. This connection not only records the learning results, but also optimizes the learning algorithm. It simplifies the components needed, and saves energy consumption. For example, deep learning is very brilliant, but at present, it cannot be used for intelligent driving. There are some problems with the automatic driving vehicle crossing pedestrians. The speed of deep learning cannot keep up, so the computation speed is not good at present. If you go further and build tens of billions of neural units by simulating the brain, according to the current technology, it can't afford the power consumption. So there are a lot of limitations in the computing technology, computing speed, power consumption, etc., which is why we should carry out brain-like research.


Fourth, we should pay attention to the learning and evolution model of the computer. Of course, it is easier to realize it with software, but don't forget that the structure need to be changed. If both hardware and software can evolve, the future results will be immeasurable.

Fifthly, researchers of artificial intelligence should abide by the corresponding ethical rules, have self-discipline and formulate basic rules, because science is a double-edged sword.

Please correct me more. Thank you!


Chinese Academy of Artificial Intelligence
2016-06-20

Copyright © 2010 中国人工智能学会 京ICP备06029423号
地址: 北京市海淀区西土城路10号 邮编: 100876 电话: 010-62281360 传真: 010-62282983

二维码