Zhang, academician of the National Academy of Sciences, the full text of the speech: Artificial intelligence three pillars of development _ AI

Zhang, academician of the National Academy of Sciences: The three pillars of artificial intelligence

2017-07-23 Li-Qi Science and education observation
June 21 Morning, Stanford University physics professor, American National Academy of Sciences Zhang on the "three pillars of AI" for the title of the speech.

Zhang is employed at Stanford University Physics department. This paper mainly studies condensed matter physics, focusing on topological insulators, and obtains a great deal of world-class research results in the research direction of HTS, quantum Hall effect, spin electronics and strong associated electronic systems.

Lei Feng Network AI Technology Review organized the scene shorthand, and did not change the original intent of the collation and editing. The following is the content of the speech:

Good morning, everyone. I am very honored to have the opportunity today to share with you, I as Stanford professor, is also the founder of Dan Hua Capital, we have made a lot of investment in the field of artificial intelligence, today want to share with you is my understanding of the three pillars of artificial intelligence and prospects.

Ai did come to a very exciting time, because we humans have evolved and evolved on Earth for almost 100,000 years, and we have been the most intelligent species on earth for the past 100,000 years, but now we have a challenge, as the previous guests have mentioned, Perhaps there will be a new generation of species, which is a challenge to our entire human intelligence and leadership on Earth.

The reason why artificial intelligence can develop to today is mainly the collection of three big trends. One of the most fundamental laws of the information society is Moore's Law, our computational power, which has doubled over the past fifty or sixty years every 18 months. Now whether artificial intelligence can go to the next level, the key depends on whether the Moore theorem can continue to move forward. As a physicist, my view of Moore's law is here to share.

Artificial intelligence is due to the emergence of large data, especially the emergence of the internet, China has BAT, the United States has Silicon Valley and face book, produced a large number of data provided to machine learning. There is also an important aspect, is the advancement of algorithms, these three pillars is to make artificial intelligence to advance the fundamental principle.

But I think the fundamental point is whether Moore's law can move forward. In the past fifty or sixty years, every 18 months of computing power will be doubled, this is a very alarming trend of development. In the past fifty or sixty years, since the development of integrated circuits, has been in accordance with the direction of Moore's law. But now there is a crisis. Mainly we are talking about the information highway, but to the top of the chip no information superhighway. A lot of electricity is needed in the cloud computing room, these electricity in the top-level operation into a kind of heat, so that the thermal energy can not continue to emit, semiconductor integration can not be integrated, but the most basic principle is that the electronics at the chip level, its use is chaotic, electronic and electronic interaction between, Electrons and the surrounding impurities have interaction, in the collision process so that the original electricity into useless heat, so continue the words, the heat emitted out is doubled.

We have put forward a new idea, that is, at the level of the chip, the electrons can run like highways, so that they can do their own way, without interfering. This is a big crisis for us. But also an opportunity, Moore's law to find the "crisis" is also a great opportunity, each city is thinking about whether to become the next Silicon Valley, but to become the next Silicon Valley how to seize a new special opportunity, Moore's Law of the "crisis" is the opportunity we bring today.

My scientific research has found a very magical quantum phenomenon, so that electrons at the chip level can have an optional orbit and the way to achieve the different, non-interference in the operation of the highway, so that electrons can be at the chip level almost no energy consumption in the movement. So Tencent's future cloud computing may not need to spend so much power, nor need so many smart devices. If we're going to move this field forward, the material science has also brought a great challenge, the recent theoretical predictions and the introduction of a new material, a bit like graphene, is the atom to replace, at the level of the single atom operation of the electron can be like the highway without interference, each line.

Today's achievement is also a trans-boundary thinking. We put a very tall detour in mathematics, called the concept of topology applied above, so that we can really achieve the electronic chip-level highway operation. This is a startling discovery because of all the material in the past hundreds of thousands of years of human development process, every landmark of the history of human civilization, are named after a material, such as Paleolithic, Neolithic age, Bronze Age, iron, silicon era, But all the materials in the past have been discovered under experimental conditions, and this time we are theoretically predicting the possibility of such material formation, and later in the laboratory.

It is now discovered that human perception of materials has reached the next level, and what kind of revolution these materials will eventually bring to cloud computing. Just now Mr. Michael Jordan also talked about the computer doing some problems can do very well, but to do some other problems is very slow, for example, you give a large number of two to multiply it, the classic computer can do very quickly, but I give you a large number, you have to break it into two numbers of the product, This computer is very difficult to calculate, it can only use the exhaustive method to see whether this number can be 2 apart, by 3 apart, by 5 apart, by 7 apart, so that it is poor to lift it again, this process is very slow.

But nature has a very magical and perfect world, this is the so-called quantum world, the quantum world has a very wonderful phenomenon, for example, there is an elementary particle, it to pass through two holes, in the back of the hole to form a disturbance of the stripes, a classical particles either from the left through, or from the right through, But quantum particles pass through two of holes at the same time, this is a very magical quantum world, which shows that the quantum world has a parallelism, we can use this quantum parallelism to do a calculation, so once the quantum computer is formed, it is equal to the use of quantum parallelism, a poor lift all the possibilities, So this is a very worthy of our pursuit, a perfect quantum world,

I've had a very important scientific discovery lately, the biggest problem with quantum computers is that qubits are susceptible to the environment, and we propose a completely new idea to think of a quantum bit as the smallest unit, we split it in half, far away, and the surrounding environment interacts, It is impossible to influence the same direction at the same time, so it is impossible to destroy this qubit. It is thought that quantum computers can be produced in the next 50 years, but perhaps there is a scientific development in which we can actually push quantum computing into application scenarios.

Next I want to share with you my opinion about AI. Let me give you a metaphor, for example, when we humans see birds flying, we feel very magical, we also want to learn birds fly, we start to fly the way is imitation, is also a kind of bionics, is that we tie two wings on the arm, think they can fly up, this is a simple bionic, But then we really understood the scientific principles of flight, is one of the most fundamental fluid mechanics of physics, so that once there is a theoretical understanding of fluid mechanics, so that we can design the plane, it must be far better than the bird, but it does not necessarily resemble the bird, so this plane is really due to the understanding of the basis of our theory , and it really pushes the development of it, not the simple bionic.

In this sense, I think AI is now in a relatively simple bionic phase, we still use a neuron simply to imitate a brain, I think the next big development of artificial intelligence, like from seeing birds fly into airplanes, we must on the basis of theory, thoroughly understand the theoretical basis of intelligence, Once understood, we are able to make some very magical new technologies.

Besides being a professor at Stanford University, I'm also investing in technology, especially in colleges and universities in the United States, where people know that it's possible to be a great target in the field of artificial intelligence, but when you're doing it, you're simply imitating the technology that Google creates. It requires a three-dimensional map of high definition and a laser radar. I think this is completely unnecessary, we can ask a simple question, because people can also drive, people do not need to drive a high three-dimensional map, the human brain will not emit laser radar. So we voted for a company that was developed by a very famous Princeton professor who used a common device in a car without the need for laser radar and three-dimensional maps to achieve unmanned driving, which is truly revolutionary. We need to look at the forefront of artificial intelligence with a scientific perspective, we want to see our technical route must have to be scalable, because the LIDAR and HD three-dimensional maps are not scalable, people do not need these, so in principle, computer vision to a certain extent, you can achieve this scenario.

I told you that artificial intelligence needed very good algorithms, and that Moore's law needed to be pushed forward, and we contributed in all two areas. But the most needed for artificial intelligence is big data, now we need big data in the areas of finance, education, and health, but it's not possible to have a very good market for big data, because some people have the data, but the data has its privacy, especially in terms of finance and health. But we have a very good artificial intelligence algorithm, so in the normal environment, it is impossible to have a good market for data providers and data analysts, in this market can solve the problem of privacy problems.

In the algorithm recently, academia has a major invention, its English name is called Homomorphic encryption, Chinese is called the same state encryption, we can do an operation on the encrypted data, so the analytic ability does not necessarily have to understand the data itself, the provider of data, although not analytical ability, His privacy was protected, and I felt the need for such technology. I think we must use technical methods, rather than the legal approach, to solve the problem of personal privacy, in order to enable us to achieve a large market data, large data must have a large market, the big market must have a technical route to truly protect our field.

Before you have talked about how artificial intelligence can really surpass people, there are different kinds of statements, in front of Michael Jordan professor also mentioned, for example, the understanding of natural language. But natural language is a human language, we use computers to measure, just as we measure airplanes than birds fly better, not because the plane is like a bird, but it is better in machine learning than birds, we have to find a criterion, on which day to really say that artificial intelligence has surpassed people. Today I gave them a vision of how to measure AI over people. Because the world of nature is an objective world, the reason why people have the development of science and technology today, because we have discovered the laws of these sciences, such as Einstein wrote the 1=MC2, Einstein is a person, which day the machine can write a new discovery, can be in front of people, For example, it predicts a gravitational wave phenomenon, we prove that the computer is indeed right, and more than people, this day is the real artificial intelligence. But this is a very great prospect, I think that day is not too far, I think in the next 10 years, 20, we can find in the artificial intelligence, we can find a new scientific theorem, we will feel that the discovery of this machine, a bit like we worship Albert Einstein's discovery, To worship the scientific discoveries of artificial intelligence.

In addition, I would like to mention a very exciting field of artificial intelligence, the development of which must have a very close association between academia and industry. Today we come to the Tencent Cloud computing summit, we feel very honored, this afternoon we will hear a lot of industry about artificial intelligence, but I invited Michael Jordan professor and me, the past development of artificial intelligence is due to the industry and academia very close union, I am a professor at Stanford University, in fact, I also made a small contribution to cloud computing, I was an angel investor, I invested in a cloud computing company, the whole cloud computing core technology is provided by this company. This is a very exciting time and we need to discuss how to use a new mechanism to enable academia and industry to collaborate more closely.

In this day and age, artificial intelligence has brought us a lot of imaginary space, the rise of artificial intelligence as a new intelligent species and the rise of China almost at the same time, academia needs to make great contributions, I, as a representative of academia, very honored to come to the General Assembly of Tencent, I hope we in this collision also produced a new spark, Thank you.

SOURCE | Lei Feng Net
Finishing | Wilson Hin
Edit | Roginmio