Artificial Intelligence Keynote Speech

Netease Technology News July 9 news, a few days ago Zhongguancun business forum, Digital Chief Scientist Xie Yun of artificial intelligence conducted a keynote speech; speech, Xie Yun shared several stages of artificial intelligence research and said, The second key shift in the IT industry came as computing speeds grew faster, "and there was another major turning point in the path to making computers faster and faster - today's foundational technology platform In most cases rich enough, no longer a bottleneck in the application.

Xie Yun believes that in 1956, a group of cattle from neurology, logic, mathematics and computers, etc., opened two rounds in the United States to come up with the concept of artificial intelligence, which is a symbolic time of artificial intelligence research; In the 1980s, scientists always thought that to simulate the human brain as a network, intelligence would emerge. However, this hit a major setback in the 1990s. In the 1990s, a historical turning point occurred in the IT industry, that is, From the pursuit of intelligence back to the pursuit of computing power, and so far, the IT industry is still making the machine faster and faster, the machine is smaller and smaller, using the machine to connect all the equipment; Therefore, Yun think "Computers can solve everything that can be calculated, and computers can not be solved if one does not have the ability to turn one problem into a computational one."

Xie Yun believes that big data has triggered an infinite reverie of intelligence. "Everyone feels that the cost of using cloud computing can be reduced. In fact, this can be the reverse because the cost reduction can only be provided by cloud computing today. The nature of cloud computing is a change in the service level of the industry, not because cloud computing technology has broken through and started to engage in cloud computing, but because the technologies and products in it constantly improve, resulting in a substantial reduction in the total computing resources, communication resources, and storage resource costs So, today started to engage in cloud computing, which is the logic of the real development of the industry.My big data is the same, because the omnipresence of big data, it is because computing resources are very cheap, the processor is very cheap, so it will produce A lot of data.I alone brought several computers, and now the phone is a computer, now the ability of mobile computers, compared to the astronomical investment in 1980 made the first Galaxy One giant computer's ability to fast More than a single order of magnitude, today's computing resources are so rich that they are generated Computing and big data. "

The following is Xie Yun speech Record:

Ladies and gentlemen, everyone! Just painted the teacher big data situation with everyone to do a good introduction, and Tu teacher coming to an end, talked about big data will enter into a smart society. My topic begins with intelligence.

Earlier this year, I attended a reading session of a young friend. One of our friends from the Central Organization Department asked me a question. "Thank you, Teacher, now all the big data age. Can you design a system for us? The material and so on all the information in the system, and then you can tell me this person can not be deputy minister, future minister? "Big data does trigger this infinite reverie of intelligence. At that time I gave him the answer, if one thing no one knows how to do it, in fact, the computer will not do for you. Do you understand this sentence? The friends here may be very strange, computer do a lot of things I do not know how to do ah, not for me to do Well. But there must be somebody who knows how to do this and the computer will do it.

Why talk about this topic? That is, big data has triggered a lot of talk about intelligence, so I wanted to talk about this topic today to see if I can make it clear. This is also my personal point of view, not necessarily for your reference.

In fact, the tools created by mankind are divided into two categories, one is aided by physical tools, and computers are actually used as a computational tool to assist people in their intelligent activities. Can not replace smart activities later, this is the dream activity of many people, I will do some analysis below.

Because computers are a tool to assist in intelligence, and computing itself is actually part of a person's intellectual activity. After the birth of computers in the 1940s, the entire industry soon turned its interest in making computable machines to making intelligent machines. An iconic time for this matter was that in 1956 a group of cattle from neurology, logic, math computers, etc., two in the United States, came up with the concept of artificial intelligence. At that time, mankind was still very optimistic about making intelligent machines, which are two typical cattle. One is dead, called hebert, he won the Nobel future learning, 1965 20 years to say the machine can do all the work one can do. Another cow, still alive, called Minsky, said in 1970 that in a period of three to eight years he will have a machine with average human intelligence, what are they saying? We all know that the previous period when a U.S. machine won the same question, and one machine passed the Turing test, did they say that it was not such a machine? No, what we're seeing now is not what we talked about when we talk about artificial intelligence.

What does it mean when the machine can do all the work a person can do? This means that when scientists wanted to build a machine, the machine built some mechanisms from the beginning, just like people, but it was ignorant and did not know anything about it. Let this machine and the outside world can learn to interact with, can have the wisdom, to obtain knowledge, this is the most primitive idea when discussing the intelligence of the year. In the past that this machine can be made to spend 20 to 20 years, the result is what? This effort suffered a major setback in the 1990s. The next platform is Japan's fifth-generation computer, started in 1981, when it caused a sensation worldwide. At that time, the computer was built on Japan's all-powerful power. They believed that mankind has accumulated a great deal of experience in the field of artificial intelligence since the nineties. After a decade, I can create such a machine. This machine only needs the built-in basic mechanism and artificial interface, it will gain wisdom and knowledge by itself, and does not need you to program. By 1990, this project completely failed.

The artificial neural network in the 1980s, which claims to simulate the brain of a human being as a network, is intelligent. When I was a Ph.D. who was studying this issue, it cooled down around 1990 and proved to be unable to meet our expectations. Therefore, in fact, the IT industry as a whole has undergone a historic turning point in 1990. That is, the pursuit of intelligence has come back to the pursuit of computing power. Found that a computer made with the same human brain does not fly, what fly? Let the computer faster and more reliable. The pattern we see today in the IT industry was such that a fundamental turnaround took place in 1990 when the quest for intelligence was very hot.

This is what we see today IT industry is doing? Is to make the machine more quickly, the more the smaller the machine, the machine can connect all the equipment, and why? Because the computer is getting smaller and smaller. Computer core is still doing calculations, whenever it can become a computational problem, the computer can be solved. If a person does not have the ability to turn a problem into a computational problem, the computer can not solve it. People from the Central Organization Department just asked me if a director can serve as deputy minister. If no one can treat this as a problem of calculation, I am sorry that I can not solve the problem. It is of no use for computer to charge much data. Someone must know how to do this. Computer can count.

As the faster and faster the computer, so today the entire IT industry ushered in the second key change is to make the computer faster and faster on the road there has been a major turning point, what turning point? It is today that the basic technology platform is rich enough in the vast majority of cases and no longer a bottleneck in application. Therefore, the IT industry will look to the future from now on. The stage where the basic technology platform dominates the development of the entire industry has come to an end. In the nineties buy PC will ask how much frequency, frequency not enough, some vendors say you can give you frequency, and why? Because the computing power is not enough. But until today to buy a cell phone or Pad when the clock frequency is how much? Nobody cares, why? The capabilities of the underlying technology platform are already abundant enough in the vast majority of cases and are not a bottleneck in application constraints. Starting today in the IT industry, the basic technology platform will no longer dominate the future development of the industry, the real dominate the future development of the industry will be the application. Why did this happen? It is because the development of integrated circuit technology led to this.

Now that ICs have a minimum line width of 22 nanometers, there are hundreds of millions of transistors that can be placed on a pin's needles, and why they are so small has been made in three dimensions. The original transistor is planar two-dimensional, and now it is three-dimensional.

Going further, the integrated circuit can do much smaller? I talked to Intel people who said that a minimum of 2 nanometers can be reduced by an order of magnitude in the next ten years. Silicon will come to an end in ten years and can be used for something else.

The second turning point in the IT industry is because integrated circuits come to this level, and today the IT industry really entered the youth, that is, it really started the golden age. Before that no matter what incredible miracles the IT industry has created, it is a prelude. The real miracle will be developed in the future, the grand prospect of the entire IT industry has just opened a corner. Nowadays many people are talking about the substitution of industries. Who is driving the development of society after IT? I think it is too early to talk about this topic. The IT industry has just begun to have an impact on the entire society.

Take a very simple example, "Avatar" have seen it right? IT in the subversion of a very traditional industry, is our movie, you can do without an actor. Can not be done before, why? Because of the need for supercomputing capabilities, the original basic computing power can not, it can be very expensive. Now that the basic computing power is rich enough, I can make it out of my computer completely without the need for an actor. In the future, it is possible for an actor to do business. This is a very typical example. IT in the future will have in all areas may not be expected today phenomenon.

In addition the pattern of the entire industry will not be like today, such as the database by Oracle to rule, as long as you want to do this thing we must use Oracle. The future will spread, the surface of the bulk will be very large, in all areas will have a unique and unique technology companies in a real application to do a good job. This is the basis for me to see the IT industry in the future, and it will really mature when the underlying technologies in one area are no longer the bottleneck. Our automobile industry is like that. Our aviation industry is like that. For example, Boeing and Airbus are doing What? Do applications, it is not to do the core technology, does not produce the engine. Airbus wants to create any aircraft, the engine company can give you a suitable engine, which is the basic technology, the core technology is no longer a bottleneck, the industry is really prosperous.

Today, I think IT industry has just entered the threshold, the future will be very lively, we can do well in this industry.

For this reason, big data and cloud computing have been created. We all think that the cost of cloud computing can be reduced. In fact, this can be the reverse because the cost reduction can only be provided by cloud computing today. The nature of cloud computing is a change in the service level of the industry, not because cloud computing technology has broken through and started to engage in cloud computing, but because the technologies and products in it constantly improve, resulting in a substantial reduction in the total computing resources, communication resources, and storage resource costs So, today began to engage in cloud computing, this is the real development of this industry logic. Big data is the same, because big data is generated everywhere, because computing resources are cheap and processors are cheap, so a lot of data is generated. I brought myself several computers, and now the mobile phone is also a computer. Now the capacity of mobile computers is more than the capacity of the first Galaxy One giant computer in comparison with the astronomical investment made in 1980. Level, today's computing resources is so rich, so have a cloud computing and big data.

Big data leads everyone to the reverie of intelligence, the following specialized talk about smart, how do I see smart, including in the big data environment in the future how to do. These two figures on the one hand is the computer system, while the human system, the human brain is not calculated? There is no way to answer this question now. We do not know whether the underlying mechanism in the human brain is calculated or not. However, the computer is very simple. The lowest level mechanism in the computer is computing. The bottom is computing power. Go up there is an intermediate layer, the middle layer is between the computing power and application, including the operating system, database, etc., itself is not directly applied, it has an intermediate layer. Above the middle layer is the various applications we see, what is the brain? The bottom of the brain is a neuron, a neuron connected together, psychophysiologists can now measure these signals, but it is not clear exactly whether it is calculated, but can measure some of the signals.

There is another layer in the brain, and this layer is even more unclear. Above this level is the usual performance. We speak, study, and have various psychological processes, namely the so-called psychology Consciousness process, this process can be monitored, analyzed, can also be seen. What is the most important thing in the middle of the brain? 2009 Central Ten put a program is talking about the semi-brain, the child got sick at the age of one year, and finally the doctor said he had meningitis, in fact, not, but he had epilepsy. At the beginning of teens, the child started to have epilepsy, which has been normal and without any major problems. Epilepsy convulsions is ventilation, we say it is claw wind. It turned out to be a year to a year and a half, and later intervals shorter and shorter, once or two months guilty once, to seventeen-year-old parents can not stand, take him to the hospital for testing and found the child is epilepsy. The most interesting is this photo, we can see the left brain has shrunk, can not see. The doctor finally decided to give him a craniotomy, what happened? Cut his left brain, cut off after what happened? This child is basically still normal. During the period of childhood illness, the brain has been completely rebuilt. Because the brain is divided between labor and logic, the most important thing is that the right-hand movement is controlled by the left brain and the left-hand movement is controlled by the right brain. Therefore, people After stroke, see which side of the head stroke, this side of the head stroke, limb movement on this side are affected. But now the doctor cut the entire brain half without being affected? What does this mean? That the brain has a middle layer. After the underlying neuronal layer fails, like a cloud computing center a cloud platform fault reconstruction, to ensure that the system above the application can run normally, in fact, the brain has such a function, that is, there is a layer in the middle. In the middle of this layer is what? In fact, now neuropsychology, physiologists do not know.

Why do you say this? Our understanding of the essence of intelligence today is still very superficial, and understanding of the whole process is very superficial to what happens in the human brain. Including many winter scientists in the world who passed the so-called artificial intelligence in the nineties. Now that the era of big data has begun to take shape, it is trying to construct a machine that claims to be as smart, learning, independent and intelligent as people. I heard that a company put a machine like this into the university classroom to follow the students to study and think it is still quite possible to create machines that are as intelligent as human beings. I think this ideal is still very far away and I can not know whether it can be made in the future. However, from the limitations of human understanding of the entire intelligent process, I think we are not enough to support such a system.

The so-called intelligence we are able to achieve now, or just above, is the intelligent activity of people who can be observed, measured, analyzed. I am sorry, people's intelligence activities is more important is the bottom of this layer, why? Just now I said that this child has just been born ignorant. There is nothing in his mind to grow up solely in interaction with the outside world. But today's computer can not, we have to build a lot of things.

Here are some basic ways in which people are smart on the road, in fact, there are three. The first is formal imitation, I just said that the artificial neural network, what does that mean? We found that human neurons have a large number of neurons, a large number of connections between them. A group of people who started artificial intelligence from the fifties believed that I also made some very simple processing units and then connected a large number of processing units together, which I think might be smart. This sentence sounds a bit joking, we do not surprise, scientists often do not fly things. If you read the history of science carefully, in fact, there are many unscrupulous things in the history of scientific development, including that famous scientists also do very poor things. Therefore, we must remember not to turn science into a religious superstition.

This form of things to imitate so far did not produce very good results, the so-called result is a non-linear function, the nonlinear function of the sample set according to the smallest difference to fit the process, there are hundreds of years, so the form does not imitate Results.

The real success is the last two, namely the mechanism of imitation and mechanism of substitution. What is the mechanism to imitate? We do one thing entirely in accordance with the way the human brain do things, such as playing chess, such as the computer can challenge the world champion, won the world champion in Russia then, the machine how to play chess? Will be deduced, because the calculation is much faster than the process of deduction, and finally won the world champion. This is achieved through the mechanism of imitation to achieve specific activities, the premise is to understand this mechanism.

The second is the mechanism instead. This person how to do, I may not know. For example, human vision, human visual process is still very complicated, do not know what the final visual process. It does not matter, I can use another mechanism to achieve the same effect, so the process of handwriting recognition and our identification of Chinese characters is not the same, we identify the Chinese character is to take a Chinese character to identify. Handwriting recognition will record the entire process of writing, the use of writing process information and stroke information to identify, so the whole mechanism and we are not the same, but the effect can be a drink.

These three are the core of the three ways to achieve intelligence, the two are the most important.

The following is a mechanism to achieve intelligent how to go, this and big data have a relationship. At the very beginning of realizing intelligence, people hope to invent some basic mechanisms. With these basic mechanisms, machines and people can learn smartly through the interaction with the outside world. This is typical of Japan's fifth-generation computers. However, This idea later hit badly beaten, it turned out to be impossible. In the seventies and eighties of last century in pursuit of intelligence, people took a step back. It seems that we have no way to understand what is the underlying mechanism of human intelligence and can not be achieved by this means. What is a step back? It is a hybrid way to put people's knowledge into the machine and add some mechanism to realize intelligence. Machine is actually prefabricated people's knowledge in it, what is this? This is called an expert system, which was very hot in the 1980s. This time we will find the mechanism and information combined, and the data began to combine, knowledge is a kind of data, a kind of information.

Today, we are talking about the application of big data and so on. Actually, this is the third step. We will have some mechanism, coupled with knowledge, plus a large amount of data to achieve our goal. This is the evolution of human beings in pursuit of intelligent process and mechanism to realize intelligent process. At first I just wanted to find some intelligent mechanism, and later found no, so these mechanisms coupled with knowledge and data to do. Throughout the process, the concept of intelligence is continuously relaxed and relaxed, so the intelligence we talk about today and the intelligence in the 1950s and the 1960s are actually not exactly the same. Because we hit the wall, so the intelligence to relax. Today when we talk about many smart applications and so on, many things are actually in history, such as these algorithms or mechanisms, and they are not historically regarded as intelligent things because at that time Smart things require relatively high. Just say that later can not be achieved, so these things have slowly become a smart thing, into the concept of intelligence.

Today we talk about the concept of intelligence is still a very broad concept, to accommodate a lot of traditional things, including the present things in it, which is the realization of the entire intelligent mechanism of evolution.

Today we have gone to the era of data + mechanism and knowledge. Looking back to the evolution of data in the history of computer applications, the computer starts with numbers and values. The data used in numerical calculation is the physical quantity. Just as the teacher applied the measurement of the external world, it is a kind of microscopic response to the local physical values. This is the data that the computer initially uses. Later, the computer into the so-called online transaction processing, online transactions, such as banking systems, ERP systems, aircraft ticketing system, this time the computer is used when? Using so-called structured data, complex than a single value, no longer a single micro-data, I said that the data into a meso. An electronic ticket is the most typical structured data, you can use a fixed field to describe all the air tickets, air tickets for all people are in this format, which is structured data. This is the application of computers from the fifties and sixties to today's mainstream status. The data used by computers are evolving, evolving from micro to meso, and their connotations are complicated. Today there are a lot of unstructured data to achieve, such as online chat and take photos and so on.

The two phases are very different from the first two phases. The data in the first two phases are tied to the concrete application. The data service is used in the application and the application needs you to collect the data. But today is not, today a large number of data appear may not have a specific purpose. For example, take pictures, used to think that this attraction is particularly good, let's take a leave to commemorate. This year is to take the camera to a photo here, where to want to general photo Zhang. We have a lot of data coming out and we do not know what it means to be worth, so we just say we can not dig something out of the data. This is the computer data changes.

Data and information I have been versatile here, I think the more complicated information, simple data, it is not careful study. Here is a very important concept to talk about when most people talk about big data today. Most people are talking about how to use the data. I want to talk about here, data, information from generation to use is divided into different stages. First of all, there is a generation phase acquisition, and then there are storage and transmission phase, storage, transmission is not immediately after the application, in fact, there is a management process, how to manage the information. The storage of information does not equal management, or storage is the most extensive management. But really want to make good use of information, in fact, information is the need for more complex management, according to the content of these information and logic really up. Oracle engaged in the database is to manage the data, the real effective management of information up.

Look back at the human brain and computer system, and then compare this picture. Computer system data and applications are one to one, each application has its own data, so we talk about computers today are an island of information. Such as banks have their own data, the bank's application is to use their own data, the application of the ticket is to use their own data, they have different data, each data is prepared for a specific application, which is today The most typical feature in computer applications.

We look at the human brain is not like this? The human brain is not like this. What is the information stored in our head? Is the whole. It is not in our heads that this message is used for this purpose, that it is for another application. With people under the brain there is a unified layer of information management, all of the information in the human brain is the overall structure of existence together, from the perspective of the human brain and computer biggest difference. When today's computer system has changed, after entering the so-called era, how do we manage information in the end? Or in accordance with the original way to put the data in accordance with the application up, or it is a block, let's piled up to say, dug again when used.

My suggestion here will be followed by another one of my ideas. Big data here would like to talk about what kind of technology in the future, what kind of application mode. I think it is very complicated and there will be many different approaches. I will say only one of them, not all of them. However, I think this approach may be very important. Even this practice has subverted the entire IT industry. However, it is not a panacea. It is not a panacea. Some people sell this medicine, but I do not sell it.

In the big data era information should be how to manage? Let's talk about the application, the first information on how to manage in the end. The traditional management is to manage the data according to application one by one. There is no clear application or explicit application of large amount of data in the era of big data, but the value of this data lies not only in the application of this point. In the end we should be how to control this data? Because good data management can have good data, even if the pipe is not good, talk about big data applications a little blind, or only in a very shallow level. I personally have a basic point of view. In this era, we should put these figures in terms of subject, people, business, city, and subject-matter definition. How to manage? According to the data, the logic of the reaction between the information, rather than stacked like a logic.

This is the picture I just saw. What should I do with an individual? The message in the human mind is not disorganized, the message is not a pile of junk in the mind, it is inherently logical and relevant. What should we control when we have personal information? Should we logically organize what we perceive as having these messages and finally form what? The virtual image that forms part of the human brain creates an image of a collection of human brain information in a computer. It is not a repository, nor is it a garbage collector.

It is a very challenging task to build this kind of image in order to put the information around the topic. You have to understand what the information is and the information is not a simple picture, it needs additional information to make it meaningful to use. At the same time, it is important to understand how the object comprehends the information, which information reflects the internal logic of your object. This is the mathematical structure for personal information that I spent three years building. This approach addresses a specific set of issues in the Big Data era and in the future of IT development, but it does not solve every problem.

Said these things may be too mysterious, in this way to effectively manage the information up, on the basis of the application of the system and the nature of the traditional structure in the end what is the difference? Here is a typical example of a smart city. This is a smart city management system in which you can see the management system of smart traffic, smart travel, urban safety, and so on. How to build a traditional IT system? For example, a smart transportation system collects data on its own, and a city security system collects data on its own, resulting in a location where multiple cameras can be put in the city because of different systems. This is a very traditional practice, that is, the data and application are closely bound together. The data is generated for application. The application also uses the data, so we complain about the information silos and so on.

If, in turn, the information is managed according to a city's main body, the entire system becomes like this, with the lowest level of all information gathering devices, including cameras, sensors, and virtual sensors included on the network to monitor activity throughout the network . All of the data is collected not by the application so that the application can use the data, but by organizing the data according to the internal logic of the various parts of the city. What has become organized together? Became a reflection of the city in the virtual world. This data is not a garbage dump, this data has the inherent meaning of the whole. Based on this data can be more fully utilized, you can do all kinds of analysis, a variety of simulation. And then the above application can go to the image to get a variety of information needed, rather than just the original confined to their own systems.

In this way the value of the information will be more fully tapped, and applications can get more complete and rich information, and changes in the upper application does not affect the underlying. So this approach is completely different from traditional IT system construction. Personally, I think many IT systems may be built in this way in the future. Of course, not all, and not all of the information is used to build a virtual image. But this can become a very core, or very important, application-built model that can have a major impact on the IT industry and IT applications. It will not only change the application, but will also generate some of the core technology, basic products, basic platform, the future of the entire IT industry may change far beyond our imagination.

Returning to the personal example, if the personal application is also built in this way, the bottom is a personal virtual image, the middle of which is a service that is fused on the basis of this image, with a more natural interaction at the interface layer. This is not a smart process that can not be seen in the lower part of the brain. It is designed to be used in conjunction with what I can feel and see the parts of the brain that work together seamlessly. This is the way in which applications are built in the future. Computers and You Not only in the human-computer interface, you will find that computers are very handy, and that the use of computers, especially the way computers store and manage information, And your thinking habits, lifestyle, will be very natural together. Because the information is organized and managed according to your habits and ways of thinking, information systems and human seams are smaller and ultimately form a seamless connection.

Into the entire network in the world what is the scene? In the entire online world may have different meanings, there are different applications from the main core structure of the system, these applications are based on the main virtual image-based, and finally the Internet world to connect them together. Therefore, the online world will become more orderly and more like a real-world image, not as chaotic as today.

Computer application, Tu Zipei teacher just talked about driverless cars, but also the car and people seamless integration of car driving and personal information management system will be combined, information applications and information control systems will be seamlessly connected. The most typical seamless connection took this year's World Cup took a disabled kick-off, I did not find the video. This is a manipulator, this hand is directly connected to the human nerve, is completely controlled by the nerve, the hand has entered the clinical application stage. Take a grapes in this hand, grapes will not be crumbed.

Future development if summed up in one sentence, that is, to assist human physical tools and auxiliary human intelligence tools, and people are gradually seamlessly pick up, and this is what I think the future of intelligence. Not to say that we made a machine so that it was just like me, same intelligence as me, and finally challenged me. I think this goal may be realized for hundreds of years, but in the coming decades and hundreds of years later, I think that intelligence will go more in this direction instead of creating Independent personality and people have exactly the same ability of the machine. No matter how big the data is, I think this goal is hard to come by so far.

I probably take the time to understand the whole process of intelligence to tell you some, are not necessarily personal point of view, for your reference. And certainly not the package cure medicine, but I think the IT industry in the future is very important main line, and this main line will have a great impact on the entire industry.

thank you all!