He thinks, the premise is to understand the deep learning of artificial intelligence, machine is to imitate the human brain to learn. In 1985, humans have put forward the artificial intelligence, but can't do that, then calculate the amount is too big, and now the technology is mature, mainly displays in the ascension of computing power. In this context, to locate yourself.
In artificial intelligence and the practice union, whether can reduce the level of artificial intelligence. Artificial intelligence has three levels:
Based on this background, what kind of person easy to be replaced? If human relatively closed environment, the work is the result of more standardization, information demand is less, then the more replaced by machines. In contrast, the broader cognitive boundary, demand information, the more of this work, it is not easy to be replaced.
Imagination is the machine can't replace. But if scientists to create life attitude is to do it, it won't be the same, to understand the artificial intelligence and create life this is not the same concept. In the future, the progress of human is together with the machine, if we clear goal, the environment is closed, the machine will be instead of some of our hard work, liberating us, technology will bring and the fusion of people, let us raise living standards.
The following is wang xiaochuan share points record:
Today, I looked up first saw is we speak of "ecological" new hardware. We in a word, called intelligent hardware, in fact, every time I mentioned the word I anxiety instead, we want to understand the word sometimes got no do this thing, can bring a lot of risk.
As we did a lot of companies in a technology, this technology can do, can't do anything, no judgement, or fear, or produce blind worship, don't know what's the meaning, may the Angle of investment or doing things. When we went back to the "new hardware" instead, give me space to interpret what is intelligence.
March 8, began a week of time have an Alpha Go lee se-dol man-machine war, along with the top technology companies fought a battle. Before the game is very interesting, we review, but Google released a I am excited, because two years ago, I saw the development of their biological science, it is a pity that you didn't have the aura saved up, my colleagues and with tsinghua lab would talk about the idea, but do not go the matter, said it was too difficult.
Not involved in this matter is quite sad things, there are a lot of friends say you how positive in this matter, including in February on zhihu said Google would have write files, didn't participate, onlookers can always, so the things more involved in the state of mind. Before the game when I found that most people, there are two kinds of people, the first kind of people is a go player, especially in the world competition to paragraph 9 players, including wei-ping nie.
I attended two innings in five innings chess, I tell you to go I understand different, although know the rules, but there is no way to judge the situation is good, so the chess game I saw one thing will know who will win the game. I will look at the coach's face, and his face more ugly machine the odds, the greater the last coach crash, finally won the machine.
I speak this example is that we seem to be facing some threats, ever good at thinking began to invade by machines. A go player is a machine to your success and proud of things instead of, this is what a kind of fear. I think you more or less likely to experience a little later, including tencent started to change BUG, we may have this kind of pressure.
Before the game, a lot is the representative figure of the Internet, and even have a technical representative of people believe that people will win, the machine can't win, a dozen people, including our company we ask, what people give my answer will be the future machines will win, but the people will win. They think go there will be a difference, think later machine, there is no problem, but the timing was not mature now, unfortunately, the machine did win over the people.
When we work to remember that not enough feelings, even if to do science and technology of people did not think of this all of a sudden, so this is the thing to feel.
But I think we don't have fear or a romantic caring, we know what it can do, can't do anything, it is helpful to our attitude towards life and work. Not to say that this thing how to make money, a lot of friends at the end of the day is business opportunities, how we understand the matter, including one's own ascension, our understanding of the machine could be more in the long run.
Don't know if you have heard the word "deep learning". Most have heard, because the word like Alpha Go, speak to a mysterious concept in particular, is the depth of the machine learning or intelligence.
Deep learning is about two concepts, the first concept is inside the machine learning neural language was used to simulate the brain language model for training or machine identification, is your input into vector, intermediate results through iteration do. The second is the network result of iterative deep, is not a layer can be done, need to multilayer. Later the researchers improve the model.
This concept proposed early in 1985, this theory has been mature, it is a back propagation, the machine how to training the already have. Then there is a problem with this machine, amount of calculation is too big, can't do. Ten several nodes at the time when machine is not enough, but now the largest theoretical system is not change, but the calculations of ascension.
This also tells us, from the theory of artificial intelligence to the deep learning approach, including the machines before understanding has been mature, today we use the method of no more than when the theoretical framework and computing model. This is not a new thing.
What has changed? Actually change the two things, one thing is a tremendous improvements in the computing power. Alpha Go machine force is calculated by deep blue 25000 times. The second is that we collected a large amount of data, data acquisition is difficult, there are big data, data have how old? Actually go there is not much, basically data basically with 300000 units have been under go for training.
Without the Internet did not dare to think, after the Internet, go abroad website have corresponding data, 300000 units, each about 100 steps, so a total of 30 million positions do the training. How do we put this theory to go, this is Google's innovation. First thing CNN network (sound), use the method of point at the picture to play chess, said before pieces is the logical analysis, rather than the network.
Just like looking at photographs board now, so the machine have move feeling. The last five years has one of the biggest improvement is face recognition, before is completely don't know how, to identify their eyebrows? Identify the eyes? Everyone here may have write programs, it with what rules do you want to go to describe the person's face, but today we use the feeling of CNN image did.
So Google's first CNN network was carried out on the machine is used to describe the innovation, makes the machine has the feeling of the body. The second is associated with deep search as a rational way, combination with CNN's perceptual, this is the second innovation. The third thing is to use powerful learning, let the machine with himself, as machines become smarter after you can talk with myself, so in this promotion. It does not bring theoretical breakthrough, but made great contribution in innovative applications. So Alpha Go behind the victory is a blend of the engineer's ability in again.
It's really important is what place? Is not the technology itself, but to all people in the care of our own positioning. I compare the activities of this week as the original decades the results of the Renaissance. A week past, I like a lifetime ago. You know what a week machine understanding and attitude, has changed a lot after a week.
Our relationship between person and person, including we read the bible and human equality, closer. Now how do we see the machine? Before the game we think machines are stupid and what to do, after the game there are two important changes, the first is the ability of our machine with higher evaluation, the machine can be overcome. Said when we see a doctor, before you make a film, machine diagnosis, there is nothing to tell you do we find it difficult to accept, very don't believe it.
Machine now tell you what a result, we may feel better than people. But can think of this change? Our machine because after this event, the ability to have a great recognition. This makes it more engineers and more entrepreneurial company, more capital to artificial intelligence.
I joke that a-share artificial intelligence concepts may continue for several harden board, we see the faith of artificial intelligence. But very skillful is Alpha Go not five innings are victory, lost the game, but everyone still turn to. It represents the whole excel common victory, is the machine into the other factions, we still have their own dignity.
Think of the movie "independence day" twenty years ago, when human pilots are faced with the spacecraft, the spacecraft in the spacecraft eroded, so in the face of the machine we have dignity. More people, many of our young people begin to take Alpha Go call a dog. 90 00 or after we think young people will think robot will become friends, there are also called the teacher.
Machine is not actually to invincible, the first is to believe in it, the second to accept it. To reject it is very difficult, we have to accept it. This is our key to finishing the man-machine war, how we use and how to make friends with it, later, so it's very important to the enlightenment.
Technological progress has three levels, no matter from software to hardware. What is the three level? Traditional earliest is intelligent, is actually the rules to the machines. For example, we do an electric rice cooker, refrigerator, intelligent it doing here? Our programmers to write programs, when the temperature to 103 degrees when I tripped. In fact we can put things complicated enough to the machines, the wisdom of human to the machines. Unfortunately, teachers tell students if give his rules, his ability will decline, so this time machine behind people, intelligence less than one.
There is a kind of situation, we do not know what are the rules, we are feeling. Like just about face recognition, this is a classic problem. Everyone thinks is very simple, may face blindness struggling a bit, but there is no problem, most don't like everyone to learn a foreign language so hard, just may be memory, recognition: no problem. When the problem to the machine we encountered obstacles, decades in the aspect of image recognition we struggling, abroad have done the data points, progress is very slow, every year make images of people basically couldn't find a job, because is not practical.
For a long time with artificial intelligence theory is disjointed. A few years ago we chat with tsinghua do academician of artificial intelligence, said the artificial intelligence in combination with practice is low, because the connection is not on, but not now, now connected to a, why? Because we came to the second stage, we began to speak rules with the machine. Depth study of the beauty is corresponding to the machines, we put the question and answer told him it was zhang SAN's face, the face is li si, don't tell why he is zhang SAN and li si machine.
Through a large amount of data training can learn to machine. Just like we teach children, through these methods, step by step, and here we give to the machine, along with all the images and sound field has been very good. Last year began to face recognition machine over the images, the accuracy of more than one times. We can tell the machine, the machine can learn myself. It is not enough, there are even some problems we even the answer is not enough. We said 30 million steps on go to learn the answer, the basic rule of machine learns to play chess, take 6 to 9 paragraphs way to tell it, it reached the 6 segment level.
Google later for the two random machine, after not tell you win or lose, machine through optimization algorithm to find a better answer. The second thing is the answer to the machine, the third is to tell you machine to give you an answer, I evaluate the answer is good or bad. This is three levels of progress. Especially the third thing, people like Google or Microsoft's top team, and may even have a religious color, when we give the machine a goal, a machine if I learn to find rules and answers, I just give him a target, instead of telling him what to do, this is very important step in evolution.
I saw a literature, they want to training a Alpha Go machine, a began to not let it learn how, starting with two pieces of white paper of Alpha Go, with himself, only tell it goal is to win, check, can see the training of new players. I think it is very meaningful.
A person learned all powers in central plains, it will converge, and then to improve. There was another man who had never been to the earth and the central plains, to learn martial arts, you said it's skill with people? This is the curious to see what wisdom will grow again. This is made of three layers.
Based on three levels just now, let's think, what kind of person what kind of occupation, work itself is more likely to be replaced? We can be regarded as a chance, also can be seen as a challenge to itself.
There are two easy to replace, the first is closed your work and environment. Means that when making decisions, decision making is closed and limited sources of information and even is structured. Such as chess this decision is very closed, just need to know the rules of the anticipation on can be decided.
The doctor will be a lot of hard, your doctor will know the patient's medical history and the current state of the patient; But as a teacher may face more complex environment and make decisions that source may be enough to open, your answer the standard the more likely it is to be replaced by machines. The second thing is the goal, the first thing is you deal with the information of the open or closed. From this we know, some decision-making information need machine is easy to do less, information need for the machine can do.
We do a translation or a writer, need a lot of life experience, the author is to read thousands of books, the view of practice. So open environment for machine is a big challenge. On the other hand, if the open the answer has to do with you, the machine is easy to do. That is whether the machine can do a good job, people will replace a standard.
Back to the question of whether people will be replaced, what is the concept? Person's goal is to own survival or reproduction, the machine is more simple, I do the diagnosis, the next chess, or to do a speech recognition. Man has come to the expansion of the large space, our machines are only in a limited space to work, basically see the training of the machine space is how much, if the algorithm is again good also cannot from the target and the scope of the machine to adapt so today's technology is far less than this step, the second is we're not going to build a machine to set a goal how to survive, will not say to adapt to the environment has a particularly large space.
We even have the ability, also have no incentive to make a person to replace the machine, we don't think the machine itself will evolve to a survival ability. On the other hand, if have ambitious scientists to do one thing, to say to want to create a smart machine, this machine has the concept of survival, can in the face of the whole earth environment.
Artificial intelligence in fact we are not doing, we are creating a kind of life. So you want to clear the concept, if your attitude toward the creation of life to do, the machine may have a kind of life consciousness, know their presence. In turn, today we can rest assured, our goal to do these things simple enough, such as Alpha Go machine, the board from 19 lines into 20 x 20, human beings can understand and learn, but the environment changed, Alpha Go nothing.
Another problem is the imagination, the difference in human and animal. There is a book called "a brief history of mankind, this is related to the historical development, it is also a path, this is I don't think people will be replaced at the core of the two judgment standard.
Artificial intelligence and what is the relationship? Have the technology we may become more powerful, but it is also possible technology let us become more weak. Here a lot of people wear glasses, glasses is a technology, when using the glasses after our vision to become better, more powerful, but leave the glasses we are weaker.
We're leaving technology after the isolated line not line, put down the phone, pad, car transport is better? We found to be weak. We become weakened due to machine, we mastered the energy strength was replaced, today's planting is replaced by a machine.
After goals clear, relatively closed environment, the machine can do we can to it, we can do it using Google search engine, by mobile phone into a clairvoyant clairaudient, it is a trend. Future apparel equipment may become implanted, like Google glasses, a lot of people eye myopia also can have a lot of people try, not including the young girl will cosmetic, these things will bring new implants.
From the artificial intelligence fusion with people that I think will bring new species, you need not fear, you ask a monkey you will become a person? Through our understanding of the artificial intelligence and the understanding of technology, technology can bring with the fusion of people, can put the person's ability to ascend, may also reduce the person of ability, it is our future evolution.