Chen Pei, chairman and president of Zhongsou Network, is a famous artificial intelligence expert, executive director of China Computer Association, third-generation search chief designer, and also an amateur champion of Beijing Go.
Dear friends, good evening, I am Chen Pei. The theme shared for you tonight is "Will the machine rule humans? See the rise of artificial intelligence from the human-machine war.
We all know that the man-machine battle in 2017 has been opened. In the first game of Ke Jie and AlphaGo, we have all seen the result. AlphaGo1/4 won Ke Jie. In fact, 1/4 of the game is the smallest one in Win, there is no smaller difference between winning and losing, but the game is actually one-sided. In fact, AlphaGo has always occupied an advantage. Although Ke Jie stepped hard in the official stage and tried his best, he finally lost the game.
Before this, the media asked me "Ke Jie wins the geometry?", I am sorry to predict that "Ke Jie is negative in all three battles." Today is the first set and we have a chance to watch the second and third sets. But the game itself is very interesting. In my opinion, it is beyond the scope of Go. It is not a simple Go game, but a competition between humans and the machines we humans make, but this competition is A very unique field is unfolding in Go.
Let me share this theme, I personally think it is quite appropriate. Because I personally started playing chess from the age of six or seven, won the championship of Zhejiang University, won the championship of the General Staff, and won the championship of Beijing Amateur Go, so I have some understanding of Go, it should be said that A better person in the amateur. On the other hand, I have also done a lot of artificial intelligence projects, including expert systems. In particular, I have also written a Gobang chess program to express and use human knowledge in a computer language, and then compete with humans. Therefore, the battle between AlphaGo and Ke Jie has aroused the attention of the world, and it has also aroused my personal interest. I think that through this matter, we can better see the rise of artificial intelligence.
Repeated several "human-machine wars"
Artificial intelligence is a hot spot nowadays, and it has become a key word for brushing. In fact, the game between the machine and the Go is not the first time. In the early days, there was such a game of Go and the game of the people. There was even a computer game in the game every year, but it did not cause a sensation like this. One of the biggest reasons I think is that the rise of artificial intelligence has brought about a significant increase in Go software and has begun to challenge the top players in humanity.
In 1997, IBM's deep blue victory over human chess champion Kasparov, it should be said that the machine challenged human chess games and defeated the top human masters, but at that time did not particularly cause the sensation of the entire human race and attention. At that time, it was a game between IBM's IS6000 and human world champion Kasparov. This game did use machines, but it did not use artificial intelligence technology. It used more hardware acceleration and violent calculation. , defeated the champion of mankind. Chess and chess are much worse in magnitude than the Go, so that the machine has the chance to defeat the human world champion through its own powerful calculations.
This time, the challenge is the complexity of Go and the change of Go. Some people describe it as the sum of the numbers of the atoms. It is probably more than 120 digits. The changes are extremely complicated and complicated until we humans can never be clear. The future is not likely to be clear. Excessively complex so that the machine is not clear, so in Go we must use some human intelligence, such as analysis, judgment, decision, etc., rather than the computing power of the machine.
Many people began to say: "The machine can be so fast, of course, people can't do it." In fact, although the machine is relatively fast, it is not entirely based on calculations to defeat humans. It has its own neural network and has its own deep learning. It regards human beings as teachers and then continuously learns and improves, including the right and left struggles. In this process, the machine accumulates a large amount of human knowledge of Go, and at this time it can truly defeat humanity. So this time AlphaGo has a milestone significance in defeating mankind. It may have occupied a commanding height in all human intellectual games, so that after the game is finished, humans may have no chance to challenge the machine again, according to Ke Jie’s words. Said: "This is the last three games between me and the machine."
When IBM's IS6000 was "a deeper blue," the next day after defeating Kasparov, he announced his retirement. In other words, "If I can win you now, you will never have a chance to win me anymore, and it will be meaningless to follow you, so I will not play with you." Why did you have a chance to do this with Ke Jie this time? What about man-machine wars? It was because in April last year, AlphaGo and Li Shishi carried out the first "human-machine war". The game machine defeated humans 4:1, and Li Shishi made the so-called "hand of God" in the fourth set. The system of the whole machine was disrupted, and the logic behind it was disordered. I personally participated in the explanation. I found that the machine made some unconventional mistakes in this process. It should not be knowledge or ability.
From the content of chess, last year's game machine is indeed better than the top human masters like Li Shishi, but the good is still very limited. We humans may not be convinced: "While you won me, I didn't think you were awesome." At the beginning of the special competition, the human side thought that the machine was playing chess but not humans, including myself, and it was misjudged, but in fact the machine had already defeated human representative Li Shishi in the last game.
Li Shishi won the 14 world champions of human beings and should rule the history of human chess players for ten years. He represents the human beings, but objectively speaking, he has passed his peak period, not the best competitive state, so it is not necessarily It can represent the highest level of Go in humanity. Ke Jie is now the world's number one, so it is still challenging to arrange a human world first and AlphaGo to play another game. However, this challenge does not seem to be a machine challenge to humans, more like a human challenge. machine.
In addition to the field of Go, there have been "Cold Masters" and humans in the Texas Hold'em tournament, and later defeated humans. In fact, earlier, in fact, after IBM’s deep blue victory over Kasparov, he did not continue to study in Go, but went to participate in some human knowledge contests. The system was called Watson and later defeated. The champion of human knowledge quiz, so the machine challenges human wisdom in different ways on different occasions.
How does AlphaGo compete with humans?
There are many people who ask me how AlphaGo plays chess. What technologies does AlphaGo use?
cloud computing
As you think, it really has to be locked in the calculations. I don't know what the situation is like this time, but the last time I played with Li Shishi, AlphaGo used 2000 servers and performed calculations. It should be said that it uses the cloud. Computing technology concentrates a significant portion of computing resources connected through the network. In comparison, if you only need one server to do the operation for one second, 2000 can greatly reduce the time it judges each time, and even make the whole game, so using cloud computing Google should be able to use a lot of The server, not only 2000, but also 20,000 can be used, so now it is necessary to use two thousand servers to simultaneously participate in such a top game.
Big Data
It should be said that the previous version of AlphaGo is version 1.0. It has entered a large number of Go's chess games for human top players. It can be said that it inherits and learns the knowledge of Go that can be learned in the history of human Go, and then through deep learning, it becomes able to The knowledge of understanding is then applied to its own actual combat, so it is clear that this is also the result of a big data.
Of course, many people may not necessarily play Go, and many people do not know how to artificial intelligence. So AlphaGo actually has a very good set of artificial intelligence technology to adapt to the game of Go. Simply put, there are two networks - one is the network of value judgment, the other is the network of search, and then find the best that every decision makes. One point, this point is based on the highest point of victory as a basis for decision making.
We can easily understand how AlphaGo plays chess with people. When Ke Jie took a step, it would make a basic judgment in all possible points to determine which pieces should be considered, but what kind of chess should be considered? It may require a search method to verify, for example, if this point is often used by human chess players, or an important recommendation point that may be generated by its own learning system this time, according to this recommendation point will trigger a series of The evolution, and these evolutions are constantly verified by the number of searches of Mentocco. The two sides are equivalent to the way of mutual respect, and both sides follow the best response of the other side to evolve some of the following changes. When these changes have evolved to a certain number of steps, such as steps of 20 steps and 25 steps, it evaluates the results and then feeds back the results.
It is said that when AlphaGo version 1.0, its learning process is to use a large number of human chess as the main basis. It is the kind of point that human beings choose to think and judge in such a situation, and to search and deduct. This time, AlphaGo2.0 does not actually use the existing chess game. Instead, it uses two AlphaGos to learn from each other to generate knowledge recommendation points. I think this process is very easy to understand. Because the early AlphaGo did not have human knowledge at all, he needed a lot of chess. After waiting for it to pass the 1.0 game, it has become the top master of humanity, so the two AlphaGo's own mutual praise can be used as the basis for recommendation, so it is likely that this time it is produced by its own way. Go knowledge.
High-voltage Submersible Motor Winding Wire
This products is made of solid copper wire conductor, crossed-linked polyethylene insulation, optional shielding layer, polyamid jacket. It has fine performance for water-resistant, mechanical, electric conducting and acid-alkali-salt resistance. It used to all types water-filled submersible motor, oil-filled high-voltage submersible motor or similar working electronics, and applied to standard of environmental-friendly.
The product can be used for Voltage 1140V, 3000V, 6000V, 10KV temperature -40℃-90℃, water pressure 1.5MPa, light acid-alkali-salt media or oil-media, for long time use with stable liability.
High-Voltage Submersible Motor Winding Wire,Filled Water Submersible Motor Wire,6Kv Submersible Motor Winding Wire,10Kv Submersible Motor Winding Wire
HEBEI GUOQIAN WIRES CO., LTD , https://www.guoqianwires.com