No. 285 Crushing Lee Sedol (7)

Wei Dongsheng smiled and waved to the audience to thank him: "The Rotten Go Challenge has ended here, and I will make a summary report." ”

"As you may already know, I wrote three papers on the most cutting-edge deep learning, and I compiled the main program of Ranke Go. What is Rotten Ke Go and what it can do, I know very well in my heart: half a month before public opinion despised Rotten Ke Go, I knew that Rotten Ke Go had a great probability of winning Lee Sedol. I made such a subjective judgment not because I knew Go or Lee Sedol, but because I understood deep learning and bad Go. ”

"The 7:0 record is also what I expected."

"There is a joke that there are two kinds of ball games in China that have the least sense of expectation, one is football, which is difficult for anyone to win; One is table tennis, which is difficult for anyone to win. To put it mildly, the gap between Rotten Go and Go players such as Lee Sedol is like Chinese table tennis and foreign table tennis, and I don't have the slightest sense of expectation for such a must-win game. I came here in person, not to celebrate the 7:0 victory of Ranco Go, but to apologize to Mr. Lee Sedol. ”

After that, Wei Dongsheng formally apologized to Lee Sedol.

The audience can be simply divided into two categories: one is the artificial intelligence faction who subjectively supports the game of Go, who cheers for the achievements of the game and specially comes here to participate in the historical process; The other group is the Go enthusiasts who subjectively support Go, who have witnessed the collapse of the Go myth in a very short period of time, on the one hand, they are shocked by the achievements of the game, and on the other hand, they are depressed, and if they lose something, they fantasize that a miracle can happen today.

After seven rounds of play, Go lovers are no longer reconciled, and they can only admit that Rotten Go is indeed powerful. However, when they heard Wei Dongsheng's remarks of beating the water dogs and disdainful of Lee Sedol and even the entire Go world, they still couldn't help but feel disgusted in their hearts, feeling that Wei Dongsheng was too technologically supreme.

Just when such emotions were slowly growing, Wei Dongsheng suddenly said that he wanted to apologize to Lee Sedol.

The audience was stunned one after another.

The artificial intelligence faction doesn't understand, and doesn't understand why it apologizes after winning Go.

Go lovers don't understand, and they don't understand why Wei Dongsheng apologized while despising Lee Sedol.

Wei Dongsheng stood in front of the stage and calmly resolved his doubts: "The reason why I apologize to Lee Sedol is because the members of the Rotten Go team are relatively young and act a little recklessly. ”

"In the first six games of the Ranke Go Challenge, the Ranko Go team tested the game ability of Ranko Go in the stages of layout, mid-game, and official. The team found that the winning rate of the game between Ranco Go and Lee Sedol was extremely high, often achieving a 90% win rate in the mid-game, so they wanted to test the limits of Ranco Go even further. The Rotten Go team visited Mr. Lee Sedol, hoping that Lee Sedol would accept the Rotten Go team in the seventh game, but Mr. Lee Sedol rejected the proposal of the Rotten Go team out of some concerns. At this time, the Rotten Go team made a mistake, they ignored Mr. Lee Sedol's refusal and privately adjusted the winning goal of Rotten Go, so that Rotten Go deliberately made a wrong move, allowing Mr. Lee Sedol to make 9 moves. The above is the truth of the fatal mistake of the 78th hand of Rotten Ke Go. ”

"Spring and Autumn Search is accustomed to emphasizing the spirit of contract, and with the authorization of users, we can analyze big data."

"A thief who doesn't tell him."

"This is not the value that Spring and Autumn Search adheres to."

"However, remembering that many of the members of the Ranke Go team are fresh graduates who have just joined the Spring and Autumn Search for less than a year, I hereby ask for Mr. Lee Sedol's forgiveness: please forgive their youth, please forgive their recklessness, I guarantee that the Ranko Go team will never be stronger in the future."

The audience was in an uproar.

Go enthusiasts who regarded the 78th move as the hope of mankind to defeat Rotten Go were even more shocked: the hope they regarded as a treasure was shocked that it was actually the handicap of Rotten Go.

This is embarrassing.

Wei Dongsheng seemed to be apologizing modestly, but these few words of apology made Lee Sedol's face hot and painful.

There is no hand of God at all!

Rotten Ke Go disliked that the opponent was too weak and uninteresting, so he took the initiative to let Lee Sedol 9 eyes.

Wei Dongsheng's apology is more like stepping on Lee Sedol on the ground and rubbing it.

Go lovers were shocked and then doubted: "Is Rotten Go really powerful to this extent?" How can a dead thing without emotion be so strong that it is invulnerable! Wei Dongsheng is telling lies, right? ”

Wei Dongsheng immediately gave an example to explain the basic principle of Rotten Ke Go: "What do you think of the quality of the seven games of the Rotten Ke Go Challenge? ”

"Before the challenge, some people questioned that Ranco Go lacked experience in playing against top players."

"That's wrong."

"From its birth to playing against Lee Sedol, it has played 200 million games of chess against itself."

"Two hundred million!"

"You heard it right, it's 200 million games. Moreover, the quality of each game after a million games is not weaker than the seven games of the Rotten Ke Go Challenge. Can a top human chess player like Lee Sedol play 10,000 high-quality chess games at the peak of his chess skills? The obvious answer is no, Lee Sedol doesn't have that much time and doesn't have that much energy. Therefore, the most important thing that Rotten Go lacks is experience, and to compete with it is to find its advantages to fight a tough battle. ”

"Everyone initially felt stupid because its chess path was inconsistent with the concept of Go in the world; But these chess moves, which conflict with the concept of Go in the world, can win against top players such as Lee Sedol. These previously unknown chess paths may be the unknown rules of Go discovered by Rotten Go by collecting 200 million high-quality chess games. ”

"And just like the 78th move is not caused by the loophole of Rotten Go itself, there are many things in the world, even if you see it with your own eyes, the result is not the truth."

"The framework of Rotten Ke Go is like this, the little tiger-shaped machine that plays face-to-face with Lee Sedol is only responsible for the power functions such as the robotic arm, which solves the problem of the embodiment of Go to the fall and is responsible for placing the chess pieces in the correct position; The real body of the bumpy Go is parked in the basement of this small building, which is very large, composed of 960 CPUs and 100 GPUs, and is responsible for calculating the real-time chess game picture seen by the little tiger-shaped machine camera, and searching the Internet to find the best place to land the next move; When Ranco Go needs additional calculations, the supercomputer Cambrian 2010-1, which is several kilometers away, can also become a foreign aid for Ranco Go. ”

"From everyone's point of view, it seems that Rotten Go has only played seven games with Lee Sedol."

In fact, from May 18th to June 5th, Rotten Ke played more than 700,000 games of chess with Lee Sedol. Well, to be precise, it should be Rotten Ke Go and its imaginary Lee Sedol who played more than 700,000 games of chess. ”

Wei Dongsheng asked people to start the projector and call up the chess game picture stored in the background of Rotten Ke Go: "Taking the 78th hand of the seventh game as an example, Rotten Ke Go played nearly 100 games of chess in advance. After Lee Sedol's 71st move, Ranco Go began to think about how to concede, how many chess, and where to concede. This kind of thinking is not a boring pure hard calculation, but an deduction based on existing data models. What the data model is, more on that later. Let's take a look at Figure 1 first, Rotten Ke Go has a plan to let Lee Sedol 10 mesheks here, but Lee Sedol's 73rd move is topped here, so that Rotten Ke Go can no longer play here. ”

、

"Let's look at Figure 2 again, Rotten Ke Go is ready to put the 'wrong chess' here, and it is calculated that Lee Sedol's winning rate will increase to 51.8%; Figure 3, Ranco Go is ready to put the 'wrong move' here, and it is calculated that Lee Sedol's winning rate will increase to 49.4%...... And so on, Rotten Ke Go was compared and screened more than ten times before deciding to play the 78th 'wrong move'. Moreover, after the 78th move, Ranco Go still holds the initiative in the chess game, and the winning rate is firmly controlled at more than 50%. ”

"Speaking of which, everyone should understand why Rotten Go made the wrong 78th move."

"However, this is still not the whole picture of Rotten Ke Go."

"Spring and Autumn Search is a search engine company, an artificial intelligence company, but not a Go club. The development of Rotten Go is not only to shatter the myth of the invincibility of Go, but also to verify the cutting-edge technology within Spring and Autumn Search in practice. ”

"For example, we have added big data analysis capabilities to Ranke Go."

"Before the Ranke Go Challenge, we entered all of Lee Sedol's published chess games into a sample database, and asked the Ranko Go big data to analyze Lee Sedol's past chess games and derive the Lee Sedol model. Then refine the sample model of Lee Sedol to obtain a variety of data models such as Lee Sedol in 2009, Lee Sedol in 2008, Lee Sedol in 2007, Lee Sedol in the Cup, and Lee Sedol in the Weijia League. Deep big data mining the difference between Lee Sedol in 2001 and Lee Sedol in 2009, the Rotten Go team and Rotten Go found many small details of changes, and I am afraid that there are many details that even Lee Sedol himself did not notice. In the fifth and sixth games, Rotten Go integrated and used Lee Sedol's habit of detailed layout that had formed a stereotyped thinking to quietly set up an ambush, so that he could continuously slay the dragon and create miracles. ”

"These data models are not simply analyzing the past, but 'learning from history' to predict future developments."

"Of course, data analytics is just data analysis, and it can't really predict the future. However, even the 'pseudo-future' of data analytics is of great value. For example, if Lee Sedol realizes and corrects the defects of Lee Sedol's stereotyped thinking habits that Rotten Go used in the fifth and sixth games, he will not be so miserable that he will be slaughtered again and again when he plays against Rotten Ke Go. ”

"Back to the win rate I just said."

"There are actually several hidden data in the winning rate prediction system of Rotten Ke Go."

After the 78th hand, the backstage showed that Lee Sedol's win rate instantly increased to 42.2%. It is said that Lee Sedol's winning rate is actually not Lee Sedol's winning rate, and Lee Sedol's real winning rate has never exceeded 20%. ”