In the 4th game of the man-machine war, Lee Sedol endured until the second half of the game fell into a bitter battle, and he was forced to play chess in the middle abdomen, and after more than 20 minutes of the long examination, he played 78 hands to dig. Gu Li exclaimed that he wanted to be selected as the annual magic trick; Ke Jie laughed and said that it was made with his life and must be carefully arranged. Then Alphago played a series of problem hands, and the situation was quickly out of control, and Lee Sedol's total score was 1-3. So, is the 78th hand really one of God's hands?
Deepmind leader Hassabis tweeted that alphago miscalculated the 78th hand, believing that the winning rate of the 79th hand was as high as 70%, and it was not until the 87th hand that it reacted that it was his own mistake. Go is so deep that the two moves of 78 hand digging and 37 hand spikes in the second game will be discussed for a long time.
Dr. Huang Shijie, who replaced the alphago, pointed out: According to the preliminary understanding, alphago missed a series of combination punches in the middle of Li Shishi. Lee Sedol is really powerful! Alphago basically thought there was no chance after being played in the middle. Anyway, today's chess is very interesting, I think it's really not easy to win a game, but Lee Sedol did it! Alphago is still progressing, we haven't seen the top yet, and future versions will be stronger and harder to deal with. Happy for Lee Sedol.
Lee Sedol said after the game that the 78 hand was the only place he could see if he could drop.
But can this hand really play chess? Professional chess players such as Zhang Dongyue questioned.
After Lee Sedol 78 dug, Alphago quickly fell 79 and retreated, and then walked out of the black 83 to wait for a series of problem hands, running dead more and more losses.
In fact, black 79 can be played in the 1st place as shown in the picture, until black 7 is strongly blocked, and the white dragon has no way to live. Notice that there is an insignificant robbery here.
After the 79th block, the black still has a chance, the black 83 is pictured in the 1 tiger, and the black 7 is calmly restrained, and the situation is still black leading.
Black 83 can also play as shown in figure 1, and white can start robbing as desired, but black can jump out of 7 places, and when the white robbery black seals the lower right, the situation is also ahead.
Even if, as Hassabis said, alphago found the miscalculation in the 87 hand, but still had time to turn back, at this time the black series of moves were considered by the spectators to be accumulating looting materials, ready to play as shown in figure 1, and white 2 to fight between the two sides.
Until the black loss to the right of the second way, 97 hands such as Gu Li recommended to play in figure 1, abandon the two sons can still stop the loss in time, the victory or defeat depends on the official competition.
However, the actual combat black 97 dug up the unbelievable, and the white loss was a chess hand, and the situation was finally out of control.
According to the analysis of computer science expert Chen Jing and others, alphago relies on the algorithm research pipeline established by the Google team and huge computing resources, and achieves improvement through continuous improvement of strategies, value networks and other coefficients. There are still some module codes that are the results of the first author David Silva and Huang Shijie accumulated over the years, if there are defects, they may be covered up by alphago's neural networks, self-game, reinforcement deep learning, etc., which is difficult to find, which requires a top "software test engineer" like Lee Sedol.
Through the first three games of competition, Lee Sedol has learned about testing opponents from as many aspects as possible, and roughly found that alphago killing is not strong, minor officials make mistakes, especially in unnecessary situations, do not like to rob and so on. So in the 4th game, Lee Sedol changed his previous style of chess and played chess unusually stoicly, until he decided to win or lose through the complex killing of the open area of the mid-abdomen. After a long examination, Lee Sedol's 78th hand did not choose the span recommended by Ke Jie, but was not successful in the eyes of many chess players. It is precisely this originally untenable hand that triggered a series of alphago mistakes?
Chess friend Frogley boldly speculated that after the exchange of more than a dozen hands such as the normal fight of white 78 digging black, there was an irrelevant robbery in the local area, and white could not play chess. However, it is possible that Alphago will preferentially use the search tree to cut out the longer robbery branch of the decision chain for brevity, and as a result, the correct response of White 78 Mining was cut, resulting in Alphago's mistake. In addition, because deep learning is statistical knowledge rather than precise logical knowledge, there may be blind spots in complex hedges and complex robberies.
When the mistake is realized, alphago's unexpected moves after that may be to create robbery materials (such as 97 hands), or to play the general selection point after the defeat has been determined (such as 101 hands), or to break the can after the situation is very poor (such as 113 hands), but it will not admit defeat until the winning rate drops to the bottom line (180 hands).
Artificial intelligence will be used as a tool to benefit mankind in many fields, and I believe alphago is also aiming for this. Li Shishi could not get more information such as opponents' chess scores, and after what kind of thinking he chose 78 digging, which induced alphago mistakes in the dark, providing a valuable case for human beings, worthy of god's hand.
(Li Xinzhou)