Why did Ke Jie say "I lost my temper after losing"? 8 questions to interpret the first game of the man-machine match

After more than four hours of play, Ke Jie lost to AlphaGo by a quarter of a point.

After the game, Fan Hui, a chess player who had participated in the training of AlphaGo, and Chang Hao, a Go 9th Dan, reviewed the entire game process. Ke Jie also accepted media interviews with Deepmind founder Demis Hassabis and AlphaGo's main programmer David Silver to talk about their thoughts on the game.

AlphaGo focuses on the development of the overall situation

Fan Hui mentioned that after Master's 60-game match at the beginning of the year, many chess players also tried to use the point three three method. And Ke Jie also imitated this opening.

Chang Hao said in his analysis that what impressed him most in the entire game was the big flying move made by White in the 24th move shortly after the start of the game. This showed that AlphaGo paid more attention to the development of the central force and the thickness and development of the chess game, because this move would be more efficient in future games.

Another impressive move by AlphaGo was a break move at the upper left corner of the board shortly after the start of the game. Before the break, the black pieces (Ke Jie) looked thick, but then the white pieces (AlphaGo) also became thick. Chang Hao said, "This was a late move, but it left a lot of aftertaste."

The 24th move is "Big Flying", the 54th move is "Broken"

"This move brings us new ideas. Some chess moves are not impossible to play. After playing them, we will have new ideas. It provides an innovative way of thinking. We can study it together and bring more development to the way of playing Go."

Although the four characters surrounded by the Go pieces seem hopeless, AlphaGo will still make more use of these four pieces. It will want to utilize the value of all the pieces on the board.

Ke Jie: I lost my temper after losing

When reviewing the game, Ke Jie also praised the break in move 54. He said, "I was shocked. This is not possible in human play." After the break, he solidified his steps and became thicker, killing two birds with one stone.

Ke Jie sighed, "It was really a great game. I lost without getting angry."

Demis Hassabis said after the game that it was a very exciting game and he respected Ke Jie very much. He thought Ke Jie showed tenacity and wonderful strength and also did a lot of preparation before the game.

"Ke Jie prepared both strategy and tactics, such as the three-three move, which was his summary of the 60 games against Master at the beginning of the year. He has his own strategy, and several of his moves are very beautiful. The 24th move is a move that saves the whole game. There are many more games to come, and I respect Ke Jie very much. Even AlphaGo must go all out to deal with you."

Although the result of the game is out, the outside world's curiosity about AlphaGo and Ke Jie has not been satisfied. In the interview, both parties also answered many questions that everyone is concerned about.

Why did Ke Jie smile after the game?

Ke Jie said that it was a bitter smile, not a very happy smile.

"It knew it would lose very early on. Each of AlphaGo's moves was at a constant speed, and by the time it reached the final stage it already knew how much it would lose."

Ke Jie commented on himself, "AlphaGo played brilliantly, but I also tried my best. It played so well that there are many things worth learning and exploring for players. It has impacted the concept of Go and changed our initial way of playing Go. There is nothing that cannot be played. It has also changed our thinking about AlphaGo. We can be bold to innovate, expand our thinking, and play Go freely. Today, I also boldly expanded my thinking."

Ke Jie also wanted to use AlphaGo's strategy to deal with it. "My impression is that AlphaGo is very greedy for the ground, such as the three-three method, and I also used this strategy today. But this intention was defeated by AlphaGo, and it completely entered its tune. It played really well."

Ke Jie believes that AlphaGo's chess is completely different from last year. The first time it played very close to humans, but now it is more and more like "God".

Does AlphaGo have weaknesses?

Ke Jie believes that no weaknesses have been found yet. There were weaknesses in the past, but now AlphaGo's understanding and judgment of chess are far better than ours. If it wins, it can only find some bugs.

But he also said, "Always have confidence in yourself."

Hassabis said that he participated in the competition again after Lee Sedol in the hope that chess players could find AlphaGo's weaknesses and defects that it was unaware of when playing against them.

"Lee Sedol found this gap in the fourth game, and we later improved the system. Of course, there will be new areas that neither we nor AlphaGo know, so we will participate in the competition."

David Silver explained that AlphaGo's approach is to maximize its chances of winning, rather than winning the game by proximity. It will look for some certain routes to achieve the chance of winning with less risk. This is also a trade-off issue, to decide how to expand the proportion of victory. Of course, it has some other goals, such as taking a bigger step towards victory. If the focus is only on winning, it may reduce the risk and pay attention to even small risks.

Impact on Go?

Hassabis believes that Go is a very interesting subject with great potential. Through competition, we can improve our understanding of Go. This is a tool we can use to help great players discover more mysteries of Go.

Does AlphaGo rely entirely on the training of chess players?

David Silver explained that AlphaGo learned from humans in the beginning. When it learned by playing against itself, it also had to play against human experts, because it could only improve by playing against the best players. The new version learns more by playing against itself, realizing self-search and relying less on human data. DeepMind's goal is to make its rules go beyond Go and apply them to other fields.

After defeating Ke Jie, the top Go player, will AlphaGo retire?

Hassabis did not respond directly to this question, but said that the next step will be announced later this week. Just like the first version published technical details in academic journals, they will also publish more details of new versions in the future, and will disclose more technical details to allow other teams and laboratories to make their own versions.

Technical details? How many TPUs? Standalone version?

The AlphaGo used in this competition is a new version, which has improved the new algorithm. The main improvement is that the amount of calculation is reduced by ten times, and the self-playing ability is stronger.

Hassabis said that the program was run on a single machine during the competition, which was different from last year when it was distributed. This time there is a more powerful algorithm, which works simpler and better, and the speed is faster.

Hassabis said that computing power can be obtained "in the Google cloud, using TPU, ten processing units, ten times less computing power." In short, this AlphaGo is a stand-alone version.

Will Ke Jie never compete with artificial intelligence again?

Before the game, Ke Jie posted on Weibo that this match was his third match against artificial intelligence. After the game, he reconfirmed this decision.

"I've thought about it for a long time. AlphaGo is improving too fast. Every step is a huge improvement. It will be even better in the future. The gap between humans and it cannot be made up by our own efforts. I believe in the power of technology. The future belongs to artificial intelligence. As for Go, I still prefer to play with humans, because the gap between machines and humans will become larger and larger, and the probability of humans winning will become smaller and smaller. It's too painful to play Go like this. I can't win a single game."

He said that he would not train machines in the future, as this was a three-game chess game.

“I hope to continue playing chess with humans in the future, and view the software as a teacher and an object of learning.”

He said that he was a little dissatisfied with his performance and felt that he could do better, but sometimes he couldn't do it no matter how hard he tried. So this time, he hoped to play a game that satisfied him without any regrets.

"It would be good to make AlphaGo's host computer get hot."

So what if human chess players don’t want to play against AlphaGo anymore?

Hassabis said that from their perspective, they want AlphaGo to be a tool for Go that can be used by humans, so they will share the knowledge of the architecture so that everyone can know about it. They also hope that it can enter the homes of ordinary people and find application scenarios in medicine and science.