Interview with Hui Fan: AlphaGo may already have found a different beauty of GO that beyond human’s imagination （1）
The game between AlphaGo ( the AI Go player program from Google Deepmin) and Lee Sedol ( the South Korean professional Go player of 9 dan ) might be written in the history. This is the first time an AI player challenges the top human player in Go game, the game that has been recognized as the most complex game in human history. The show of AlpahGo was started last year by the game against Hui Fan, the European Professional Go Champion.
Synced is proud to present this special interview with Hui Fan. In the interview, Hui Fan took us back to the spot of the game he played with AlphaGo, as well as shared his perspective on Artificial Intelligence and reflections on the meaning of life with the game of Go .
Hui Fan will judge the final board for the “AlphaGo VS Lee Sedol” game. We decided to warm up this upcoming “Fight of the Century” by sharing these profound reflections from Hui Fan, and we believe this exciting extensive piece will inspire us to go beyond just this single game. To give some thoughts with these grandiose propositions: machine v.s. human, technology v.s. culture, game v.s. philosophy, past v.s. future.
Part 1 — The game with AlphaGo
It was never a fair game. AlphaGo does not have feeling. He is like a ghost. He knows everything about you, but you can’t catch anything about him. — Fan Hui
Synced: What leads you to the game against AlphaGo? Any interesting story to share with us?
Hui Fan: Last August, I came back to Bordeaux for vacation after the European Go Championship or Congress (EGC) in a a small town of Czech. Soon around late August or early September, I received an email invitation from Deepmind to visit the company. I had very little idea about this company at that time. However, I still accepted the invitation as I am the kind of people who are open to experience.
I got to know that they were actually an Artificial Intelligence Research company owned by Google after a Skype call with them. During the call, they only mentioned that they were working on an exciting research project that is relevant to me, but nothing concert about the project or about the purpose of the trip, and of course, nothing about AlphaGo either. I signed an NDA with them. Then, it took place.
As for the NDA, I can not disclose some of the details. I guess the reason why they chose me is because of my performance on the European Go championship. I have been wining the European championship for the recent 3 years. Firstly, they need some one that is a professional Go player; and then this player better owning some nice title; preferably he’s not on the other side of the planet.
The first game against AlphaGo was in October, 2015. The result is known on the spot, and I knew how powerful it is. The world know nothing about it and I can’t tell anyone about it either. While I had to keep the silence, there were two thing became very interesting to me. One is that I went back to China to watch a Go program competition and find the system is totally not on the same level of AlphaGo.But I can’t say anything; The other happens in November 2015. There was an amateur Go player of 7-dan rank from South Korea, who was also a top player in Europe. He said that he played with Crazystone a while a go, wining easily with a 3 stone handicap, and believe AI is far from wining human on Go even though it has made a lot of progress recently. I had to keep it to myself at that time. It is not just for him, most people believe Go programs have to take another 10 years to beat top human players. Well, they had made it already.
Synced: You thought it was just a tour at first, but then you knew they want you to play game with AlphaGo. Did you have any concert at that time？ Have you ever thought that you might lose? Have you ever thought of rejecting?
Hui Fan: No one can win every single game. Sure I have thought of the possibility that I might lose it. I decide to accept the invitation because I didn’t thought there is much chance that I will lose; But even if I lose, so what? The game result does not mean anything to me. I am never afraid of losing. As long as you are playing the Game of Go, you must go with losing and winning. It doesn’t matter who your opponent is. I just try my best. It is good if I can win, but it is just another game if I lost.
When I was young, my father told me many similar stories, such as those about Lee Chang-ho. Lee worried about losing as well, but he keep a simple principle to make himself calm — just play the best for every single hand. In the end, wining or losing that game doesn’t really matter for me. Actually for all these years I always play with this attitude in Europe. It is interesting that it was actually DeepMind who were very worried. They were afraid of me rejecting. It was just a research project for them, but for a professional player like me, I might lose my reputation. But I told them not to worry, as to a professional player, wining or losing is nothing special.
Synced: But this is not just another Go game. How you feel when you play those 5 games? Were you under a lot of pressure or very relaxed in this game against machine?
Hui Fan: No one would like to lose a game that has wining or losing, especially as a professional player. I tried my best to win. There are some posts on the internet suspecting that I lost the game intentionally. I don’t want to comment on that, but I must admit that I did underestimated AlphaGo in the first game. As the opponent was an AI program, and we would play 5 games in total, I decided to play the first game simple and try something else in the later games. If I win the first game easily, then I know how to suppress him. If it didn’t go well, I can still switch back. Another reason is that I believe integrity is the key in the game of Go. It is hard to evaluate the state of the plate clearly. If I play with simple strategy, it might be actually harder for it to evaluate. That’s what I thought at that time. Its moves surprised me, — at least it never played the weird “computer hand”. It did pretty well on evaluating the plate. Before I made the mistake, I thought I would win. But after I made the mistake, I knew I was going to lose. Its control over the end game was amazing.
The first mistake is that I missed a Tesuji and caused a big loss. Another bad move was not hard to notice, but I was really careless because I thought I was leading and going to win. When that bad move appears, I know i might lose and there is no much chance to win. That was the first game, I didn’t really care too much about it but DeepMind shows big interest.
I was very upset after the first game. Why? It is simple, that was the historical moment that computer program beats a professional human player. It was the firs time and I lost it completely. Losing a game is not a problem, but I felt that I am incapable of beating him. I just made one mistake and I lost.
There are some more issue with my mentality in the later games. I changed my strategy starting from the second game. I designed some complex changes and trying to fight against it. The first Joseki was Avalanche, one of the most complex transformation in the game of Go. I caught some good chance at that time but then I made another mistake in the middle of the game. It caught me again, accurately as always. Once it caught me, the game fell into his hand and I could never turned it back. The later games went the same way, but it never made obvious mistakes.
Synced: Based on your evaluation on the plate, will it be a hard game if you didn’t make the mistake when you were leading at that time?
Hui Fan: It will be very tough. But the biggest problem is that human make mistakes while machine make less. Therefore it became a burden later on. It doesn’t really matter if i’m leading or losing, I’m afraid that I’m gonna make mistake. But it is not a problem for AlphaGo. I actually lost my mind because of that. When I am leading, I didn’t believe I could keep it up; When I was losing, I kept thinking that I was going to lose. Then it actually affect my performance. Why people don’t think I was playing well at that time? Because to an experienced Go player, the performance in the game usually doesn’t matches your ability due to psychological factors. For example. Lee Chang-ho, the famous player who has beated many top Go players in the world. He was known as “the Stone Buddha” because of his poker face and stable mentality. You can’t read anything from him. But what I want to point out is that who can be a “Stone Buddha” comparing to computer? When facing the computer, no one can be more poker face than it. It does not has feeling like human players, and I suffered in the later games. It is void. You can’t get him but he knows all about you. Totally imbalanced.
Synced: Did AlphaGO missed any mistake that you made in the real game？
Hui Fan: Yes, it neglected some mistakes that I made but I don’t think id didn’t discover. It rather let me go as even it didn’t catch me, it will win. It decided to let me took an area that I was supposed to lose. There would be more risk if it took that piece from me and the game will be more complex. It decided to not took my stone so that it could win the game easily. It chose a less risky strategy to win easily. Some Go TV show replicated the game between us in a later time, a professional Go player had the same view. As in the real game, I resigned soon after he let me took that area. I might continue fighting if it took my stone and that area. The game might be more complicated. Just like how Jie Ke said, he could not tell who is human and who is computer. No one can tell.
Synced: Was it a problem if it made mistake? During the 5 games, which mistakes that you see as vital? How frequent did it make mistakes?
Hui Fan: I can’t answer this question as it is covered in the NDA that I signed. You can ask other professional Go players about their opinions.
To Be Continued …
Author : Wei Zhao, Yunfeng Zhao, Rita Chen | Editor : Chain Zhang
