Lee Sedol talks about his battle with AlphaGo

“I don’t want to compete with AlphaGo any more”, said Lee Sedol on the Summer Davos this year.

Shortly after he appeared at the conference hall, Lee Sedol was completely surrounded by cameras. Since there were too many people while the space was limited, almost half of the audiences were left outside.

3 months ago, Lee Sedol lost to AlphaGo- Google’s artificial intelligence algorithm by 1:4 during a man vs. machine Go competition.

And 3 months later, he said on a round-table conference in Davos that he doesn’t want to compete with AlphaGo anymore.

Screen-Shot-2016-03-09-at-1.10.59-PM

We have compiled the content Lee Sedol talked about on the conference, including his feelings about AlphaGo and understandings about artificial intelligence.

As the most direct participant of the man vs. machine competition, Lee Sedol’s altitude might be similar to most of us, on one hand, he fears the completely cold-blooded thinking machine, but on the other hand, he hopes the technology behind it would bring some benefit to our life.

In the following, H refers to the host Li Mei, a director of Shanghai Tech University, while L refers to Lee Sedol. To ensure the continuity, the order of questions has been adjusted.

H: How did you feel while you were competing with AlphaGo?

L: I was very surprised as there are so much uncertainty in a Go game. I thought computers nowadays are unable to reach the level and we all know the result, I lost the game by 1:4.

I always thought that the game of Go requires people’s intuition and prediction, at that time, I felt that AlphaGo wasn’t playing. But it had vulnerabilities while it met with unexpected moves. AlphaGo surprises me, but still it has limitations.

H: AlphaGo had applied massive GPU and CPU to calculate, it also had learnt from more than 1 million games. The whole process cost it 1000 watts, while it cost a person 100 watts. This means you are competing with 10 persons and behind them there are 100 scientists, do you think it’s fair?

L: I didn’t know much about “deep learning” before competing with AlphaGo, so I wasn’t taking it so seriously. Back then I had watched its competition with Fan Hui, but I thought it couldn’t defeat me and I should be able to win. I never thought deep learning would be able to make that big progress within 6 months.

I think we cannot say it’s unfair, only that computers have their own features. It’s unfortunate that I knew too little about deep learning and that was an error judgement. I thought it was unable to reach the level of a top Go player, that’s the main reason I failed.

H: During the first round, one move of AlphaGo had surprised you a lot and you had displayed your feelings. However, AlphaGo never show any emotion or expression, does that influence you?

L: After all, a human being would have psychological sway, even though he knows the answer, he might choose another way and he will consider the whole situation.

So a human being is at a disadvantage while competing with a machine. For example, if my rival is a person, he definitely would make changes in each round. But AlphaGo always stays the same, it even couldn’t realize which round it is in and how the whole situation has become.

Thus I want no more competition with AlphaGo. This is a surprising experience for me, if there’s another chance, it would still be very tough.

H: What difference were you faced with during the first 2 rounds? And what’s the difference with AlphaGo?

L: The difference is clear, the player using black must attack actively because he is in a disadvantage with the win rate of 48%. I was being too active so that I went wrong.

AlphaGo had made perfect preparation, it has too much information. If I could compete with it again, I would have a better chance to win, but I really don’t want to play with it again.

Although I won in the 4th round, that was because AlphaGo had a fatal bug. It still had bug for the 5th round but I lost. Generally speaking, I think it’s difficult for a person to defeat a machine.

H: Why don’t you want to play with it anymore?

L: Because I completely had no emotional exchange with it and although I had imagined it beforehand, my pressure during the actual game was unimaginable. To compete with an impersonal machine is really a tough thing to do.

H: If you are going to compete with AlphaGo again, what preparation would you make?

L: Before the game, I had had a lot of practice in my mind and imagined what might happen during the game. Just like I said, I made an error judgement.

Although I hate to face it, on the other hand, it’s a rival that could arouse my desire to challenge.

AlphaGo-Lee-Sedol-first-move

H: Some people say that AlphaGo is actually not that good on Go, and you should have tried harder in the game, what do you say?

L: I agree with this opinion to some degree, if AlphaGo now remains the level it used to be while competing with me, I might be able to defeat it. I mean it has vulnerabilities, but the question is that it can make huge progress within 6 months, and now it’s 3 months later, I think there’s no way people could defeat it now.

H: Sometimes, AlphaGo played neat tricks, is it really creative?

L: AlphaGo often made stunning moves. I didn’t think it has this kind of creativity, but it made it. So what we should consider is that what the so-called creativity is? The so-called creativity is a result of our learning and it only appears in our cognitive domain.

H: Would you agree if your daughter wants to learn Go from AlphaGo?

L: Of course I would agree. I admit many excellent educators would make an opposite choice. But as a father, I’m so scared of this (human’s) uncertainty. However, AlphaGo is rather accurate, it would fulfil a task within the prescribed scope, my attitude on it is absolutely positive.

Leave a Reply

Social media & sharing icons powered by UltimatelySocial