벤처 핸즈업!

The Go Master Strikes Back After a Losing Streak 본문

For English Audience/Tech Trends / Events in South Korea

The Go Master Strikes Back After a Losing Streak

알 수 없는 사용자 2016. 3. 14. 01:35

Hello, I'm back with the latest news on Lee Se-dol vs. Deep Mind's AlphaGo. 


Related Articles



After a striking losing streak, Lee Se-dol finally found himself a much needed victory over the ingenious software, AlphaGo. Let me review the first three games. In the first game, Lee Se-dol was very confident and also doubtful of the capability of AlphaGo. This attitude mostly came from experiencing Go softwares in the market that were not even close to the level of professional Go players. And the AlphaGo's matches with the European Go champion revealed that it was far smarter than the existing Go softwares, but was still immature in comparison to the top notch Go players.


That was why Se-dol tried some unordinary moves in the beginning of the game to see how AlphaGo responded. But in fact, AlphaGo not only responded, but also showed surprisingly strong moves that were deemed unorthodox to the norm of Go playing. The whole effort to find how prepared the software was went astray and brought the first loss for Se-dol. 


Se-dol seemed quite perturbed by the first game. His style was less aggressive in the second game. The game proceeded in favor of AlphaGo from the beginning to the end of the game. However, Se-dol admitted that the loss was due to his inability to take advantage of several chances that were actually proven quite favorable to him. Despite the continued loss, Se-dol was not without a gain, which was finding out AlphaGo had a weakness.    


For his third try, Se-dol came out with a strategy of utilizing Ko fights, which can significantly increase the number of possible outcomes. He was definitely under extreme pressure as he was just one game short of losing the historic man versus AI fight. I can't even fathom how much pressure he might have felt since there was so much to lose on his side. Nonetheless AlphaGo proved itself to the world again. It was capable of playing Ko fights to his advantage.




So, let's talk about the fourth game. Lee Se-dol has already lost three games, which made AlphaGo the clear winner. The positive side was that he probably felt less pressure going up to a match. Because of this, he could finally show the best of his style The turning point was made in the fight in the center where Se-dol tried to invade black's big territory. The build-up to the 87th move, which not many people expected, was called a God's move by Gu Li who is one of the best Go players in the world. After the unexpected move, the AlphaGo's value net, the winning percentage for the next move, plummeted from 70% to about 60%. Since this move, AlphaGo's moves didn't seem sharp at all. I believe this is one limitation of the algorithm based on the value net. When there are several moves with a value net surrounding 50%, it has a difficulty selecting the best move out of seemingly bad moves.    



Anyways, Lee Se-dol strongly finished the match and won a victory over the AI. I was very glad somehow. Maybe all humans were except for the Google people. The interview with Se-dol and Hassabis at the very end is fun to watch. Just look at the smile from Lee Se-dol.



I'll be back with Watson vs. Deep Mind next time. And also I'll be writing articles regarding IT and startups in Korea, so please keep following if you are interested. Thanks for reading :)


 Follow Venture Hands Up



벤처 핸즈업을 구독하세요.


Comments