AlphaGo wins second match in series of three with Chinese Go master
World No 1 player Ke Jie said to be ‘out of his league’ in competition considered a benchmark in measuring the encroachment by artificial intelligence on human endeavours
AlphaGo, the artificial intelligence programme developed by Google DeepMind, won the second of three planned matches with world No 1 player Ke Jie on Thursday, in what had been billed as the ultimate battle of wits between man and machine.
The match, which took much less time compared with the first game on Tuesday, suggests that AlphaGo has been improving its Go-playing skills ever since it stunned the world in 2016 by beating South Korean grand master Lee Sedol, the then reigning Go world champion, in a 4-1 match.
The 19-year-old Ke played well at the beginning and even successfully predicated several moves made by AlphaGo. But as the match continued, AlphaGo turned the situation around by handling Ke’s multi-angled assault and eventfully forcing him to resign.
“In the middle of the game, I thought I was very near to a win. But because I was too nervous, I later made some bad moves,” Ke said in the post match press conference.
AlphaGo dispatched Ke in 3.5 hours, an unexpectedly short period of time. The first game in competition lasted about 4.5 hours.
A planned third match between Ke and AlphaGo is slated for Saturday.
Developed by DeepMind, a London-based AI firm acquired by Google in 2014, AlphaGo is the first AI-programme to outclass humans at Go, which is dubbed as the world’s most complex board game because it requires players to make their move based on intuition rather than calculation.
“If you ask a chess player, why you make this move？he or she will probably tell you a plan ‘because A, B and C’. But if you ask a Go player the same question, the answer is usually ‘because it feels right,’” said Demis Hassabis, co-founder and CEO of DeepMind.
Rather than pre-programming its system, AlphaGo uses two different neural networks, or brains, one for selecting the next move to play and the other for valuing the move and predicting the winner of the game. This helps to ensure it can adapt to the gaming situation in real time. Apart from training with millions of games played by experts, it can also sharpen its skills by playing against itself.
Because of its computing power and self-learning capability, AlphaGo can think about 50 moves ahead. It also uses a game strategy radically different from human players.
After his first loss to AlphaGo, Ke described the AI as “a God-like Go player” saying that many of its moves have never been seen in contests between humans.
Thursday’s game is the second of three between AlphaGo and Ke in a week-long “Future of Go Summit” held by Google in Wuzhen, Zhejiang province.
However, Nie Weiping, a Go master in China, said it was unlikely that Ke would win on Saturday, noting that AlphaGo is “clearly out of his league”.
David Silver, lead researcher for AlphaGo at Deepmind said that the current version of AlphaGo uses 10 times less computing power than last year when it defeated South Korea’s Lee Sedol. He said that the current programme learns much more quickly than its earlier version.
