Google’s Go-playing AI AlphaGo has defeated world champion Ke Jie in the first game of a best-of-three match series in China.
DeepMind’s AI software AlphaGo beat Ke by a close margin in terms of points, but was dominant in terms of actual gameplay in a four-hour match held during a five-day AI symposium dubbed the Future Go Summit in Wuzhen, China, which is being jointly held by Alphabet, China Go Association, and the local government.
After the match, DeepMind said the AI is now being powered by Google’s second-generation cloud Tenser Processing Unit (TPU) unveiled at its I/O conference held earlier this month.
The system previous used multiple previous-generation TPUs when it beat Korean champion Lee Se-dol by four games to one in a game last year after beating the European Go champion by five games to nil to become the first computer to defeat a human champion in Go.
During the summit, AlphaGo will be involved in a variety of matches, with AlphaGo holding its first team matches on Friday. The teams will consist of AlphaGo and a human player, and will be played over five-match series.
AlphaGo and Ke’s remaining games are planned for Thursday and Saturday, local time,
The summit, which will be attended by AI experts from Google and China, will “explore how the technologies behind AlphaGo, machine learning and artificial intelligence, are bringing solutions to some of the world’s greatest challenges” such as energy and medical usage, Demis Hassabis, CEO and co-founder of DeepMind, said in a blog post.
AlphaGo had already defeated Ke once; in January, it was revealed that while playing under the user name “Master”, the AI had played online matches against several Go champions through online gaming platforms Tygem and FoxGo and clinched 60 straight wins.
Hassabis said the online games were played to test the new prototype version of AlphaGo.
“Now that our unofficial testing is complete, we’re looking forward to playing some official, full-length games there this year,” Hassabis said at the start of the year.
AlphaGo played its online “test” matches between December 29 and January 4, defeating Ke during that time, along with world number two Park Jung-hwan from South Korea, Japanese champion Iyama Yuta, and a litany of other world-renowned champions.
Ke at the time said human players need to learn computer patterns to improve, while world number 35 Gu Li offered 100,000 yuan to whoever can beat AlphaGo to protect the “cultural treasure” invented by their ancestors.
Go, an ancient board game originating from China, has 10^761 possible games in comparison to the 10^120 possible games for chess.