AlphaGo, developed by Google’s DeepMind AI division, has defeated the current world number one Go player in all three matches played during the « Future of Go Summit » in China this week.
AlphaGo is Google’s DeepMind AI-based program that learned by itself how to play Go, a game invented nearly 3,000 years ago in China that requires some level of instinct to succeed. AlphaGo defeated the current world number one Go player, Ke Jie, in their first match played last Tuesday at DeepMind’s « Future of Go Summit » in Wuzhen, China.
Two other matches followed since then between the two contenders, one on last Thursday and another this Saturday. AlphaGo has managed to win both of these matches as well, sweeping this week’s three-match series and marking « the highest possible pinnacle for AlphaGo as a competitive program », as DeepMind has stated in a blog post.
It is not the first time AlphaGo has defeated a high-ranked Go player, though. Last year, the program managed to defeat Lee Se-dol, one of the world’s top Go players, by defeating him in four out of five games, and winning the match. But after this week’s achievements, DeepMind has decided to retire AlphaGo from playing against humans. According to DeepMind’s blog post:
The AlphaGo team also plans to publish one final academic paper detailing the improvements made to the algorithms’ efficiency and its potential to be applied to other problems. Furthermore, the team will release a teaching tool based on AlphaGo’s analysis of Go positions and its underlying thinking, a top requests from the Go community. The development of the tool will count with the help of grandmaster Ke Jie, who will work with the team on a study of his match against AlphaGo.
Finally, DeepMind will publish a special set of 50 AlphaGo vs AlphaGo games « played at full length time controls » to inspire the Go community. The first ten games can be accessed here.
Source: DeepMind via Phys.org | Image via MIT Technology Review
