Google DeepMind’s AlphaGo beats Ke Jie
Google DeepMind’s AlphaGo beats Ke Jie again by resignation, officially winning all three games in the latest series of matches putting human against machine. For cultural sensitivity reasons, due to Go’s 2000+ year history in the region, China decided to censor the livestream.
At time of resignation, Ke Jie had roughly 32 minutes of time remaining, while AlphaGo had 1 hour and 30 minutes remaining on its clock. A press conference discussing the match series and AlphaGo Project, along with the collaboration which made the match series possible, was held shortly after the event.
Ke Jie, the world’s #1 human Go player, a title the player has held each year consecutively since 2014.
The complex nature of Go has produced remarkable variation in play styles over time, with some players being more aggressive and unpredictable, while others have effectively played traditional, conservative strategies. Variation in skill is so great that even though the game has over a dozen unofficial rankings, nine dan players are considered vastly superior to almost all lower-ranking players.
In particular, Ke Jie has been a wildcard, known to be arrogant and confident. After renowned Korean player Lee Sedol lost to AlphaGo in 2016, Ke Jie had claimed confidently – possibly correctly at the time – that he would have been able to beat the AI.
However, AlphaGo’s skills have improved tremendously even since then. AlphaGo anonymously signed up in the TyGem and later FoxGo community servers, a Go community popular among even the world’s greatest professionals, using the name “Magist” (and later “Master”). There, it acquired a streak over 60 wins, beating nearly all top world players (including Ke Jie at some point) in rapid-fire online matches.
Its feats were so great that Go player Gu Lie offered a $14,400 bounty for the first player that could beat “Master.” Its status as an AI was suspected early on, due to the fact that it would play up to 10 games a day with few breaks between.
This remarkable pace of growth has been welcomed, but it comes as a cultural shock to the human game and AI communities, which had not expected that machines could compete with humans in this realm for at least another decade.
At the press conference, we may hear additional information about what DeepMind’s plans for AlphaGo are. AlphaGo represents a rapid advancement in artificial intelligence, and DeepMind is a subsidiary of Google, which uses artificial intelligence in many of its products, including search and Waymo – which was formerly Google’s self-driving car project.
This latest series of games is part of Google’s “Future of Go Summit,” and AlphaGo has also faced against a team of collaborating Chinese professional players. A third challenge involved AlphaGo vs AlphaGo, with a human collaborator each having AlphaGo available to assist them. AlphaGo won all games. This collaboration demonstrates not only the power of AI as a silo, but as an extension of human capabilities. As AI learns from humans, humans will learn from AI.