Even after beating the world’s best Go player last year, DeepMind is still working on improving AlphaGo. Over the past few days, a “new prototype version” of the AI has been playing in secret and winning against several dozen Go players online, according to DeepMind CEO Demis Hassabis.
With the user name Magister(P) and Master(P), this newer AlphaGo iteration only accepted matches from professional Go players in Asia on the Tygem and FoxGo server. Beating the top players in China, Japan, and South Korea, the mysterious player eventually defeated the current number one ranked player in the world, Ke Jie.
By the time Hassabis revealed the identity, AlphaGo’s record was over 50 wins and had only one tie due to a timed out connection on the part of the human competitor.
Sending shock waves through the community, there was even a 100,000RMB ($14,391.59) bounty for a human to beat the player. Offered by Go grandmaster Gu Li, he later opined after playing AlphaGo how “Together, humans and AI will soon uncover the deeper mysteries of Go.”
Excited by the results, the DeepMind CEO announced that with “unofficial testing” complete AlphaGo will go on to play official, full-length games later this year, with more announcements coming up shortly.
(Thanks to our tipster)