Who's next? AlphaGo secretly returns, blitzing past online opponents
By Gong Zhe

2017-01-05 10:30 GMT+8

Scientists have never ceased in their efforts to make artificial intelligence (AI) better than its natural counterpart. In March 2016, a milestone was set in the great man vs. machine debate, when AlphaGo, an AI developed by Google-owned DeepMind, beat one of the best players at the game of Go, an ancient strategic and complicated board game which had long been considered as impossible for AI to conquer.
But AlphaGo's win against South Korean professional player Lee Se-dol was not the ultimate goal of DeepMind. They nicknamed an updated version of the AI as "Master", and secretly put it in an online arena last Thursday to test its strength.
Lee Se-dol (C, seated), reviews the game after losing to AlphaGo on March 15, 2016. /CFP Photo
The arena is well known to Go lovers for hosting professional players. Many famous players from China, South Korea and Japan have accounts on the platform, and play regularly, including the current world number one, China's Ke Jie.
But none of them managed to beat this mysterious new-comer. After 61 matches played during testing from December 29 last year to January 5 this year, "Master" teared down all of its opponents, except for a game that ended in a draw caused by a network dropout. After the test drive, DeepMind announced on Wednesday that the account is powered by their machine.
Announcement made by DeepMind CEO Demis Hassabis on Thursday. /Twitter
So far, the account has only taken part in "fast play" games, a version with very strict time limits. Statistically, in this short time period AlphaGo has already become the best Go player in the "fast play" category.
After losing online to this AlphaGo version 2.0, Ke Jie said he analyzed the matches, and realized "the computer has told humans that our thousand-year-old strategies of Go-playing are totally wrong." He also vowed to do all he could to face the "new storm" of AI players.
Ke Jie's Weibo announcement. /Weibo