New version of AlphaGo can master Weiqi without human help

 

‍After defeating Ke Jie, considered the world’s best Weiqi, or Go player in May, the artificial intelligence (AI) program AlphaGo announced its retirement from the Chinese ancient board game by its creator, Google's DeepMind. 
However, on Wednesday the company staged a comeback, published in the journal Nature, with a better version of the AI phenom. 
Named AlphaGo Zero, the new AI program defeated its predecessor, which previously beat top Weiqi player Lee Sedol, by 100 games to 0 after just three days of self-learning from scratch with no human help beyond being told the rules.

Lee Sedol played against AlphaGo in March 2016. /Photo via Reuters 

According to DeepMind, the previous AlphaGo initially trained on thousands of human amateur and professional games to learn how to play the board game. AlphaGo Zero on the other hand, skipped this step and learned to play simply by playing games against itself, starting from completely random play. 
The program not only understands the common patterns that have been accumulated by humans over thousands of years of playing, the new version can also invent better strategies beyond any human’s knowledge of Weiqi.
“For us, the idea of AlphaGo is not to go out and defeat humans, but actually to discover what it means to do science,” said Professor David Silver, lead researcher for AlphaGo.
The company is excited about extending the algorithm into the real world to tackle some challenging problems for humanity, Silver added.

Ke Jie's comment on AlphaGo Zero via Weibo. /Screenshot via Weibo

Chinese Weiqi master Ke Jie commented on the remarkable accomplishments of the new program via his Weibo account, "A pure self-learning AlphaGo is the strongest, humans seem redundant in front of its self-improvement.”

AI Weiqi programs in China

Chinese AI scientists have also been developing programs that could master the ancient board game. 
Developed by Chinese tech giant Tencent, an AI Weiqi program named Jueyi, or FineArt, is considered one of the major rivals of AlphaGo after taking the crown in the 10th Computer Go UEC Cup in Japan this March.

Jueyi at the 10th Computer Go UEC Cup in Japan, March 2017. /Photo via ThePaper.cn

According to Tencent, Jueyi was trained based on big data, similar to AlphaGo.
In addition to Jueyi, other Chinese AI teams have also demonstrated their work in the first World AI Go Open, a two-day AI and Go conference held in the city of Ordos in north China's Inner Mongolia Autonomous in August.
Post a Comment (0)
Previous Post Next Post