AI trumps humans again after online Go master revealed to be Google programme
Artificial intelligence has already beaten human masters of strategic board game 60 times over the past week
Artificial intelligence has scored another milestone victory over humans, after Google-controlled AI developer DeepMind revealed the identity behind a recent secret online winner in the world of Go.
Speculations were put to rest by the confirmation posted on Twitter by DeepMind founder Demis Hassabis on Thursday that a 60-time winner with the online alias of “Master” is an improved version of AlphaGo.
— Demis Hassabis (@demishassabis) January 4, 2017
AlphaGo made news headlines in January 2016 after beating South Korean’s 9-dan professional player Lee Sedol, one of the best Go masters in the world, in a five-game match. Lee later said in an interview that he was shocked at how perfectly AlphaGo performed.
Racking up victories over 60 contestants over the past week in a fast-paced version of the ancient game of strategy, Master’s victims included top names.
Ke Jie, China’s No 1 player, was sceptical when AlphaGo defeated Lee.
“AlphaGo can defeat Lee Sedol, it can’t defeat me,” The 19-year-old grandmaster once boasted on Weibo, China’s popular Twitter-like social platform.
He was proved wrong when he failed to beat the AI programme. In a Weibo post on December 31, Ke called the programme the arrival of a “new storm”.
“I have been studying Go software since March … Humans have been practising and playing Go for thousands of years, but the computer now tells us that we are all wrong,” Ke wrote. “From now on, we Go players will need to work with the computers to enter a new area and a new level.”
He revealed on Wednesday that he had been aware of Master’s identity.
“I thank AlphaGo’s newest version that brought shock to the Go world. … I hope very much that humans can win one game,” Ke wrote on Weibo.
In addition to household names in the Go community, including China’s Chang Hao and Gu Li, South Korea’s Park Jeong-hwan and Japan’s Yuta Iyama, China’s Nie Weiping was also defeated by Master.
Nie, known in China as a “Go god” and a childhood friend of President Xi Jinping, was quoted by Chinese media as saying AlphaGo subverted the orthodoxy of the game.
“Go is far beyond the simplicity we imagined – a huge space is still out there for us humans to explore,” Nie said in an interview. “AlphaGo or Master, they are sent to lead our way by the god of Go.”
Bought by Google in 2014, the London-based AI development company announced that the new version of AlphaGo was ready for challengers of “official full-length games”, according to a tweet posted by DeepMind’s Hassabis.
AlphaGo was first trained using a database of about 30 million moves collected from expert human players. It then completed thousands of games against itself. The large quantity of intelligence accumulated from that data enables AlphaGo to evaluate different moves and their impact on the probability of winning at each stage of the game, and thus decide the next move to make. The latest tweaks in the technology were not revealed.
Go, an ancient abstract strategy board game, originated China and flourished in eastern Asian countries for more than 2,500 years. Its tactics require intense concentration and thoughtful calculation and prediction of movements. It has been described by Hassabis as “the most complex game ever devised by man”.