AlphaGo vanquishes world’s top Go player, marking AI’s superiority over human mind
Humanity’s most complex game was thought to be the final frontier that takes another decade for machines to conquer. That changed this week with AlphaGo’s 3-0 victory over Ke Jie
Just a year ago, the Chinese master of the complex game known as Go was insisting no computer could ever outplay him.
Yesterday, Ke Jie was in tears, calling the artificial intelligence programme developed by Google that beat him in three straight matches “perfect, flawless, without any emotions”.
“I’m very sorry I lost,” 19-year-old Ke told the post-match press conference in Wuzhen in Zhejiang province. “I wish I could have done better,” he said.
Go, also known as weiqi, is played on a 19x19 grid board by two opponents. With more permutations than the total number of estimated atoms in the visible universe, it has been a benchmark for measuring the human mind against artificial intelligence after IBM’s Deep Blue beat chess grand master Garry Kasparov in 1997.
Computer scientists and futurists had predicted AI would need at least a decade before it could decisively conquer the game.
The final match took less than four hours to decide.
The programme, developed by Google’s DeepMind Lab and called AlphaGo, uses two neural networks to analyse the game – one that selects its next move while the other evaluates the decision. This allows it to adapt to the situation in real time, planning 50 moves ahead and even play out hundreds of thousands of games against itself, all the while learning and improving.
It won Tuesday’s initial match in 289 moves played out over four and half hours. Ke declared AlphaGo “like God” after it pulled several surprise moves. Those surprises kept coming all week, with the programme making “unusual” and “interesting” moves against him. “I made several bad moves, moves that I regret, because I wanted to play well too much,” Ke said, conceding he would never be able to beat AlphaGo in his lifetime.
The odds were stacked against Ke from the start. He had already lost to AlphaGo in an online match in January, and last year, the programme dispatched South Korea’s Lee Sedol, the ranking Go champion at the time, in a 4-1 match. Following the latest matches, AlphaGo was awarded the highest honour by the Chinese Weiqi Association.
DeepMind’s co-founder and chief executive Demis Hassabis said AlphaGo would stop retire as the top players in China represented the highest possible level of human achievement in the game.
For the mainland, the Wuzhen showdown was ripe with irony and symbolism. China is the birthplace of wieqi, with records showing it was played as early as the Zhou dynasty (1046BC-256BC).
Wuzhen is also host of the annual World Internet Conference, where the government touts its version of an internet walled off from the rest of the world. With Google banned on the mainland, censors had been managing the expectations of the home audience since the competition began.
Live broadcasts, originally scheduled with CCTV5, were cancelled. The game could not be found on many of the most popular streaming services on the mainland, such as Tencent and Sina.
Still, DeepMind organised a five-day Go festival where players held matches against artificial intelligence or formed teams against a single machine player.
“AlphaGo’s successes hint at the possibility for general AI to be applied to a wide range of tasks and areas, to perhaps find solutions to problems that we as human experts may not have considered,” Hassabis said.