Go masters stunned as AI passes supreme test of intuition

The elation was in stark contrast to Lee’s remarks at the start of the best-of-five series, when he dismissed his non-human opponent as being unready for such a complicated game. The expert predicted then that he would defeat the program 5-0 or 4-1, saying that it had not yet mastered the “human intuition” needed for Go.

Lee quickly suffered three straight defeats, losing the US$1 million series, before finally scoring an “invaluable” consolation win. The grandmaster – and the Go community as a whole – was flabbergasted. After his first defeat a visibly shocked Lee told press, “I never thought I would lose.” Self-belief dented, he lowered his chances of winning to 50-50.

Lee’s initial confidence didn’t stem from arrogance. His belief in human superiority at the Go board was shared by many even after AlphaGo beat European champion Fan Hui in October 2015. The European champion is among the top amateur players while Lee occupies a higher spot among the global Go elite.

When supercomputer Deep Blue beat world chess champion Garry Kasparov in 1997, Go was seen as the one of the key remaining unconquered frontiers for artificial intelligence. The astronomical number of possible moves in Go – more than the number of atoms in the universe – makes it hugely difficult for a machine to crack by brute computing power. It takes too much time and calculating power for a program to run all the possible moves in order to find the best one, which is what Deep Blue did in 1997. To defeat a human Go master, the program would need to “learn” the game.

AlphaGo achieved exactly that by studying the moves made by millions of opponents in tens of thousands of online Go games, narrowing down the number of possible moves it needed to consider. The astonishingly effective result caught more than just the world champion by surprise: A commentator on the second game of the series apologised to the audience for being unable to comprehend AlphaGo’s moves. “I am sorry but I do not know what Lee Sedol’s losing move was,” said Song Tae-kon, a Go master himself. “From a human’s point of view, AlphaGo is making frequent mistakes.”

Fellow master player Lee Hee-sung commented that AlphaGo is completely unpredictable as its moves showed no pattern at all. “A lot of its moves appeared to be wrong,” said the expert, who confessed to being bamboozled over exactly how AlphaGo’s strategy won the game.

AlphaGo’s victory is a historic event the world is still trying to comprehend. The Korean Go association has joined an IT lawyer in demanding that Google apologise for gaining an “unfair advantage”, pointing out that the program was able to analyse all of Lee’s past moves from the master’s game records. The lawyer, meanwhile, pointed out that unlike Deep Blue, AlphaGo was connected to the Internet, through which it drew on the power of hundreds of computers to study Lee’s moves in real time. Even if these assertions were true, they only describe why AlphaGo won. The key issue of the series is not whether the program should be regarded as a normal Go player, but whether it can defeat a human expert. 

Even Google has failed – deliberately or not – to understand the true meaning of AlphaGo’s victory. Google Taiwan managing director Chien Lee-feng pointed out that while AlphaGo represents a key step in the company’s development of learning machines, human beings should not be too worried since machines still can’t replicate human intuition and associative ability.

The AlphaGo program didn’t just beat Lee, it did so in a way that the best Go players in the world cannot even comprehend. Its victory shows that a machine does not need to pass the Turing test for thinking like a human to pose a challenge to humankind. A machine might lack the multipurpose, intuition-oriented edge of a human being, but it can be far superior in a well-defined competition. For Lee, the fact that AlphaGo couldn’t crack a joke made it no less fearsome as a Go opponent.

The idea that artificial intelligence should only be a worry to humans when it reaches self-awareness is misguided. Human beings might not be enslaved by machines anytime soon but AI is already making its impact felt on our lives.