It was a result that sent shockwaves through the Go community. AlphaGo, the computer created by DeepMind, the artificial intelligence (AI) arm of Google, thrashed European Go champion Fan Hui 5-0 — the first time a computer program has beaten a professional player of the ancient Chinese game.
Played on a board with a 19x19 grid of black lines, Go is such a complex game that enthusiasts hoped it would be years, or perhaps decades, before machines would be able to triumph over the best human players.
But now that time scale is shortening and AlphaGo is scheduled to play the world’s top player, Lee Sedol, over five games in March.
Mr Lee is a stronger player than Mr Fan and for now remains confident. "This is the first time that a computer has challenged a top human pro in an even game," he said. "I have heard that Google DeepMind’s AI is surprisingly strong and getting stronger, but I am confident that I can win, at least this time." It’s the "at least" that’s significant here. The parallels with chess are ominous. IBM’s Deep Blue lost 4-2 to Gary Kasparov the first time they played in Philadelphia in 1996, but triumphed 3.5-2.5 a year later in New York. Mr Lee may not succumb the first or even the second time, but in the end he or a successor will and another bastion will have fallen.
Complexities of Go
Go is such a complicated game that until recently the programs could defeat only amateurs and the Google team had to use a new approach. It now looks ahead by playing out the rest of the game in its imagination, many times over.
The program involves two neural networks, software that mimics the structure of the human brain. It was trained by observing millions of games of Go and evolved to predict expert moves 57% of the time. The network was then set to play against itself, learning from its victories and losses as it carried out more than a million individual games over the course of a day.
This is only possible, of course, due to the huge improvements in computing power in recent decades. And the bottom line is that the machines are, or will soon be, able to defeat the best humans at most games.
We in the chess community have had to deal with this for nearly two decades now, and the solution has been to accept that they will beat us in single combat but work around it.
We know that as human beings we will make small mistakes, however well we play. Chess playing computers ("engines") are uniquely well placed to exploit these and once they have a material advantage are almost totally unplayable. But we can console ourselves that they still don’t create that much themselves and rather than banging our heads against a brick wall, we can use them as superb training agents.
We use "engines" extensively in preparing for games — training in which the crucial element is that the human must lead the machine rather than following.
And after these practice games we always check with the "engines" to find all the mistakes we’ve made and hidden tactics we’ve missed.
Sprinters don’t run against race cars and there’s no reason that human Go or chess players should compete directly against computers.
The Go community will have to adjust but if, like us chess players, they learn to use the new technology then a new generation will surely arise with an understanding and aesthetic that is different and in some ways superior to their predecessors. And with free top class training at their finger tips the best players will develop much younger.
It’s a shock to the human ego that machines can emulate our intelligence. But rather than fight against them we should embrace the opportunities they bring.