The best human Go player made a surprise move that a machine would never make ahead of his final match with AlphaGo
The two are in a best-of-five match of Go, the ancient Chinese board game in which no AI program had ever defeated a top-ranked human player without a handicap - until this series. AlphaGo beat Lee three times last week before losing the fourth.
But ahead of the fifth and final match, Lee made a surprise request to AlphaGo's creator Demis Hassabis: he wants to play with the black stones.
The move was a surprise given Lee told the press that he found AlphaGo to struggle more when it played with white stones. And it's a move that no machine learning program would have made, since it makes decisions based on probability, not intuition or for the purpose of challenging itself.
"AlphaGo finds it more difficult when it's playing with black as opposed to white," Lee said during the press conference after Game Four. "But since I won with the white, I really do hope that in the fifth match I could win with black, as winning with black would be much more valuable."
It's hard to prove whether AlphaGo really found playing black more difficult. But Lee's only win did come when AlphaGo played black.
Hassabis didn't hesitate to accept the request.
In Go, each player chooses either black or white stones, with black playing first. There's no point disadvantage for playing second, but it could alter the strategy and certain moves.
Regardless, Lee doesn't blame the color of the stones he played with, or any other external factor for his losses.