From Mar. 9 to Mar. 15, Google DeepMind Challenge Math was held at Four Seasons Hotel, Seoul. A match between the top ranking 9-dan Korean Go player and the artificial intelligence with tree searching algorithm that assembles humans’ way of thinking may well be considered as a match of the century.
1st Match
During the early phase of the game, AlphaGo dexterously encounters most of Sedol Lee’s offensive moves. Especially, AlphaGo’s 32nd move to cut Lee’s black alignments in the upper side was critical to Lee, letting him give up making any further moves in the upper side. During the middle phase of the game, AlphaGo abruptly makes a ‘loose’ 80th move, leaving the left lower side vacant so that Lee can take a huge house. However, while Lee is satisfied with the lower black house, AlphaGo makes the 102nd move, taking up the entire right side and covering up the lower black house that it gave to Lee, finishing the game with the 139th move.
2nd Match
AlphaGo makes a provocative 43rd move, trying to save its 9th and 23rd moves, and the game in overall tilts to Lee. Then, AlphaGo makes the 61st move which connects the upper black alignments and the 63rd move to take up the middle house. As Lee puts the 64th move to contain the middle black house, AlphaGo makes the 65th and 67th moves to protect the upper edge. As Lee attacks black 47th move with 68th move, AlphaGo rats out with the 69th move as if it thinks the 61th move is enough to protect the middle middle of the 47th move. As a result, while Lee has been busy catching up with AlphaGo’s previous moves, AlphaGo has swiftly created a huge middle black house. AlphaGo finishes the game with the 211th move.
4th Match
AlphaGo starts the game with similar moves as the 2nd match. With the same alignments, Lee places the 12th and the 14th move to distract AlphaGo and takes up the upper edge first. With the 23rd and 25th moves, AlphaGo aims for the left side. Then, Lee puts the 40th move to start creating a middle point based white house. While the 51st, 53rd black moves and 52nd, 56th white moves are making crossing over of their houses, Lee makes an exquisite 78th move to protect his white moves on the right side and spontaneously claim the white house at the upper edge. As if taken aback by the 78th move, AlphaGo consistently makes several mistakes, ending with Lee’s first victory at the 180th move.
5th Match
With early mistakes of AlphaGo at the right bottom side, Lee starts from an advantageous position for the first time. It is common sense among professional Go players that focusing on the center is not desirable since the center involves too many uncertainties and possible moves. However, AlphaGo manages the middle side with no mistakes until the end, taking the 5th match. Overall, AlphaGo defeated Lee four games to one.
Another professional 9-dan Go player, Chang-ho Lee, said, “When I feel confident of taking major houses, I allow opponents to kill my minor moves, as much as he or she wants. AlphaGo is a cold hearted player that does not care about its past moves but only reacts to opponents’ moves to make moves with the highest possibility of winning. In that sense, AlphaGo’s playing style highly resembles mine”.