AlphaGo strikes back to wind up Weiqi game series with fourth win

Lee Sedol loses the last match, which was close and intense.

After losing one game to its human opponent on Sunday, Google's AlphaGo struck back on Tuesday to wind up the five game series with an impressive 4-1 score. The computer program made up for a big mistake early in the game and went on to win in what commentators called a close and intense match. Both players used the full two hours allotted to each. After 280 moves, Sedol resigned.

After his first win on Sunday in the fourth game, Sedol had opted for a challenge in the final game by choosing to play with black, despite knowing it gave AlphaGo an edge, from observations on previous games. Did black do him in? The games will be discussed and studied in detail back at DeepMind. For now, artificial intelligence has scored a historic win over man.

According to Google, the techniques from AlphaGo will be applied to challenges like instant translation, smartphone assistants and advances in health care.

Lee Sedol said at the post-game press conference, "I am very sorry that the Google DeepMind Challenge Match is over. Today I wanted to bring the match to a successful conclusion. Personally, I am regretful about the result, but would like to express my gratitude to everyone who supported and encouraged me throughout the match." To a question, Lee said, "I have questioned at some points in my life whether I truly enjoy the game of Go, but I admit that I enjoyed all five games against AlphaGo. After my experience with AlphaGo, I have come to question the classical beliefs a little bit, so I have more study to do."

Demis Hassabis, Co-Founder and CEO of DeepMind, said, "We wanted to see if we could build a system that could learn to play and beat the best Go players by just providing the games of professional players. We are thrilled to have achieved this milestone, which has been a lifelong dream of mine. Our hope is that in the future we can apply these techniques to other challenges — from instant translation to smartphone assistants to advances in health care."

Earlier, South Korean Lee Sedol redeemed some status for humans after a resounding win against artificial intelligence in the complex Chinese game of Weiqi. On Sunday, he beat Google's computer program AlphaGo which had defeated Sedol in three games.

Surprisingly, Sedol who was under pressure having lost three straight games got his moves right and the machine seemed to fumble. According to Google, AlphaGo played baffling moves that were confused between mistakes and strategies.

According to tweets from DeepMind founder Demis Hassabis, AlphaGo made mistakes in game 4. The AI got confused on move 79 and realized its error only by move 87, Hassabis said. The program adjusts its playing style based on its evaluation of how the game is progressing.

An elated Lee Sedol said at the press conference on Sunday that the win was so valuable that he wouldn't exchange it "for anything in the world". Sedol who won playing with white offered to play black, or first, which was a bigger challenge as AlphaGo fared better playing white, or second.

"I really do hope I can win with black," he said, "because winning with black is much more valuable."

The game Go is played primarily through intuition and feel, making it that much difficult for a machine. What looks as simple as players placing white or black stones by turn on a board and trying to capture the opponent's stones or surround empty space to make points of territory is complex. Go is a game of profound complexity and according to Google there are more possible positions in Go than there are atoms in the universe.

AlphaGo works on a combination of deep neural networks as in the human brain and machine learning. Trained by supervised learning both from human experts and by reinforcement from self-play, it is the first computer program to beat a human at the 2500 year old Chinese game.

Google DeepMind will donate the UDS$1 million in prize money to UNICEF, STEM (science, technology, engineering, and math) charities, and Go organizations.

Related topics : Artificial intelligence