简体中文
Artificial intelligence (AI) shocked the world again on Thursday as Google's computer program AlphaGo beat human Go champion Lee Sedol of South Korea in the second of a historic five-game match, taking a 2-0 lead at the ancient Chinese board game match.
The "match of the century" between a human and AI entered its second phase at 1 p.m. local time (0400 GMT) on Thursday at a hotel in Seoul. The match, which kicked off Wednesday, will last until next Tuesday. The third game is set to be held on Saturday after a one-day break.
AlphaGo, a computer program developed by Google's London-based AI arm DeepMind, showed a variety of "anomalies"humans cannot imagine and played in an "incomprehensible"manner that was seen by even professional Go players as mistakes at the first sight.
Those mistakes turned out to be made intentionally and from a broader perspective, which one commentator described as a way of thinking clearly different from humans.
AlphaGo's moves looked very strange and incomprehensible at the first sight, but those moves drove Lee's position very unfavorable in the end, another commentator said.
The landmark match was broadcast live on TVs and through Internet websites, including Google's YouTube, attracting hundreds of thousands of viewers around the world.
Defeats of Lee, one of the greatest Go players who has won 18 world championships for 21 years of his professional career, was a shock to the world as GO has been seen as the last game humans can dominate over machines due to its complex, creative and intuitive nature.
AlphaGo's two straight victories proved that such hopes were wrong though three games are left for Lee. It demonstrated a step forward further in AI's development after AlphaGo defeated European Go champion Fan Hui in October last year, marking the first AI victory over a Go professional.
After exchanging 211 moves with AlphaGo that lasted four and a half hours, Lee accepted his second loss. During the first face-off, Lee resigned an hour earlier after trading 186 moves.
Lee should win all of the three remaining games in order to win 1 million U.S. dollars in prize and to keep his reputation as well as hopes for a human victory over machines.
AlphaGo played in an incomprehensible and anomalous manner, which commentators said could be mistakes or algorithm errors at first sight.
The computer program started off with anomalies. Playing Black, it put its starter at the upper-right flower spot. In turn, Lee also placed his first white stone at a flower.
About one and half minutes later, AlphaGo put the third marker at a position beside the flower spot, an unconventional move different from the previous day's strategy. AlphaGo placed its first two pieces on the flower spots on Wednesday.
Exchanging conventional moves with each other, AlphaGo put its 13th marker at an anomalous position in the middle of the upper board, which a commentator said was a move he had never seen in the human Go world.
Lee"stably and calmly"responded to the irregular moves in an early phase. Commentators said he had been slightly dominating by the middle of the match.
Alphago, however, took counterattacks at the middle areas of the white built up by Lee, rattling his emotional calmness. In exchange for submitting a part of the areas, the human Go champion won a smaller part of the black stones set by AlphaGo.
The tight race was abruptly broken as AlphaGo's another mistakes began to be viewed as an intentional strategy from a broader perspective. At the first glance, commentators described the mistakes even as algorithm errors. In the end, AlphaGo's choices turned out right and calculated moves.
AlphaGo's moves may look like mistakes from humans' perspectives and according to ways that human players learned, but the result showed AlphaGo's broader perspective only aimed at winning an entire game despite losses in part, experts noted.
The AI boasts of a deep learning capability that enables the computer program to discover new strategies by playing games against itself and adjusting neural networks based on a trial-and- error process known as reinforcement learning.
Lee, 33, is seen as one of the greatest Go players in the world as he won 18 world championships for 21 years of his professional career. He recorded a winning rate of about 70 percent with 47 victories in professional matches, while AlphaGo posted a 99.8 percent rate with 504 victories and only one defeat.
Go originated from China more than 2,500 years ago. It involves two players who take turns putting markers on a grid-shaped board to gain more areas on it. One can occupy the markers of the opponent by surrounding the pieces of the other.
