Posted May. 26, 2017 07:10,
Updated May. 26, 2017 07:21
“One of the innovations of AlphaGo Master is that it actually relies much more on learning from itself. So in this version, AlphaGo has actually become its own teacher, learning from moves which are taken from examples of its own searches. That relies much less actually on human data than previous versions.”
Demis Hassabis, the founder and chief executive of Google’s DeepMind who developed the AlphaGo Master, an upgraded version of the AlphaGo artificial intelligence (AI), made the remarks at the Internet International Conference and Exhibition Center in the ancient Chinese city of Wuzhen on Thursday. That was his clear answer to the question on how much AlphaGo had improved since its defeat of South Korean Go grandmaster Lee Se-dol last year.
AlphaGo had a perfect victory in the second match of three over Ke Jie, the Chinese world Go champion after winning the first competition on Tuesday. Ke did his best to recover from his first defeat, only to fail. The AI defeated the human champion at the 155th move with tactics that were previously unimaginable
On Friday, AlphaGo will face a group of five players, all playing together to defeat the AI. Each side will be given 2.5 hours and three 60-second chances to calculate moves. Kim Sung-ryong, a Korean Go master who visited the site, said that human players have the advantage of using group wisdom but that the time limit would be too short for humans.
Hassabis said that the group matches were intended to “truly test what (AlphaGo) knows” and to provide Go masters to have an opportunity to have “the most beautiful” Go games. Asked whether AlphaGo would play against more humans, he said he would think about it after the current event is over.