Supercomputer with innovative software beats Go Professional

MoGo vs Kim Myungwan 8p

For the very first time in history, and after 40 years of research, a program defeated a professional Go player in a 9 stones handicap game. This success was the result of revolutionary algorithms that were invented in the period 2006-2008. The match involved the collaboration between French researchers, who developed most of the software, and Dutch researchers, who provided the cutting-edge Supercomputer Huygens and contributed to the code. “MoGo played really well,” said Kim, who estimated its current strength at “two or maybe three dan,” though he noted that the program – which had a processing power more than 1000 times higher than the chess program Deep Blue – “made some 5-dan moves.” “I think there’s no chance on nine stones, it would even be difficult with eight stones. MoGo played really well; after getting a lead, every time I played aggressively, it just played safely, even when it meant sacrificing some stones. It didn’t try to maximize the win and just played the most sure way to win. It’s like a machine.” The game was played live the 7th of August at the U.S. Go Congress, with over 500 watching online on the internet Go server KGS.

“Congratulations on making history today,” game organizer Peter Drake told Olivier Teytaud, one of MoGo’s programmers.

The game in sgf format can be downloaded here

 

The middle of the game. The professional plays white and MoGo plays black.

Kim Myungwan

Kim Myungwan was born in 1978 in South Korea and became professional go player in 1994. He improved steadily from 1 dan to reach nowadays 8 dan, which is the second best rating that can be reached by professionals (the first being 9 dan). He participated in numerous international competitions, and won the US OPEN on the 9th of August 2008, two days after the match against MoGo.

MoGo

The MoGo project was started in 2006 as a Master project at the LRI in the University of Paris-Sud. Less than six month later, the program developed by Sylvain Gelly and Yizao Wang was already the highest ranked program on the 9x9 Computer Go Server. It still holds the first place on this server since August 2006. In June 2007, MoGo also proved its strength on full boards by winning the 12th Computer Olympiads in Amsterdam. A few month later, it was the first program to win a 9x9 game against a Go professional. The fast level increase of the program is due to new algorithms, denoted as “Monte-Carlo Tree Search”. Many innovations were invented by the developers of MoGo. The new algorithms are remarkably scalable with computing power, which makes the use of supercomputers particularly fruitful. The program is developed mainly by the French research organizations INRIA and CNRS. The Dutch research project GoForGo, funded by NWO, is collaborating on MoGo. The current MoGo team is constituted of Guillaume Chaslot, Christophe Fiter, Jean-Baptiste Hoock, Julien Perez, Arpad Rimmel, and Olivier Teytaud. A list of all the people who contributed to MoGo can be found here.

The Supercomputer Huygens

MoGo was running on the SuperComputer "Huygens", provided by the Dutch research organizations SARA and NCF. MoGo was using 25 out of the 104 nodes of the Supercomputer, i.e., 800 cores at 4.7GHz, with a floating point processing power of 15 Teraflop (more than 1000 times Deep-Blue). This was the most powerful supercomputer ever used by a board game Artificial Intelligence. Several people worked to grant the MoGo team an access to Huygens (e.g., Jaap van den Herik, Peter Michielse), to adapt the code and make it run on this architecture (e.g., Vincent van den Elzen, Walter Lioen, Thomas Herault and George Bosilca), and to test its level of play (Bernard Helmstetter, Fabien Lips).