AlphaGo won a Go match versus European 2013, 2014, & 2015 champ Fan Hui (naturalized French citizen, formerly from China, presumably one of the strongest humans outside of China/Korea/Japan) in a London match in October by 5 games to zero. No handicap. At the same time they also played an "informal" 5 game match with shorter time controls. AlphaGo won it by 3-2. Now a match versus Lee Sedol is planned in March in Seoul with $1 million prize to be streamed live on YouTube. Fan Hui is a pro 2-dan. Lee Sedol is a pro 9-dan and one of the top 10 players in the world. DeepMind Technologies, later acquired by Google-Alphabet, wrote the program. It self-learns via self-play as well as play with foreign entities. AlphaGo also played test matches versus other top go programs, namely CrazyStone, Zen, Pachi, Fuego, and GnuGo, as well as playing the first 3 of these with 4 handicap stone disadvantage. Zen and Pachi each are rated 6 amateur dan on KGS. The results were massive wins for AlphaGo in every case. It is about 1000 Elo stronger than the best among those programs on even terms, beating them over 99% of the time; and 350 to 1100 Elo stronger even with 4 handicap stones beating them 77 to 99% of the time. And then on top of that, AlphaGo can run highly parallel, with one configuration using 1202 CPUs and 176 GPUs. This high-parallel version was the one used against Fan Hui and planned to be used versus Lee Sedol. "The Alphabet approach relies on the newest so-called deep learning approach combined with a more traditional type of algorithm known as a Monte Carlo, which is designed to exhaustively explore large numbers of possible combinations of moves. The researchers said they had also trained their program using input from expert human Go players." "Two deep-learning networks were used in AlphaGo: one network learned to predict the next move, and the other learned to predict the outcome from different arrangements on the board. The two networks were combined using a more conventional AI algorithm to look ahead in the game for possible moves." Incredibly, AlphaGo WITHOUT any search, simply using its trained neural network based eval, will beat Pachi using 100000 monte carlo rollouts per move, over 80% of the time. Lee Sedol is a 9-dan pro from Korea and probably really is more like 10 or 11 dan; the rating system stops at 9 but the plain fact is, some 9-dans win a hell of a lot more tournaments than others. Lee Sedol is one of the top 10 players in the world and is 600 Elo above Fan Hui. I would tend to doubt AlphaGo could beat Lee Sedol just based on the fact Lee Sedol would probably have beat Fan Hui 5-0 and 5-0, not 5-0 and 3-2. I think it is an amazing, incredible advance that AlphaGo can kick the butt of a pro 2-dan and even consider trying to take on Lee Sedol. Just a few years ago both seemed way out of reach. 20-author paper: http://www.nature.com/nature/journal/v529/n7587/full/nature16961.html Video: http://www.nature.com/news/google-ai-algorithm-masters-ancient-game-of-go-1.... Another piece with a video: http://www.nature.com/news/game-playing-software-holds-lessons-for-neuroscie... -- Warren D. Smith http://RangeVoting.org <-- add your endorsement (by clicking "endorse" as 1st step)