Winning at Go
Nicolas Sabouret · 2020
Monte Carlo algorithms have been used in this way since the 1990s to win at Go. In 2008, the program MoGo, written by researchers at Inria, the French Institute for Research in Digital Science and Technology, achieved a new feat by repeatedly beating professional players. By 2015, programs had already made a lot of progress in official competitions, although they had failed to beat the champions. When young children learn to catch a ball, they learn by trial and error. They move their hands without much rhyme or reason at first, but they gradually learn the right movements to sync up with the ball’s movement. In this learning process, children do not rely on examples. They correct their errors over subsequent attempts. This is the idea behind reinforcement learning: the computer calculates a “score” for each possible action in each situation by “trial and error.”.