After 3 losses last week in a 5-game Go tournament, Lee Sedol, world champion for the board game of Go, has just won a game against Google DeepMind’s AlphaGo AI. This brings the score from 3-0 to 3-1, showing that while AlphaGo is perhaps better than any human at playing Go, it is not a perfect AI.
AlphaGo has already defeated the European Champion at Go, and has already won a majority of its games against Sedol due to the combination of two artificial intelligence techniques deep learning and Monte Claro Tree Search, which allow for highly successful, but evidently imperfect, strategies. Even this is an amazing feat of programming, however; Google has noted that “There are… more [possible positions in the game of Go] than the number of atoms in the universe, and more than a googol times larger than chess.”
While AlphaGo is extremely impressive, we are still quite a bit away from developing an AI that can pass the Turing Test. Such an AI would not only be able to process large number of outcomes, but also learn and solve all kinds of problems, not just pertaining to board games.
The final match in the AlphaGo series is set to take place on March 15, and can be viewed live on Youtube.