Software Wins Go Match: Google's Deepmind AI Beats Grandmaster

Software Wins Go Match: Google’s Deepmind AI Beats Grandmaster

Artificial intelligence software has been taken to the next level after Google’s Deepmind AI software won a best out of five Go match against Lee Sedol, the greatest Go player in the world, by beating him in the first three games. The remaining two games will still be played in order to determine what the final match score will be, and maybe allow Sedol to save face by winning at least one match. Andrew Okun, president of the American Go Association commented on the software milestone.

“It is a huge landmark for Go, a long-anticipated moment in the West and — to a greater degree than I imagined — a shock to many people in Asia.”

Go is a strategy game that is believed to be 2,500-years-old. The game board consists of a grid that is 19-by-19 squares. Players compete against each other by placing white or black stones in order to amass the most territory on the grid. Although the rules are simple, the concept is difficult to master. Go is considered to be extremely more complex than chess.

Lee appeared to be very distraught and embarrassed after losing to the software. Lee made sure to take the time to apologize to the fans watching for his inability to win a match. Lee commented that he felt great pressure during the third match. Now that the best-of-five series has been won by the software, the pressure is off Lee and he thinks he may have a better chance at winning one of the two remaining games.

Artificial intelligence experts have been unable to explain how the AI software was able to beat the best Go player in the world. These experts believed that the software capable of what Deepmind accomplished should have been at least five-to-10 years away. With their victory, Google has placed themselves at the top of the AI foodchain.

The AlphaGo software that Google created was first introduced to the world in the January edition of the journal Nature. At the time of the publication, Google claimed the AlphaGo software had competed against, and beat, Fan Hui, the best Go player in Europe. After the October match against Hui, the software competed against itself in thousands of Go matches. Google was also able to run up to 100 versions of Go in parallel. By doing this, the software was able to get years of experience in much less time.

Software creators are now considering options in order to see how their artificial intelligence will react to certain situations. One option being considered is to alter the Go game board in order to see with the software will do as a reaction.

While playing the Go match against Sedol, the software showed that it had to ability to use a strategy that no other human player had come up with. One of the moves, in particular, left Go experts scratching their heads due to the complexity of the maneuver. Michael Redmond was the game commentator and is also an exceptionally skilled Go player. Redmond stated that “It’s playing moves that are definitely not usual moves” and the software was “coming up with the moves on its own.”

Google has come a long way in developing artificial intelligence for the world. So far, they have created AI software for writing e-mail, suggesting videos based on past preference, and most famously, automobiles that can drive themselves.

[Image Via AP Photo/Lee Jin-man]

Comments