Artificial Intelligence Defeats Top Player Of Complex Strategy Game "Go"

Go is a traditional Chinese game, and it's considered the most complex game played by humans. Tatiana Belova/Shutterstock

They might not be able to feel or think yet, but artificial intelligence is progressing at an impressive pace, solving problems thought to be unsolvable by machines.

A computer program can now defeat a professional human player in the classic strategy game Go, as reported in this week's issue of Nature by a team from Google DeepMind. Go is often considered the most complex game humans can play.

Go has been regarded as a huge challenge for artificial intelligence, and a game-solving algorithm was believed to still be a decade away. Solving it indicates that AIs have the potential to surpass humans in fields that are traditionally considered too difficult.

Go is an ancient Chinese game with over 40 million players worldwide. The game takes place on a 19x19 grid, and the two players place either black or white stones on the board and try to conquer as much territory as possible by surrounding the opponent’s stones. While there are computer versions of the game, they can be easily defeated by even the top amateur players.  

The new program, called AlphaGo, uses two cornerstones of artificial intelligence: Monte Carlo Tree Search and deep neural networks. The first allows the AI to work out each possible move with a probability, and combined with a set of rules, allows the program to predict the right moves. Neural networks, designed to resemble how the brain works, allow the AI to go beyond prediction; the machine can picture how the moves will unfold and select successful paths, and it was eventually then taught to win.

 

 

Above is a video explaining the research. YouTube

“For us the most significant aspect of all of this is that AlphaGo isn’t just an expert system built with handcrafted rules, but instead, it uses general machine learning techniques to figure out itself how to win at Go,” Demis Hassabis, from Google DeepMind, said at a press conference.

Optimizing the two techniques took a significant effort. In a game of Go, there are over 10359 legal moves, which is larger than the number of atoms in the universe. By comparison, chess "only" has 10123 legal moves. AlphaGo was able to beat European Go champion Fan Hui five matches to zero.

And AlphaGo won't stop there. It has an upcoming challenge that could be the equivalent of the famous Deep Blue versus Kasparov bout: It will play Lee Sedol, acknowledged as the top Go player in the world in last decade, in Seoul in March.

Comments

If you liked this story, you'll love these

This website uses cookies

This website uses cookies to improve user experience. By continuing to use our website you consent to all cookies in accordance with our cookie policy.