Google’s AlphaGo Defeats World’s Top Human Go Player

Computers mastered chess decades ago, but Google’s DeepMind has its sights set on Go, a more complex game that pushes machine learning to its limits. After defeating several of the top players in the world, the “AlphaGo” AI has now bested Ke Jie, the current Go world champion. It was a narrow victory, and there are still two more games to play in the series. However, DeepMind has accomplished something that would have been impossible a few years ago.

Go is considered an “intuitive” game because of the incredible complexity of the board, which is a 19 x 19 grid. That means there are 361 potential locations for a player to place a stone. The goal is to completely surround your opponent’s stones to capture them and eventually control more than 50 percent of the board.

With chess, there are only 64 squares and the pieces can only move in a certain way. It’s possible for a moderately powerful computer to brute force all possible moves. With Go, there are more possible configurations than there are atoms in the universe. That’s why AlphaGo was designed to play Go like a human. It uses a pair of neural networks to decide on several promising moves, then it simulates each one a handful of moves ahead.

AlphaGo previously defeated the European Go champion Fan Hui and former world champion Lee Sedol. The current competition with Ke Jie is seen as the ultimate test for DeepMind’s AI. If it can beat Jie, it can beat anyone. The first match was a close one with AlphaGo winning by just 0.5 points. Jie still said he was “shocked” and “deeply impressed” after being bested by the computer.

AlphaGo’s approach to learning Go has led to some interesting quirks in its play style. Jie opened the match using a pair of moves that AlphaGo is known for using. This maneuver has fallen out of favor with human players, but AlphaGo began using it to great effect as it learned. Despite tailoring his approach to AlphaGo, AI matched Jie move-for-move and ended just slightly ahead. AlphaGo is programmed to play conservatively with an eye toward winning, not completely destroying its opponent. So, this is the sort of result you’d expect when the AI is facing a highly skilled player.

It’s possible Jie will come back from his one-game deficit, but AlphaGo has already proved it can beat the best humans. This clash of man and machine is taking place at an AI conference in China. The event runs through May 27th, and will include a number of other games played by AlphaGo that follow new game formats.

Now read: What are artificial neural networks?

Let’s block ads! (Why?)

ExtremeTechExtreme – ExtremeTech