AlphaGo AI beats Go world champion in world first

Image
Video
Pictured above is a frame from a 15-minute summary from the fourth match between DeepMind's program AlphaGo and Lee Seedol.
Image credit: DeepMind
Date:16 March 2016 Author: Nikky Knijf Tags:, ,

A battle of strategic minds ended in a win for artificial intelligence yesterday when Google’s AlphaGo triumphed with a 4 – 1 win over Go world champion Lee Sedol.

Acquired by Google in 2014, DeepMind is a British company that specialises in artificial intelligence development. One of the company’s projects is AlphaGo, which uses tree search – an algorythm that uses hierarchical structure to find specific values in a set – to execute moves on the board.

Go, the game chosen to test the AI, was developed in China more than 2 500 years ago. That makes it one of the oldest board games still being played today. It is a two-player board game using stones coloured black or white. The goal of the game is to capture as many of your opponent’s stones as possible, while also cordoning off empty spaces. Captured stones and empty space are combined to score points and determine the winner. The game can also be played using a handicap. For weaker players, this involves stones being placed on specific areas on the board.

Go has been adopted by Western enthusiasts in recent years, but is still mostly played in China, Japan and South Korea.

But how is this more impressive than previous AI wins against high-ranked human players of other games? Although the rules of Go are simple, the strategy involved in playing and the vast number of possible games make it complex. On Google’s official blog the company explains the challenge as follows: “…Go is a game of profound complexity. There are 1 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 possible positions — that’s more than the number of atoms in the universe, and more than a googol times larger than chess.” And DeepMind writes: “the game of Go is widely viewed as an unsolved ‘grand challenge’ for artificial intelligence”.

The game conclusion yesterday was the first time AI had won without a handicap. Though the win is significant, what has many excited was that AlphaGo made mistakes, but learned from them. This paid off in the fifth game, when AlphaGo recovered from a mistake and won the final game.

Watch the video above for more about AlphaGo. The extended video below – it lasts more than seven hours – is gameplay and an analysis of the fifth played by Sedol and AlphaGo. Check it out to see how the game works and how AlphaGo played.​

Video credit: Google DeepMind