Google’s AlphaGo Beats Human Go Champion

  • 28/01/2016 AT 22:51 by It's a Gadget Staff
  • News, Technology

Google's AlphaGo beats human Go championGoogle’s AlphaGo beats human Go champion in a match of Go, an ancient Chinese board game. Beating humans at board games has been a great milestone in testing various computer algorithms in order to prove their competence. This tradition has started with the game of tic-tac-toe, but more complicated games have been added. Chess has been the go to game for testing algorithms for years. In the recent years, Google has also created an algorithm that can play a number of Atari games without any additional input other than the pixels on the display.

The Game of Go

The game in discussion, Go is a 2500 year old game invented in China. It is played with a board with square grid and black and white stones. The player try to surround the other player’s pieces with their pieces to capture that. In this way, one player has to capture more than 50 percent of the board to win the game.

How Google’s AlphaGo Beats Human Go Champion

It was not easy at first to beat humans at chess for computers as the 10120 number of possibilities. In case of Go, the number of possibilities becomes way past that barrier with an astonishing number of 10761 possibilities. It is not possible just to brute force through such a huge number in a game. In order to solve that Google DeepMind changed their approach. In this approach, one neural network predicts the next move that will be the most likely to achieve victory. The other neural network is used to reduce the depth of the search tree. AlphaGo was also trained on 30 million moves from expert human players followed by matches with other Go playing AIs. In those matches, AlphaGo won 499 times out of 500 matches. The final victory was achieved against Fan Hui, who is the top player of Europe. AlphaGo wins 5 out of 5 times without a single loss. It is a historical event that can be compared with 1997’s winning of Deep Blue against Garry Kasparov. DeepMind is also planning to test the potential of AlphaGo again in the next March in a match against the world’s top Go player, Lee Sedol.

Google’s AlphaGo beats human Go champion not just to be the expert in playing Go. The goal here is to create a system that can act to solve real world problems which are still too complicated for machines to solve with the present technology. The problems ranges from climate modelling to the analysis of complex diseases.


0 Comments

You can be the first one to leave a comment.

Leave a Comment