- Computers & Software»
- Computer Science & Programming»
- Programming Languages
AlphaGo: A Computer beats a Champion
What is Go?
It is an abstract board game for two players which originated in china and aims to surround more territories than its opponent. Being 2500 years old, it is the oldest board game that is enjoyed till date. It is a strategy game with vast number of possible games i.e. 10761 (more than the game of chess). It consists of two playing pieces - black and white which are placed on intersection points of a vacant 19x19 grid of lines in a board. There are different sizes of the board 9x9, 13x13, 17x17 and 19x19. Archaeological discoveries tells us that this game was played on 17x17 board before. This game is played with the help of simple rules but the strategy involved is much complex and confusing so this game is played by very less number people all round the world.
Artificial Intelligence has always left us spell bound with its ability to learn via machine learning. With the help of Artificial Intelligence, machines have always topped in intellectual games like Scrabble, Othello, Chess and many more but till date there wasn't any machine that beat human in a game of go. AI didn't leave that stone unturned too, recently Google's Deepmind came up with an AI named AlphaGo which beat the reigning champion in its own domain and also solved the mystery of the ancient game Go. It is combination of advanced tree search and deep neural networks. AlphaGo has mastered Go and achieved new milestones in the world of AI, it isn't just as expert system but it uses machine learning techniques to figure out on how to win the game.
AlphaGo - Secrets and Future Work
"10 years or so before anyone could build a computer program capable of beating a human GO champion" was stated in Wired Report two years ago but Google's Deepmind made it happen much quicker that expected. This breakthrough not only will revolutionize the world of AI but also depicts the that technology is advancing at a much faster rate than one can assume. Google acquired Deepmind- a self professed Apollo Program for AI in 2014. This morning, NATURE published a paper about AlphaGo which solved the mysteries of GO and also defeated the champion, FAN HUI.
This technology is called as deep learning as mentioned it the published paper. The researchers of Deepmind collected the strategies of expert players which had almost 60 millions move and trained AlphaGo to play the game of Go on its own. After its win against the champion, the researchers have planned to test it against LEE SEDOL.
AlphaGo uses a technique based on Monte Carlo tree search which finds the best counter move by narrowing down the possibilities using probability and statistics. It works on neural networks and other hardware which relays the message and communicates with the AI to identify and react to the scenario in front of it. It runs atop state of the art processors, GPU's which are generally used to render images for games but they also came handy with deep learning. The researchers used network of computers which consisted of 170 GPU cards and 1200 standard processors and CPU's.
At first AlphaGo had to face all of the leading Go programs where it won all of the 500 matches and proved its importance and deeplearning capabilities The researches were overwhelmed with joy and happiness when it defeated the champion. “It was one of the most exciting moments in my career,” said Nature's Chouard, who recalled the cheering from the programmers upstairs and the people near the defeated champ Fui. AlphaGo solved one to the complex game and became a champion but the researchers say that it is better for it to solve computer related problems other than solving real life problems. They say, AlphaGo has immense potential to grow and become the world's most intelligent machine learning program in the near future. They wish to create an machine with relatively great learning abilities and something that can shape the world. The scope for it's growth is very high and bright and it holds the key for the new technological advancement in robotics, machine learning and deep learning. In short, it is the pillar on which an beautiful future can be build on with machines being much more intelligent and smart to deal with real life problems to make our lives much more easier.