Google DeepMind Beats Human At Complex Asian Boardgame 'Go'
Google’s DeepMind Technologies, a British artificial intelligence company announced it has moved a step further in the quest of acquiring ultimate AI technology, it has built a system AlphaGo capable of beating the best players around at the immensely popular Asian board game ‘Go’.
Go is a game that involves placing black and white pieces on a 19×19 board and trying to surround empty space to make points of territory, and then removing opponent pieces. First played in China 2500 years ago, it has always drawn human attention and imagination courtesy the intellectual depth and subtlety required to play it. It is played by the 40 million people worldwide.
Naturally, it is as complicated as it is interesting. Even when compared to the most complex of games such as Chess, Go has a far broader array of permutations and combinations. While Chess has approximately 35 legal moves in each turn, Go has as many as 250. And, while a standard game of Chess would last 80 turns, for Go, the number is 250.
To develop AlphaGo, traditional AI methods were not adopted, rather a combination of ‘advanced tree search’ and ‘deep neural networks’ was set up. The neural networks study the Go board as an input in a twelve-layer fashion. The neural networks divide into the ‘policy network’ that selects the next move to play and the ‘value network’ that predicts the winner of the game after each turn.
DeepMind followed a three-stage process. Firstly, professional games were studied and more than 30 million moves were analysed. This automatically enabled AlphaGo to recognise almost all possible patterns and stages of the game. Now, it played itself millions of times, learning from its mistakes and get ‘incrementally better’. The final step, the Monte Carlo Tree Search, incorporates all intuitive knowledge to make long-range plans that would be used in matches. Tested against rival AIs, AlphaGo won 499 out of the 500 matches it played.
In a lopsided contest last October, the system beat three-time European champion Fan Hui in what is now seen as a symbol of its ultimate supremacy in the game. In March, AlphaGo will play legendary world champion Lee Sedol to test its proclaimed calibre.
It will be interesting to see how the 40 million Go-playing people will react on realising that the best player in the world is not one of them, if that is established.
Source: #-Link-Snipped-#
Go is a game that involves placing black and white pieces on a 19×19 board and trying to surround empty space to make points of territory, and then removing opponent pieces. First played in China 2500 years ago, it has always drawn human attention and imagination courtesy the intellectual depth and subtlety required to play it. It is played by the 40 million people worldwide.
Naturally, it is as complicated as it is interesting. Even when compared to the most complex of games such as Chess, Go has a far broader array of permutations and combinations. While Chess has approximately 35 legal moves in each turn, Go has as many as 250. And, while a standard game of Chess would last 80 turns, for Go, the number is 250.
To develop AlphaGo, traditional AI methods were not adopted, rather a combination of ‘advanced tree search’ and ‘deep neural networks’ was set up. The neural networks study the Go board as an input in a twelve-layer fashion. The neural networks divide into the ‘policy network’ that selects the next move to play and the ‘value network’ that predicts the winner of the game after each turn.
DeepMind followed a three-stage process. Firstly, professional games were studied and more than 30 million moves were analysed. This automatically enabled AlphaGo to recognise almost all possible patterns and stages of the game. Now, it played itself millions of times, learning from its mistakes and get ‘incrementally better’. The final step, the Monte Carlo Tree Search, incorporates all intuitive knowledge to make long-range plans that would be used in matches. Tested against rival AIs, AlphaGo won 499 out of the 500 matches it played.
In a lopsided contest last October, the system beat three-time European champion Fan Hui in what is now seen as a symbol of its ultimate supremacy in the game. In March, AlphaGo will play legendary world champion Lee Sedol to test its proclaimed calibre.
It will be interesting to see how the 40 million Go-playing people will react on realising that the best player in the world is not one of them, if that is established.
Source: #-Link-Snipped-#
Replies
You are reading an archived discussion.
Related Posts
I'm in 3rd year nd of mechanical branch...as no core companies come for the campus placement nd I need the job from the college campus...mentioned IT companies come and hire...
anyone plss give the equation of mass flow rate through inlet in turbojet engines and some material for engine inlet design.
Researchers from Rice University, Texas, have developed a graphene composite that has proven very effective in melting the ice build-up on helicopter rotor blades. The material could also aid in...
At NASA’s Goddard Space Flight Center in Greenbelt, Maryland, the James Webb Space Telescope team is in the process of installing the largest space telescope mirror the world has ever...
Any plz give me some ideas for mini project in mechanical