AlphaGo – Google DeepMind’s AI that beat Go champion for first time

With trevigintillion possible positions in the game Go, a computer program is able to crack that complexity and win the game

Google DeepMind has come out with its AlphaGo artificial intelligence that can crack trevigintillion (1072) possible positions in the game Go and beat a human champion.

“It was the first time a computer program has ever beaten a professional Go player,” Demis Hassabis from Google DeepMind wrote in a blog.

What makes Go a hard task in AI is the magnitude of complexity in the game. The possible positions amount to trevigintillion, according to Google DeepMind, which is: 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.

“That’s more than the number of atoms in the universe, and more than a googol times larger than chess,” wrote Hassabis.

Google DeepMind tested AlphaGo against a three-time European Go champion. AlphaGo won by 5 games to 0. This was after the AI competed with other computer programs that have mastered Go and won 499 of 500 games.

AlphaGo was trained and built using Monte Carlo tree search (MCTS) and deep neural network algorithms. MCTS is like a decision tree, but selects optimal nodes to reach a decision and uses random sampling. A deep neural network is an artificial neural network that has many hidden layers where data passes through between the input and output.

It also uses reinforcement learning, which is like a trial and error method for adjusting the connections in a neural network to achieve better results or output.

“Traditional AI methods—which construct a search tree over all possible positions—don’t have a chance in Go,” Hassabis wrote.

AlphaGo was trained on 30 million moves from games played by human experts. It was able to predict a player’s next move 57 per cent of the time, surpassing the record of 44 per cent from other AI computer programs.

“AlphaGo learned to discover new strategies for itself, by playing thousands of games between its neural networks.”

Google DeepMind wants to apply its machine learning capability – the algorithms that execute AI – in AlphaGo to solve real-world problems such as climate modelling and complex disease analysis.

“Because the methods we’ve used are general-purpose, our hope is that one day they could be extended to help us address some of society’s toughest and most pressing problems.”

Join the TechWorld newsletter!

Error: Please check your email address.

Tags Google DeepMindAIGoogledeep learningmachine learningartificial intelligence

More about Google

Show Comments

Market Place

[]