Google’s AI AlphaGo is now the world’s top chess player, defeating or drawing 100 games against a world-champion chess program, after only four hours of learning.
As the name suggests, the algorithm was first used to master the ancient Chinese game of Go.
And master it did.
As recently as October 2017, DeepMind announced that its AI had achieved superhuman performance of Go starting from scratch and learning by playing against itself.
In a newly published scientific paper, the company said it had used a similar approach adapted to the games of Chess and Shogi (Japanese chess), which are arguably not so well suited to AlphaGo’s algorithm relative to Go.
“Starting from random play, and given no domain knowledge except the game rules, AlphaZero achieved within 24 hours a superhuman level of play in the games of chess and shogi […]”, the company said.
It took the program four hours to outperform Stockfish, reportedly one of the strongest open source chess programs.
The new algorithm, called AlphaZero, is a generic version of the AlphaGo Zero algorithm the company used in the context of Go.
Unlike traditional chess programs, which rely on strategies handcrafted by human grandmasters, the AlphaGo system uses deep neural networks and a “tabula rasa” reinforcement learning from games of self-play.
This means that the program is given the freedom to “invent” its own strategy as the game advances, the only limitations being the game’s rules.
Demonstrates that a general purpose reinforcement learning algorithm can achieve “superhuman performance across many challenging domains”, the paper said.
Reinforcement learning is a type of machine learning allowing machines to automatically determine the optimal behavior within a specific context, in order to maximize its performance.
DeepMind is a British AI startup, co-founded by UK AI and neuroscience researcher Demis Hassabis. In 2014, Google bought the company for £400 million (about $660 million).
Hassabis was himself a child chess prodigy. At 13, he reached an Elo rating 2300, at the time the second highest rated player in the world under 14.
Earlier in 2017, a documentary about the Go game between AlphaGo and 18-time world champion Lee Sedol came out.
You can check out the trailer here.