Google’s DeepMind AI Becomes One Of The Best Chess Players In The World In Just Four Hours

Not content with absolutely thrashing humanity at Go, Google’s DeepMind AI has now turned its attention to chess, and as expected it has put us all to shame.

The specialist AlphaZero program was able to beat one of the world’s best chess programs a mere four hours after learning how to play.

Out of 100 games against Stockfish 8, DeepMind either drew or won all of them.

Google hasn’t officially commented on the success of the experiment but the DeepMind research team have published their paper to Cornell University’s Arxiv site.

During the experiment the team gave the rules of chess to the AlphaZero software and told it to learn the rules by playing in simulations against itself.

Each time it played itself the software would learn from its mistakes and within four hours the team felt they were ready to take on Stockfish 8.

While Stockfish 8 has been beaten by another piece of software, the real achievement here is the sheer speed at which AlphaZero was able to learn the rules and effectively become an expert.

It’s no wonder that the team described its progress as “superhuman” in the paper.

This isn’t the first time DeepMind has been able to learn a game and then teach itself to an extraordinary level of proficiency.

In fact the team had set itself the bold challenge that within 24 hours AlphaZero would learn Go, Chess and another Japanese board game Shogi and then it would convincingly defeat a world champion in each.

AlphaZero took eight hours to learn and then beat the world-champion at Go which was, well, itself.

It then took a mere two hours of self-training to beat the leading Shogi software called Elmo.

Now while this is all very impressive it should be pointed out that all of these games are in an extremely closed environment which means that its job is limited to the rules of the game.

To see how DeepMind copes with a more chaotic environment the team is now working with video game developers Blizzard who make the real-time strategy game Starcraft II.

This would be a chance for Google’s AI to branch out into a world where humans have so far, managed to reign supreme.