Google's DeepMind has been creating computer systems using AI and machine learning.
After AlphaGo defeated the Go world's champion, DeepMind researchers have their sight on chess. Using AlphaZero, which is a more general version of AlphaGo, it has been able to learn chess from scratch.
In 24 hours, AlphaZero taught itself to play chess well enough to beat one of the best existing chess programs around. And in just four hours, the AI was able to beat or tie previous top chess computer program in 100/100.matches.
Demis Hassabis, the founder and CEO of DeepMind who also loves chess, said that: "It doesn’t play like a human, and it doesn’t play like a program," Hassabis said at the Neural Information Processing Systems (NIPS) conference in Long Beach. "It plays in a third, almost alien, way."
Unlike most computer problems, Go and chess involve strategy and careful thinking of what is going to happen next.
In chess, each turn can play out in any one of 400 different ways. What this means, after just 4 moves, there are nearly 300 billion different possible positions.
Chess has a long history in AI. For example, in 1996, IBM’s Deep Blue has defeated the world champion at the time. But that program, is just like other conventional chess programs, as it required careful hand-programming.
A standard game of chess usually involves 40 moves, and this makes the number of possible outcome almost limitless. And because analyzing all those different moves is not feasible, AlphaZero AI was given only two rules of the game, and no further human input.
After playing the game against itself over and over again, and after hours, the researchers describe this as a "more human-like approach" to learning the game compared to traditional programs that run an algorithm.
The previous computer program that was the world champion, was Houdini, regarded as one of the strongest chess engine in the world, Stockfish and Komodo.