In a paper published in The Journal Science, the DeepMind team notes that AlphaZero is an improved version of its famous AlphaGo engine. After feeding basic rules of chess, shogi, and Go, it took AlphaZero, nine hours, 12 hours, and 13 days to learn the games respectively. Then It was then pitted against the world’s best AIs for these games.

Here’s how it went:

• Chess: AlphaZero won 155 games, lost just six, and drew the rest out of 1000 against Chess master StockFish.
• Shogi: DeepMind’s AI outperformed the world champion software Elmo in 91.2 percent games.
• Go: AlphaZero beat AlphaGo in 61 percent of games.

AlphaZero used a searching method called Monte Carlo Search Tree (MCST) to determine it’s next move. This method gave it an advantage over its competitors.

In an editorial in Science magazine Chess Champion Garry Kasparov praised AlphaZero’s playing style:

(…) I admit that I was pleased to see that AlphaZero had a dynamic, open style like my own. The conventional wisdom was that machines would approach perfection with endless dry maneuvering, usually leading to drawn games. But in my observation, AlphaZero prioritizes piece activity over the material, preferring positions that to my eye looked risky and aggressive. 

The secret of DeepMind’s speedy learning and expertise is Google‘s 5000 Tensor Processing Units (TPU). To put that into context, A single TPU can process 100 million photos from Google Photos in a day, so AlphaZero benefits greatly from having a number of these.

Earlier this year, we saw that Open AI’s bots beat Dota 2 players. What game will AIs master next? Any guesses?