The world's strongest Go AI · AlphaGo evolved to "AlphaZero" where you can learn any board game

byJotam Trejo

Speaking of "AlphaGo" created by DeepMind, an artificial intelligence (AI) development company with the same Alphabet as Google, is a big topic for Go AI as a program that defeated the world's strongest goographer. Such AlphaGo has evolved over and over and learn board games other than Go and become stronger AI "AlphaZeroI was born again.

AlphaZero AI beats champion chess program after teaching itself in four hours | Technology | The Guardian
https://www.theguardian.com/technology/2017/dec/07/alphazero-google-deepmind-ai-beats-champion-program-teaching-itself-to-play-four-hours

AlphaZero is an AI that is an evolutionary version of "AlphaGo Zero" developed based on the Go AI "AlphaGo" that surprised the world. AlphaGo Zero is a Go AI that allows you to freely appeal to the skill of Go without permission in the battle-style training even if you do not teach about Go, and you can read the following article on its ability to understand the details I will.

The world's strongest go program · AlphaGo's new version "AlphaGo Zero" has reached a level that can become stronger on its own - GIGAZINE


"AlphaZero" was born by applying such AlphaGo Zero's algorithm to apply it to board games other than Go. This AlphaZero has won Stockfish, the world's strongest open source chess engine with only 4 hours of learning.

Like AlphaGo Zero, AlphaZero is a board game AI that leads everything up to the way AI arbitrarily wins from how to hit, even if he does not teach how to play the game. Such AlphaZero won Stockfish, the world's strongest chess program in just 4 hours after starting chess learning. There were 100 stations in the game, but the breakdown was AlphaZero's 28 wins and 72 draws, he said he never lost.

AlphaZero is a type of machine learningReinforceI am learning board games using, for example, "Go"DefiniteThere is no need to teach expert knowledge such as.

From the following embedding, you can follow one of AlphaZero and Stockfish 's games.


AlphaZero seems to assume 80 thousands of next hands per second, but Stockfish assumes 70 million strokes in one second. In addition, data showing that the winning percentage will be higher in the first (white) is also available, and among the 28 victories of AlphaZero, the first was 25 times, and the remaining three times only succeeded in victory That's right.

Being the creator of AlphaGoDemis Hassabis"AlphaZero can acquire superhuman abilities in chess and shogi, etc. within 24 hours, even if you start with random play and do not have expertise excluding game rules, By breaking the World Champion program of each board game, it shows the accuracy as well. "


Chess' former world champion Gary Kasparov said, "After the introduction of AlphaGo, although we were expecting such outcomes, it is a remarkable achievement at all.We have a tremendous experience in chess learning I thought it was necessary, but I do not need any knowledge from humans ... ".

AlphaZero is the world's strongest shogi program with only 2 hours of learningelmoI am also victorious.