AlphaZero

AlphaZero is a <a href="/facts/Computer_program/e38jH9N9">computer program</a> developed by <a href="/facts/Artificial_intelligence/lJGauQwX">artificial intelligence</a> research company <a href="/facts/DeepMind/n5ehBwCx">DeepMind</a> to master the games of <a href="/facts/Chess/47ykvkHF">chess</a>, <a href="/facts/Shogi/z12HJ3vB">shogi</a> and <a href="/facts/Go_(game)/1fKhemww">go</a>. This <a href="/facts/Algorithm/fnl5NmRt">algorithm</a> uses an approach similar to <a href="/facts/AlphaGo_Zero/oUm1jEdz">AlphaGo Zero</a>. 
On December 5, 2017, the DeepMind team released a <a href="/facts/Preprint/ryoUUpCg">preprint</a> paper introducing AlphaZero, which would soon play three games by defeating world-champion chess engines <a href="/facts/Stockfish_(chess)/l8r2V9jK">Stockfish</a>, <a href="/facts/Elmo_(shogi_engine)/sjmgEsHK">Elmo</a>, and the three-day version of AlphaGo Zero. In each case it made use of custom <a href="/facts/Tensor_processing_unit/aVlg0LFF">tensor processing units</a> (TPUs) that the Google programs were optimized to use. AlphaZero was trained solely via <a href="/facts/Self-play_(reinforcement_learning_technique)/4t51aAn0">self-play</a> using 5,000 first-generation TPUs to generate the games and 64 second-generation TPUs to train the <a href="/facts/Neural_network/DDhJTfMc">neural networks</a>, all in <a href="/facts/Parallel_computing/nQzcpzQt">parallel</a>, with no access to <a href="/facts/Chess_opening_book_(computers)/B0r8teRD">opening books</a> or <a href="/facts/Endgame_tablebase/PSeEbcYT">endgame tables</a>. After four hours of training, DeepMind estimated AlphaZero was playing chess at a higher <a href="/facts/Elo_rating/elxHsGzk">Elo rating</a> than Stockfish 8; after nine hours of training, the algorithm defeated Stockfish 8 in a time-controlled 100-game tournament (28 wins, 0 losses, and 72 draws). The trained algorithm played on a single machine with four TPUs. 
DeepMind's paper on AlphaZero was published in the journal <a href="/facts/Science_(journal)/uTyzjkwD">Science</a> on 7 December 2018. While the actual AlphaZero program has not been released to the public, the algorithm described in the paper has been implemented in publicly available software. In 2019, DeepMind published a new paper detailing <a href="/facts/MuZero/mDU6F5rp">MuZero</a>, a new algorithm able to generalize AlphaZero's work, playing both Atari and board games without knowledge of the rules or representations of the game.

AlphaZero open-in-new

AlphaZero