AlphaZero

AlphaZero

In 2017 the chess world was shaken to its core when Stockfish (the world's strongest chess engine) was defeated in a one-sided match. It was not defeated by a human but by an unknown computer program that seemed to be otherworldly—AlphaZero. 

Let's learn more about this powerful chess entity. Here is what you need to know about AlphaZero:


What Is AlphaZero?

AlphaZero was developed by the artificial intelligence and research company DeepMind, which was acquired by Google. It is a computer program that reached a virtually unthinkable level of play using only reinforcement learning and self-play in order to train its neural networks. In other words, it was only given the rules of the game and then played against itself many millions of times (44 million games in the first nine hours, according to DeepMind).

AlphaZero uses its neural networks to make extremely advanced evaluations of positions, which negates the need to look at over 70 million positions per second (like Stockfish does). According to DeepMind, AlphaZero reached the benchmarks necessary to defeat Stockfish in a mere four hours.

AlphaZero chess
DeepMind logo. Image: deepmind.com.

AlphaZero runs on custom hardware that some have referred to as a "Google Supercomputer"—although DeepMind has since clarified that AlphaZero ran on four tensor processing units (TPUs) in its matches.

In December 2017, DeepMind published a research paper that announced that AlphaZero had easily defeated Stockfish in a 100-game match. AlphaZero would go on to defeat Stockfish in a second match consisting of 1,000 games; the results were published in a paper in late 2018.

AlphaZero stockfish chess
The AlphaZero-Stockfish matches changed the chess world.

Unfortunately, AlphaZero is not available to the public in any form. The match results versus Stockfish and AlphaZero's incredible games have led to multiple open-source neural network chess projects being created. Leela Chess Zero, Leelenstein, Alliestein, and others try to emulate AlphaZero's learning and playing style. Even Stockfish, the conventional brute-force king, has added neural networks.

In 2020 DeepMind and AlphaZero continued to contribute to the chess world in the form of different chess variants. When DeepMind and the AlphaZero team speak, the chess world listens!

From the moment it stepped onto the scene, AlphaZero has changed chess by spawning a new generation of neural network chess engines, by contributing to chess variants, and through its transcendent games.

AlphaZero’s Accomplishments

As mentioned, AlphaZero defeated the world's strongest chess engine, Stockfish, in a one-sided 100-game match in December 2017 (scoring 28 wins, 72 draws, and zero losses). The public was given 10 example games from this match, and the chess world's reaction was borderline disbelief. GM Peter Heine Nielsen likened watching AlphaZero's games to seeing a superior species landing on earth and showing us how to play chess:

Other grandmasters shared Nielsen's sentiment, including the legendary GM Garry Kasparov, who told Chess.com, "It's a remarkable achievement.... It approaches the 'Type B,' human-like approach to machine chess dreamt of by Claude Shannon and Alan Turing instead of brute force."

Others questioned the results because of the disparity of hardware used in the first match. Some also found it unfair that Stockfish was not allowed to use its opening book and its endgame tablebase.

GM Hikaru Nakamura stated: "I don't necessarily put a lot of credibility in the results simply because my understanding is that AlphaZero is basically using the Google supercomputer, and Stockfish doesn't run on that hardware; Stockfish was basically running on what would be my laptop."

AlphaZero chess
Nakamura had some doubts about the first AlphaZero-Stockfish match. Photo: Maria Emelianova/Chess.com.

Roughly one year after the first match, DeepMind published a new paper that announced an updated version of AlphaZero had defeated Stockfish in a 1,000-game match. This time, the current version of Stockfish (version 9 at the time) was used, Stockfish was able to use a strong opening book in many of the games, the time controls were adjusted (with Stockfish having large time advantages), and Stockfish was run on the same type of hardware used in the Top Chess Engine Championships (TCEC).

The results didn't change much—AlphaZero defeated Stockfish again with a score of 155 wins, 839 draws, and 6 losses.

AlphaZero And Chess Variants

In 2019 and 2020 GM Vladimir Kramnik was able to spend some time with AlphaZero and the DeepMind team to explore chess variants and co-wrote a paper with DeepMind about the exploration of new chess variants, including sideways pawns, no castling, torpedo chess (where pawns can always move forward one or more squares).

In September 2020 Chess.com hosted a roundtable discussion with Kramnik and members of the DeepMind team where they discussed variants and other topics. You can watch the full video here:

Many of these chess variants (and more) have been added to Chess.com. This article outlines the new chess variants and how to play them. If you want to check out any of these variants for yourself, simply head over to Chess.com/variants or hover your mouse over the "Play" button in the menu bar and select "Variants":

AlphaZero chess variants
Playing the new chess variants on Chess.com is easy.

After you select "Variants," you are directed to the Chess Variants Page. All you have to do is select a variant and press "Play."

AlphaZero chess
Chess.com's new variants page.

AlphaZero Games

In this first game example, we see some of the magic that AlphaZero shocked the world with in the first match. AlphaZero gambits a pawn in the opening and immediately goes on the attack. After 19...Kxh6 Stockfish is up a piece, but the king is not safe, and the entire queenside is undeveloped:

AlphaZero chess engine
Stockfish is up a piece, but the king is not safe, and the queenside is undeveloped.

AlphaZero keeps up the pressure, but its compensation for the piece is mostly unclear to us mortals. Only in hindsight can we tell that a couple of Black's pieces (most notably the a8-rook and queen's knight) will never really be part of the game. After 36.Qe6, the position has crystallized, and AlphaZero wins convincingly:

This second game example is from the second AlphaZero-Stockfish match. AlphaZero puts on a positional clinic and tortures Stockfish with the bishop pair in the endgame after 45. Bxe4. Here is the full game:

In the following video, GM Robert Hess covers this fantastic game in great detail:

Conclusion

You now know what AlphaZero is, what it has accomplished, and more. If you are interested in seeing what you can learn from AlphaZero's play, check out this great series of video lessons by Chess.com's IM Danny Rensch.