How the AI Revolution Impacted Chess (1/2)

by Joshua Doknjas
1/7/2022 – In 2017, AlphaZero shocked the chess world by crushing Stockfish in a 100-game match. Since then, chess engines have undergone a substantial revision based on deep learning methods to develop a neural network. Joshua Doknjas examines the impact of the AI revolution in chess. | Graphic: Europe Echecs

Winning starts with what you know
The new version 18 offers completely new possibilities for chess training and analysis: playing style analysis, search for strategic themes, access to 6 billion Lichess games, player preparation by matching Lichess games, download Chess.com games with built-in API, built-in cloud engine and much more.

The wave of neural network engines that AlphaZero inspired have impacted chess preparation, opening theory, and middlegame concepts. We can see this impact most clearly at the elite level because top grandmasters prepare openings and get ideas by working with modern engines. For instance, Carlsen cited AlphaZero as a source of inspiration for his remarkable play in 2019.

Neural network engines like AlphaZero learn from experience by developing patterns through numerous games against itself (known as self-play reinforcement learning) and understanding which ideas work well in different types of positions. This pattern recognition ability suggests that they are especially strong in openings and strategic middlegames where long-term factors must be assessed accurately. In these areas of chess, their experience allows them to steer the game towards positions that provide relatively high probabilities of winning.

A table of four selected engines is provided below.

Chess Engines

Engine

Type

Description

Stockfish 8

Classical

Relies on hard-wired rules and brute-force calculation of variations.

AlphaZero

Neural network

DeepMind’s revolutionary AI engine used self-play reinforcement learning to train a neural network.

Leela Chess Zero (Lc0)

Neural network

Launched in 2018 as an open-source project to follow the footsteps of AlphaZero.

Stockfish 12

(and newer versions)

Hybrid

Utilizes classical searching algorithms as well as a neural network.

The hybrid Stockfish engine aims to get the best of both types of AI: the calculation speed of classical engines and the strategic understanding of neural networks. Practice has shown that this approach is a very effective one because it consistently evaluates all types of positions accurately, from strategic middlegames to messy complications.

These two articles introduce a few concepts that the newer (i.e., neural network and hybrid) engines have influenced. Please note that the game annotations are based on work I did for my book, The AI Revolution in Chess, where I analyzed the impact of AI engines.

Clash of Styles

One of the biggest differences in understanding between older and newer engines can be found in strategic middlegames which involve long-term improvements by one side. As shown in many of the AlphaZero – Stockfish games, the older engines sometimes fail to see dangers due to their limited foresight. Relying solely on move-by-move calculation is not always enough to solve problems against the strongest opponents. This is because neural network engines excel at slowly building up pressure, making small improvements to optimize their winning chances, before gradually preparing the decisive breakthrough.

In the following game, the older engines believe that the opening outcome is quite satisfactory for Black, while the newer ones strongly disagree. Grischuk sides with the opinion of the neural network engines and understands that White’s long-term initiative is both practically and objectively extremely difficult for Black to handle.

 
New ...
Open...
Share...
Layout...
Flip Board
Settings
MoveNResultEloPlayers
Replay and check the LiveBook here
1.d4 Nf6 2.c4 e6 3.Nf3 d5 4.g3 Be7 5.Bg2 0-0 6.0-0 dxc4 7.Qc2 b5 The neural network engines have had a profound influence on this topical response to the Catalan. First of all, they clearly showed the drawbacks of 9.. .c6 as seen in the game. Secondly, they influenced a major idea from Carlsen: 9...Nd5! against Ding Liren. Finally, meeting 8.a4 with 8...Bb7!? has become a fashionable pawn sacrifice and a good attempt to neutralize White's initiative. 8.a4 The surprising 8.Ne5!? was used by Carlsen in Game 2 of the 2021 World Championship to avoid the concrete theoretical lines that may arise after 8.a4. c6 9.a4 Nd5 10.Nc3 f6 11.Nf3 Qd7 ½-½ (58) Carlsen,M (2855) -Nepomniachtchi,I (2782) Dubai UAE 2021 8...b4 Alternatively, Black can sacrifice a pawn with the fashionable 8...Bb7!? as Grischuk himself tried in 2021. Many commentators suspected that this is what Nepomniachtchi had intended to play against 8.a4. 9.axb5 a6 10.bxa6 Nxa6 11.Qxc4 Bd5 12.Qd3 Nb4 13.Rxa8 Qxa8 1/2-1/2 (40) Berkes,F (2671)-Grischuk,A (2775) Struga 2021 9.Nfd2 c6 A favourite of the older engines from the Black side, as they appreciate the extra pawn but severely underestimate White's long-term initiative. Nakamura had already played this 7...b5 line in several previous games with good opening outcomes, so Grischuk arrived to the game well-prepared. One of Carlsen's unused ideas from the 2018 World Championship Match, 9...Nd5! was later employed successfully against Ding Liren. 10.Nxc4 c5 11.dxc5 Ba6! (Carlsen's novelty) 12.Ne3 Nd7 13.Nxd5 exd5 14.c6 Rc8 15.Bf4 Nc5 0-1 (59) Ding,L (2805)-Carlsen,M (2875) Zagreb 2019 10.Nxc4 Qxd4 11.Rd1 Qc5 12.Be3 Qh5 13.Nbd2 Ng4 The alternative 13...Nd5?! does not create threats on the kingside, so White can immediately fight for some squares on the queenside. 14.Nb3 Nxe3 15.Nxe3 a6 15...a5 16.Nc4 and the a5-pawn is very vulnerable 16.Nc4 Ra7 17.Rac1 with tremendous positional compensation for the pawn. One game between the older and newer engines continued c5 18.Nba5 Qg5 19.h4! (taking squares away from the Black queen) Qf6 20.Qe4 Rc7 21.Rd3 g5 22.h5 Qg7 23.h6 and Black's position was completely crushed in Lc0-Stockfish, TCEC 2019. 14.Nf3 Nxe3 15.Nxe3 Grischuk notes in his annotations for New in Chess Magazine that the computer (older engines) gives "conflicting evidence" in these complicated middlegames, unable to reach a definitive conclusion. But in his opinion, it was clear that White can use his initiative to pose major objective and practical problems for Black. Despite the older engines' confidence in Black's setup, the level of understanding provided by human players and the neural network engines argue that the 9...c6 variation is highly dubious. a5 Another game between Lc0 and Stockfish saw 15...a6?! 16.Nc4 a5 17.Nfe5 Ra7 18.Rac1 c5 19.Qe4 Ra6 20.Nd2! Rd8 21.Nb3 with a crushing positional advantage in Lc0-Stockfish, TCEC 2019. 16.Nd4 Ba6 17.Rac1 Rc8 18.Bf3 Qg6 Grischuk pointed out the following line from his analysis: 18...Qe5 19.Ng4! Qc7 20.Qb3 Ra7 21.Be4! creating the threat of Nxe6, crashing through on the kingside. Black is forced to play a passive defensive move, for example, Bf8 22.Nxc6 Nxc6 23.Rxc6 Qb8 24.Qe3 Rxc6 25.Bxc6 Rc7 26.Ne5± 19.Be4 Qh5 20.Bf3 Repeating the position once. Qg6 21.Be4 Qh5 We have reached a type of middlegame position that the older engines really struggle with. White has clear long-term compensation, but there is no immediate way to break through. In such situations, they often overestimate the value of Black's extra material. On the other hand, the newer engines clearly see the strategic dangers in Black's position and find logical ways for White to make progress. 22.Kg2! An excellent idea which was most likely prepared by Grischuk with one of the neural network engines. Over the next few moves, Grischuk slowly improves his position by gaining space on the kingside and takes some squares away from Black's queen. 22.Qd2?! releases some of the pressure on the kingside. Following Bf6 23.Nxc6 Nxc6 24.Bxc6 Rab8 25.Bf3 Qe5 threatens b2, with a satisfactory position for Black. 22.Rd2 makes more sense, but I don't think doubling rooks on the d-file is the best way to build up the pressure. One correspondence game continued Ra7 23.Rcd1 g6 24.Nb3 or 24.Bd3 Bg5 25.Qc5 Rac7 is unpleasant, but defensible for Black 24...Qe5 25.Bd3 Bb7 26.Nc4 Qc7 and it was difficult for White to increase the pressure in F. Burmeister-S.Zielinski, correspondence 2019. 22...Ra7 23.h4 g6 24.f4 Qh6 Another correspondence game continued 24...c5 25.Bf3 Bb7 26.b3 Bxf3+ 27.Nxf3 Qh6 28.Nc4 Qf8 29.h5 and White proceeded to open the h-file and launch a decisive attack in L.Parsons-U.Burgarth, correspondence 2019. 25.Nb3?! In view of the improvement on Black's 25th move, stronger was 25.Ng4! Qg7 26.Ne5 Rac7 27.Bf3 h5 during the game, Grischuk spent a lot of time calculating 27...Bf6 28.Qc5 Bxe5 29.fxe5 Nd7 30.Qxa5 Bb7 but could not find a strong continuation here 28.Qe4 Bf6 29.Rc5± with a clear edge in A.Cubides-S.Zielinski, correspondence 2019. A very unpleasant defence for Black lies ahead. 25...Kh8? It was essential to fight for activity and not give White the chance to build up his queenside pressure. 25...c5! allows Black to reach a satisfactory middlegame. For example, 26.Nxa5 26.Nc4 Qf8 27.Nbxa5 Bd8 provides a reasonable defensive setup. 26...Bxe2 27.Qxe2 Rxa5 should only be slightly better for White. 26.Bd3! Securing the c4-square for the knight. 26.Nxa5 runs into f5 27.Bf3 e5! exploiting the poor minor pieces on f3 and e3. 26...Bb7 27.Nc4 c5+ 28.Be4 Ba6 29.Nbxa5 Qf8 30.Bf3 Rd8 31.h5 By advancing the h-pawn, White poses major problems on both sides of the board. Bf6 31...g5 may have been a better defensive try, but after 32.f5 Black remains under tremendous pressure. 32.Rxd8 Qxd8 33.Rd1 Rd7 34.Rxd7 Nxd7 35.h6! Suffocating Black's kingside. Nb6 36.Ne5 Bxe5 37.Nc6 Grischuk wins the arising endgame without any difficulties. Nc4 38.Nxd8 Ne3+ 39.Kf2 Nxc2 40.Nxf7+ Kg8 41.Nxe5 c4 42.Bg4 Nd4 43.Ke1 Kf8 44.Kd1 Ke7 45.e3 Nb3 46.Nc6+ Kf6 47.Nxb4 Bb7 48.Be2 Na5 49.Kd2 Nb3+ 50.Kc3 Nc5 51.a5 Ne4+ 52.Kxc4 Nxg3 53.Bd3 g5 54.fxg5+ 1–0
  • Start an analysis engine:
  • Try maximizing the board:
  • Use the four cursor keys to replay the game. Make moves to analyse yourself.
  • Press Ctrl-B to rotate the board.
  • Drag the split bars between window panes.
  • Download&Clip PGN/GIF/FEN/QR Codes. Share the game.
  • Games viewed here will automatically be stored in your cloud clipboard (if you are logged in). Use the cloud clipboard also in ChessBase.
  • Create an account to access the games cloud.
WhiteEloWBlackEloBResYearECOEventRnd
Grischuk,A2772Nakamura,H27611–02019E05Moscow FIDE GP3.2

Opening Developments

Perhaps the most popularized idea of the neural network engines is the h-pawn advance, where White pushes h4-h5-h6 (or Black pushes …h5-h4-h3) to cramp the opponent’s kingside by taking away some key squares. The idea itself is not at all new, but the newer engines have a much greater appreciation for it than the older ones. This has led to many new ideas in openings such as the Grunfeld, where the fianchettoed bishop on g7 can be targeted by an h-pawn attack. Tying back to the theme of long-term improvements, neural network engines understand the problems that it creates for the opponent in the long run.

Our next game surveys a cutting-edge approach against the Grunfeld. Its sharp rise in popularity from 2019 onwards coincides with the widespread use of neural network engines at the top level.

 
New ...
Open...
Share...
Layout...
Flip Board
Settings
MoveNResultEloPlayers
Replay and check the LiveBook here
1.d4 Nf6 2.c4 g6 3.h4!? A shocking idea that has become a popular and critical try against the Grunfeld over the past couple years. c5 3...Bg7 4.Nc3 d6 4...d5?! 5.h5 Nxh5 6.cxd5 e6 7.g4! Nf6 8.dxe6 Bxe6 9.e4 Bxg4 10.f3 Be6 11.Bg5 1/2-1/2 (32) Vachier Lagrave,M (2780)-Nepomniachtchi,I (2767) Jerusalem 2019 5.e4 Nc6!? was seen in two important and fascinating games at the World Cup: 0-1 (41) Fedoseev,V (2696)-Carlsen,M (2847) Sochi 2021, and 1-0 (30) Shankland,S (2709)-Svidler,P (2714) Sochi 2021. 4.d5 Bg7 Playing for a Benoni setup is one of the most logical responses to 3. h4. The pawn on h4 appears to be much more of a drawback than a strength in these structures. 5.Nc3 d6 6.e4 0-0 In view of the h-pawn advance that occurs in the game, MVL (an expert on both sides of the 3.h4 Grunfeld) introduced a new move order at the 2021 Candidates tournament. 6...e6 7.Be2 exd5 8.exd5 Nbd7!? 8...0-0 9.h5 transposes to the main game 9.Nf3 9.h5 is not so dangerous due to Nxh5 and the king safely stays on e8 for now 9...Ng4! taking advantage of the weakened g4-square. 9...0-0 10.Bf4 Qb6 11.Qd2 10.h5 Qe7 11.Bg5 Bxc3+ 12.bxc3 f6 13.Bd2 g5 14.0-0 Nge5?! 14...Nde5 should be preferred because it opens the diagonal for the bishop, covering f5 and e6. 15.Nd4 0-0 15.Nd4! and MVL soon ran into a lot of trouble in ½-½ (88) Ding,L (2805)-Vachier Lagrave,M (2767) Ekaterinburg 2021 7.Be2 e6 8.h5 exd5 9.exd5 The key idea for White in this line is based on playing h6 and cramping Black's kingside. White will argue that these long-term weaknesses are more important than his lack of development. Re8 9...Nbd7 appears to be slightly less accurate than 9...Re8. There are some lines where the knight should retain the option of developing to a6 and not block in the c8-bishop. 10.h6! Bh8 11.Bg5 Re8 12.Nf3 Qb6 13.Qd2 Ne4 opening up the diagonal for the bishop on h8, hoping to put pressure on the b2-pawn. 14.Nxe4 Rxe4 15.0-0! (ignoring the hanging b2-pawn and focussing instead on quick development) Qxb2 16.Rae1 Nf8 17.Bd3 Qxd2 18.Bxd2 Rxe1 19.Rxe1 Bg4 20.Re7 and Black's position was very unpleasant in J.de Mingo Matías-R.Khairullin, ICCF 2019. 10.h6! The critical try for an advantage. Most of the older engines prefer 10.hxg6 but this is significantly less challenging for Black to face. For example, fxg6 11.Nf3 Nbd7 12.0-0 Ne5 with a fighting middlegame ahead, where Black's chances are not worse. 10...Bh8 11.Bg5 Qb6!? Removing the queen from the pin and creating some problems for White on the queenside. 11...Nbd7 transposes to 9...Nbd7. 12.b3 Ne4 A rapid game between Grischuk and MVL saw 12...Qa5!? 13.Bd2 Qd8 and now 14.Kf1! is a very strong and important idea. The development of the g1-knight is delayed and thus Black cannot play ...Bg4. The game continued Ne4 15.Nxe4 Rxe4 16.Rc1 Rh4 17.Rxh4 Qxh4 18.g3 Qd8 18...Qh2 also leads to similar development problems after 19.Bg4 Nd7 20.Nf3 Qh1+ 21.Ke2 and the exchange of queens leads to a better endgame for White. 19.Kg2 Nd7 20.Nf3± ½-½ (39) Grischuk,A (2766)-Vachier Lagrave,M (2775) Paris (rapid) 2019. 13.Nxe4 Rxe4 14.Rc1 14.Kf1! is more accurate. One of the main points is shown in the following line: Na6 14...Re8 15.Rc1 Nd7 1/2-1/2 (40) Bacrot,E (2658) -Jones,G (2684) Malmo 2021 16.g3 Nf6 17.Bd3 15.Bd3 Rd4? 15...Re8 16.a3! and Black's knight and queen are misplaced. 16.Qe2 Bg4 17.Nf3 Rxd3 18.Re1! an important intermediate move that shows another benefit of the h5-h6 idea. Black is lost beause of the back rank mate threat. 14...Na6 15.Rh4!? White can't castle after this, but the rook on e4 was a very active piece so it made sense to exchange it. The alternative was the more natural 15.Nf3 Qa5+ 16.Qd2 Nb4 17.a4 when Black can reach a fine endgame by force after Nd3+ 18.Kf1 Qxd2 19.Nxd2 Rxe2 20.Kxe2 Nxc1+ 21.Rxc1 Bf5 15...Rxh4 16.Bxh4 Bf5 17.Bg4 In view of Black's tactical resource after 17.Bg4, it was stronger to kick back the f5-bishop with 17.g4! Bd7 18.Bg3 Re8 19.Kf1 and Black is under some pressure because of the weak d6-pawn. 17...Re8+? An extremely natural check, but it pushes White's king to safety on f1. Black could fight for counterplay with 17...Bxg4! 18.Qxg4 Nb4 bringing the knight into the game and threatening ...Nxa2 and ...Nd3+. 19.Qe2 is the only way to defend against these two threats, but Nxd5! exploits the loose bishop on h4. 20.cxd5? loses to Qb4+ 21.Kf1 Qxh4 when the extra pawn and the weak h6-pawn provide Black with a winning position. 18.Kf1 Bxg4 The aggressive 18...Re4 19.f3 Rd4 runs out of steam after 20.Qe1 Bd3+ 21.Kf2 Be5 22.Bf6! and Black's pieces in the centre are left without a purpose. 19.Qxg4 f5 Preventing Qd7, but severely weakening the e6-square. 19...Qc7 20.Nf3 followed by Re1 to exchange rooks. In the arising endgame, all of Black's pieces lack good squares and will be stuck defending. Meanwhile, White can target the weak dark squares on the kingside. 20.Qd1 Re4 21.Bg5 Nb4 22.f3 Re8 23.Qd2 Qc7 24.Nh3 Black has run out of opportunities for counterplay. The knight heads to the weak e6-square, deciding the game in White's favour. Qd7 25.Nf4 b6 26.Ne6 Na6 27.Re1 Be5 28.a3 Rc8 29.f4 Bh8 30.Ng7! The decisive breakthrough, highlighting the weaknesses around Black's king. Bxg7 31.Re7 Qxe7 32.Bxe7 Bxh6 33.Qe3 Nc7 34.Bxd6 Re8 35.Qh3 1–0
  • Start an analysis engine:
  • Try maximizing the board:
  • Use the four cursor keys to replay the game. Make moves to analyse yourself.
  • Press Ctrl-B to rotate the board.
  • Drag the split bars between window panes.
  • Download&Clip PGN/GIF/FEN/QR Codes. Share the game.
  • Games viewed here will automatically be stored in your cloud clipboard (if you are logged in). Use the cloud clipboard also in ChessBase.
  • Create an account to access the games cloud.
WhiteEloWBlackEloBResYearECOEventRnd
Paravyan,D2629Wagner,D25901–02020E73Moscow Aeroflot op-A3

The clash of chess styles between classical and neural network AI is fascinating to analyze. Many examples on this topic can be found in the famous AlphaZero – Stockfish games and in openings where the engines disagree on the evaluation, such as the Grischuk – Nakamura game. Their disagreement has led to major advancements in all popular openings, as old lines are revised, and new lines supported by modern engines are introduced into high-level practice.

Part 2 will examine another AI-inspired opening and the modern battle between two players armed with ideas from neural network engines.

Links


Joshua Doknjas is a FIDE Master from Canada and the author of two books on the Sicilian Najdorf and Ruy Lopez. He enjoys teaching, following, and writing about chess. Joshua is especially interested in the role of engines during opening preparation and understanding how AI has influenced modern chess.

Discuss

Rules for reader comments

 
 

Not registered yet? Register

We use cookies and comparable technologies to provide certain functions, to improve the user experience and to offer interest-oriented content. Depending on their intended use, analysis cookies and marketing cookies may be used in addition to technically required cookies. Here you can make detailed settings or revoke your consent (if necessary partially) with effect for the future. Further information can be found in our data protection declaration.