What Have Grandmasters Learned from Superintelligent Chess Programs?

What happens if more domains of knowledge have superhuman performance from AI?

The Elo (chess ranking estimate) of the best chess programs is about 3700 which is about 900 points beyond the 2881 maximum of Magnus Carlsen. This means Magnus might be able to get a draw in one out of 100 or 1000 games and the odds are very difficult to get a win.

The superhuman chess programs are great for teaching the best people how to get better. This is the same for GO programs. The best humans are able to interact and study the game.

Top 1%. Out of 100 random registered chess players the average top 1 out of 100 would have a 1600-1650 score.
Top 0.1%. Out of 1000 random registered chess players the average top 1 out of 1000 would have a 1900-2000 score.
Top 0.01%. Out of 10000 random registered chess players the average top 1 out of 10000 would have a 2200-2300 score.

Magnus Carlsen and other grandmasters have gained profound insights from training with and studying chess programs, fundamentally reshaping their understanding of the game. Chess engines like Stockfish, Houdini, Komodo, and later AlphaZero and Leela Chess Zero have acted as tireless sparring partners and analytical tools, revealing strategies and principles that were previously underappreciated or counterintuitive to human intuition. These lessons span positional play, pawn structures, king management, and even psychological preparation.

1. Dynamic King Management
One of the most striking lessons from engines is the nuanced role of the king, especially in the middlegame and endgame. Humans traditionally viewed the king as a piece to hide, castling early and keeping it safe behind pawns. Chess programs, however, treat the king as a flexible asset:

Middlegame Activity: Engines often delay castling or leave the king in the center if it’s safe, using it to block pawn advances or support central control. Carlsen has adopted this, occasionally forgoing castling to maintain flexibility, as seen in games where he repositions his king manually (e.g., Kf1-Kg2) rather than committing early. This reflects engine-inspired confidence in calculating safety.

Endgame Power: In endgames, engines demonstrate the king’s strength as an active piece. Carlsen, already an endgame virtuoso, refined this further—pushing his king up the board aggressively to support passed pawns or cut off the opponent’s king. For example, his 2018 World Championship games against Caruana showed engine-like king marches, a hallmark of studying Stockfish’s endgame precision.

2. Pawn Structure Fluidity and Side Pawns
Engines have revolutionized how grandmasters handle pawn structures, especially on the flanks:

Sacrificing Pawns for Activity: Programs frequently sacrifice pawns—particularly wing pawns—to open lines or gain piece activity. Carlsen has internalized this, often advancing or giving up a flank pawn (like h- or a-pawns) to create weaknesses or activate his rook. His 2016 game against Karjakin (Game 10, World Championship) featured an h-pawn push to disrupt Black’s kingside, a move engines often favor for long-term pressure.

Asymmetrical Structures: Engines don’t cling to symmetry or “perfect” pawn formations as humans once did. They’ll create imbalances (e.g., a broken queenside for a kingside attack) if the position demands it. Carlsen’s willingness to accept doubled pawns or isolated flank pawns for dynamic compensation mirrors Stockfish’s indifference to classical “weaknesses” when activity outweighs them.

Pawn Breaks: Studying engines, grandmasters learned to time flank pawn breaks (like g4 or b5) with surgical precision. Carlsen’s games often feature these breaks to challenge castled kings or open files, a tactic engines calculate flawlessly.

3. Positional Sacrifices

Engines excel at long-term positional sacrifices, which grandmasters like Carlsen have absorbed:

Material for Initiative: AlphaZero’s games against Stockfish (2017-2018) showcased sacrifices of pawns or even pieces for nebulous advantages like king safety or coordination. Carlsen, who trained with AlphaZero, began employing similar ideas—offering material to trap an opponent’s king or dominate an open file. His 2019 game against Wesley So in the Fischer Random World Championship included a pawn sac for a lasting initiative, echoing AlphaZero’s style.

Flank Pressure: Engines often push rook pawns (a- or h-) to provoke weaknesses, a tactic Carlsen now uses to unsettle opponents. This isn’t just about attack—it’s about forcing concessions, a lesson from Leela Chess Zero’s neural-net-driven creativity.

4. Concrete Calculation Over General Principles
Before engines dominated, grandmasters relied heavily on heuristics—rules like “don’t move the same piece twice in the opening” or “keep your king safe.” Engines prioritize concrete calculation, showing that exceptions abound:

Opening Flexibility: Carlsen’s eclectic opening repertoire (e.g., 1. e4, 1. d4, or sidelines like 1. b3) reflects engine training, where every move is evaluated on its merits, not tradition. He’s learned from Stockfish that “ugly” moves can work if they hold up tactically.

King on the Edge: Engines sometimes leave the king on h1 or a1 after rook lifts, a once-rare idea. Carlsen has used this to keep options open, especially in sharp positions where castling commits too much.

5. Endgame Mastery

Engines have turned endgame theory into an exact science, and Carlsen—already a prodigy in this phase—elevated his play further:

Pawn Endgames: Studying tablebases and engine lines, he’s mastered subtle king triangulations and pawn races. His 2014 game against Aronian, squeezing a win from a near-dead draw, shows engine-like precision in pawn endgames.

Flank Pawns in Endgames: Engines emphasize the power of outside passed pawns, even in rook endings. Carlsen’s ability to nurse a queenside pawn advantage (e.g., his 2018 tiebreak win over Caruana) owes much to engine insights about converting small edges.

Other Grandmasters’ Takeaways

Hikaru Nakamura: Known for speed chess, Nakamura has leaned on engines to sharpen his tactical intuition, adopting aggressive flank pawn pushes (like g4 in the Sicilian) that engines validate. His streaming career also shows him dissecting engine moves live, absorbing their logic.

Fabiano Caruana: A preparation monster, Caruana uses engines to find deep novelties, often involving king decentralization or pawn storms. His 2018 World Championship prep against Carlsen relied heavily on Stockfish and Leela to challenge classical setups.

Levon Aronian: Aronian’s creative style synced with AlphaZero’s unorthodox sacrifices, like trading queens for positional binds. He’s noted engines teaching him to delay king safety for dynamic gains.

Broader Impact
Engines didn’t just teach tactics—they rewired strategic thinking. Carlsen has said publicly (e.g., in interviews post-AlphaZero) that engines exposed how much humans underestimated certain positions—kings can survive exposure, flank pawns can be weapons, and “bad” structures can win if pieces harmonize. Other GMs, like Anish Giri, have remarked that engines made chess “more concrete,” shifting focus from aesthetics to results.

Carlsen and his peers learned to blend human intuition with machine precision—managing the king as a fighter, not a fugitive, and wielding side pawns as tools of chaos or control. The result? A generation of players who play more like engines without losing their human spark.

Living With and Improving with Superintelligence

The learning from the rest of us is how do we learn from other humans and learn from the AIs. We can work with the AIs. There can be massive flaws in our thinking on subjects. Once we get those insights then more domains of knowledge will go into the solved or mostly solved categories.

There are 3 year old with 1600 chess score and a ten year old learning with chess programs is becoming a grand master. Magnus became grand master at 13.

Humans that play with chess programs helping them can get better ranking but the best at cyborg games learn how to get even better results.