The Stockfish Swap
When Humans Become the Imitation Game
Chess has been a quiet companion throughout my life, less a passion than a persistent presence. My father collects chess sets and would display them around our home during my childhood. For a time he sponsored a chess club, mentoring local teenagers. Later, during my middle and high school years, I was required to play two games of chess daily as part of my curriculum. While I never developed the devotion to the game that my father possesses, I've maintained an appreciation for its depth and complexity. There's something compelling about its perfect information battlefield—a 64-square universe where luck plays no role, where victory and defeat depend solely upon the quality of your decisions. I follow professional chess casually but consistently, appreciating the strategic depth and psychological drama that unfold in major tournaments and championship matches.
Over the past couple of years, I've found myself increasingly preoccupied with a curious phenomenon in the chess world, something I've come to call the "stockfish swap." Named after the leading chess engine, this term describes a subtle but significant reversal in how we evaluate human chess performance. It's a transformation that I believe carries implications far beyond the checkered board.
The phenomenon is this: Chess computers are no longer compared to human players. Instead, humans are graded by how closely their moves match those calculated by a computer. The standard of judgment has inverted. The human is no longer the benchmark; the machine is.
This turning point fascinates me not only as a landmark point in chess history, but as a harbinger of similar inversions likely to unfold across domains as artificial intelligence continues its relentless advance. As we stand at this inflection point, it's worth examining how this swap occurred in chess, what it means for players and the game itself, and what it might tell us about the future of human creativity in an age of increasingly capable machines.
From Mechanical Turk to Deep Blue
Chess has long served as the quintessential testing ground for machine intelligence. As early as the 18th century, the "Mechanical Turk"—a chess-playing automaton that concealed a human operator—captivated European audiences precisely because the game was considered the epitome of human intellectual achievement. Wolfgang von Kempelen's elaborate hoax succeeded because the very idea that a machine could play chess seemed as implausible as a machine writing poetry or composing symphonies.
I find it striking how this perception persisted well into the modern era. When Claude Shannon published his groundbreaking 1950 paper "Programming a Computer for Playing Chess," he wasn't merely proposing a game-playing algorithm. He was suggesting that chess could serve as a measurable proxy for intelligence itself. The implicit assumption was clear: if a machine could play chess like a human, it would represent a significant step toward artificial intelligence.
This framing guided computer chess development for decades. Computer scientists built increasingly sophisticated chess programs with the explicit goal of matching human performance. Their success was measured in distinctly human terms—Elo ratings, titles, and competitive results against human players. Each milestone—computer defeats master, computer defeats international master, computer defeats grandmaster—was celebrated as progress toward the ultimate goal: defeating the world champion.
There's something almost mythic about how that moment finally arrived. In 1997, IBM's Deep Blue defeated Garry Kasparov, the reigning world champion and perhaps the greatest player in chess history. The headlines announced things like "Machine Beats Man," "Computer Conquers Human Champion," and "Silicon Defeats Carbon." The framing was consistent: the machine had finally reached the human standard.
Yet even in that moment of apparent machine triumph, I notice in retrospect that the comparison still flowed in one direction. Deep Blue was remarkable precisely because it could play like a world champion. The world champion wasn't remarkable because he could play like Deep Blue. The human remained the gold standard; the machine was the imitator.
The Quiet Reversal
The years following Kasparov's defeat saw rapid improvement in chess engines. Programs like Fritz, Rybka, and eventually Stockfish became commercially available, allowing players at all levels to access superhuman chess analysis. I find it telling how these programs were initially viewed as training tools, aids that could help humans improve their own play. The human remained the performer; the machine was the coach.
A subtle shift began around 2005-2010, however. As engines surpassed even the strongest human players by wider and wider margins, a shift in the language of chess commentary became noticeable. The direction of comparison began to reverse. Commentators no longer exclaimed that a program "plays like a grandmaster." Instead, they began to note that a grandmaster's move "matches the engine recommendation" or "deviates from Stockfish's top choice by only 3 centipawns."
What strikes me about this transition is how it happened without explicit acknowledgment. Nobody announced that humans would now be measured against machines rather than the other way around. It simply began to happen, a linguistic shift that reflected a deeper conceptual inversion.
This reversal accelerated dramatically with the advent of neural network-based engines like AlphaZero and it’s successor Leela Chess Zero. These engines not only outperformed humans but sometimes played in ways that contradicted centuries of human chess understanding. I find it fascinating how AlphaZero's famous victory over Stockfish in 2017 represented a double inversion: a neural network defeating not just human players but traditional chess engines, challenging the assumptions of both human and computational chess wisdom.
Suddenly, human intuition itself seemed suspect. Positional evaluations that had been accepted dogma for generations were overturned by neural networks that demonstrated superior results through methods humans struggled to comprehend. Classical principles like "knights before bishops" or "control the center with pawns" were revealed as approximations rather than absolutes. The neural networks, trained without human preconceptions, discovered different, and apparently more effective, approaches.
By 2020, I could see that the reversal was complete. During the 2021 World Chess Championship between Magnus Carlsen and Ian Nepomniachtchi, the Lichess website analyzed every world championship game in history using Stockfish, comparing human moves to computer recommendations. What stood out to me was not just the analysis itself but its framing. The headline wasn't that Stockfish played better than the world champion; it was that the world champions were playing with unprecedented accuracy, approaching Stockfish's standard. The human had become the imitator; the machine was now the model.
This is the defining feature of the stockfish swap: not the moment when machines are able to outperform humans, but the subsequent reorientation of human activity around machine-defined ideals.
I've contemplated this transformation extensively, and what stands out to me most is not that it happened, but how naturally and inevitably it seemed to occur. There was no resistance, no debate about whether machines should be the standard. Once chess engines demonstrated clearly superior results, the transition in our collective mindset was swift and complete. This readiness to adopt machine standards as our own gives me pause when considering how similar inversions might unfold in other domains.
The Price of Perfection
This reversal has profoundly transformed chess at every level. Where chess education once emphasized principles and general understanding, it now often focuses on memorizing specific lines verified by engines. Opening preparation, once a matter of developing playable positions with long-term strategic plans, has become an arms race of computer-checked novelties buried 25 moves deep.
What I find most telling is how the very aesthetic of chess has transformed. Commentators and fans increasingly judge games not by their drama, creativity, or even their outcome, but by their "accuracy percentage"—how closely the players tracked the engine's preferred moves. A messy, fighting game with mistakes on both sides might produce an exciting spectacle and a decisive result, but it will be rated lower than a perfectly played draw where both players navigate a narrow path of engine-approved moves to an inevitable half point.
This shift creates a paradoxical tension that I've observed with growing interest. When asked why he declined to defend his world championship title in 2021, Magnus Carlsen expressed his opinion that players are allowed too much time between moves. He said that players should be given less time to review their computer-driven preparation, forcing them to act more creatively and instinctively. Since then he has been one of the leading advocates for a variation of chess called Chess960. This variation of the game features randomized starting positions, effectively nullifying the computational power of modern chess engines. I find it noteworthy that the greatest (human) player in the world essentially argued that chess had become too perfect, too close to the machine ideal, and therefore less interesting as a human contest.
“The thrill used to be about using your mind creatively and working out unique and difficult solutions to strategical problems,” stated chess grandmaster Wesley So in a 2022 interview. “Not testing each other to see who has the better memorization plan.” His comment resonates with me as a reflection of how the stockfish swap has altered the very nature of chess mastery.
The impact extends beyond the elite. Club players now routinely analyze their games with engines, often focusing more on where they deviated from computer suggestions than on understanding the underlying patterns or developing their own style. Even I, a distinctly amateur player, have found myself checking my games against Stockfish, feeling a curious mixture of pride and disappointment based not only on whether I won or lost, but also on how closely my play matched the engine's recommendations.
The result is a generation of players who may be technically stronger than their predecessors but whose relationship with the game is fundamentally different: more science than art, more replication than creation. I wonder sometimes what has been gained and lost in this transformation. The average club player today almost certainly makes more "accurate" moves than players of similar ratings fifty years ago, yet I question whether they experience the same creative fulfillment, the same sense of personal expression through their play.
There's a certain irony in how chess has evolved. A game once celebrated for its endless creativity—with more possible positions than atoms in the observable universe—has in some ways narrowed. At the highest levels, certain openings have been analyzed so extensively by engines that they're considered virtually solved, leading to a reduction in diversity of play rather than an expansion. The stockfish swap hasn't just changed how we evaluate chess; it has changed chess itself.
Beyond the Chessboard
What interests me most about the stockfish swap is not its impact on chess alone, but what it might tell us about transformations likely to unfold across numerous domains as AI capabilities continue to advance. I see early signs of similar swaps in other fields traditionally considered bastions of human creativity.
Consider writing and language. Grammar checkers like Grammarly began as tools that attempted to approximate human editorial judgment, catching basic errors and suggesting improvements. Their capabilities were measured against human editors—how many errors could they catch? How close were their suggestions to what a human editor might recommend? But as large language models have become more sophisticated, I've noticed the beginning of a directional shift. Companies now market how their AI writing assistants can make prose more "clear," "engaging," or "effective," implicitly suggesting standards that transcend individual human style. Rather than simply asking "Does this AI write like a human?" we're beginning to ask "Does this human write as clearly, concisely, or effectively as an AI could?" The standard is inverting, with human prose increasingly compared to machine-generated alternatives.
This shift becomes even more apparent in professional contexts. I've observed how business writing increasingly conforms to AI-friendly patterns—shorter paragraphs, clearer structure, more standardized vocabulary—not because humans naturally write this way, but because these patterns align with what algorithms reward and recommend. Much as chess players have adapted their style to match engine evaluations, writers are beginning to adapt their prose to match algorithmic preferences and AI outputs.
In visual art, we're witnessing a similar transition, albeit at an earlier stage. Early AI art generators were evaluated based on how convincingly they could mimic human artistic styles. But with tools like DALL-E, Midjourney, and Stable Diffusion producing images of stunning quality and originality, the comparison is beginning to flip. It won't surprise me if before long human artists find their work evaluated against AI capabilities: "Could an AI have produced this more efficiently?" "Does this human artwork display the technical precision or imaginative range available through prompting?"
I've been particularly struck by conversations in design communities, where clients increasingly reference AI-generated concepts as standards for human designers to meet or exceed. "I generated this with Midjourney. Can you create something similar but better?" The human designer is no longer defining the aesthetic standard but is asked to refine or elaborate on machine-generated foundations. This echoes how chess players now often start with engine recommendations and add their human touch within the boundaries of machine-approved variations.
Even in domains like programming, where AI tools like GitHub Copilot were initially judged by how well they could predict human coding patterns, I see signs of inversion. The question is shifting from "Can AI code like a proficient human programmer?" to "Can human programmers effectively leverage AI to achieve optimal results?" Programmers are increasingly measured not only by their raw coding ability but by their skill in prompting and directing AI tools, becoming orchestrators rather than creators, with the AI establishing the standard of technical implementation.
What I find most thought-provoking about these early inversions is how they mirror the pattern I've observed in chess. They begin subtly, with changes in language and framing rather than explicit declarations of new standards. They accelerate as AI capabilities improve, creating a tipping point where the direction of comparison naturally reverses. And they often occur without resistance or even awareness from many practitioners in the field, who adapt to the new paradigm as pragmatically as chess players adapted to engine analysis.
This pattern suggests to me that we're not just having practical recalibrations, but philosophical ones as well. The question is no longer whether machines can do what humans do, but whether humans can adapt to a world where machines increasingly define the standard of performance. This is the defining feature, the tipping point of the stockfish swap: not the moment when machines are able to outperform humans in a domain, but the subsequent reorientation of human activity around machine-defined ideals.
Narrowing Paths to Excellence
As I've reflected on the stockfish swap and its implications, I've been increasingly drawn to consider how it affects diversity of approach and outcome. In chess, one of the most striking consequences of engine dominance has been a narrowing of viable playing styles at the highest levels. Where once the chess world featured players with radically different approaches—the solid positional player versus the tactical wizard, the aggressive attacker versus the subtle defender—top-level chess has become more homogenized.
This standardization occurs because engines effectively establish which approaches are objectively superior, regardless of human aesthetic preferences or psychological comfort. If the engine consistently evaluates a certain type of position as better, players naturally gravitate toward those positions, even if they might previously have avoided them based on stylistic preference.
I see parallels emerging in other creative domains. As AI tools begin to establish what constitutes "effective" writing, "compelling" visual design, or "engaging" music, I wonder if we might witness a similar narrowing of creative approaches. When tools like Grammarly or ChatGPT suggest specific sentence structures or word choices based on algorithmic assessments of clarity or engagement, they subtly push writers toward certain stylistic norms. When design tools optimize for user attention or conversion rates, they encourage visual approaches that satisfy algorithmic metrics rather than divergent aesthetic visions.
“Life has to be somewhat opaque to be bearable.”
—Bernard Williams
What concerns me about this potential standardization isn't that the AI-influenced standards are inherently bad—they may well represent genuine improvements in functional effectiveness—but that they may reduce the glorious messiness of human creativity, the willingness to pursue approaches that seem suboptimal by current metrics but might lead to revolutionary breakthroughs or simply express different values.
I think of writers like James Joyce or Cormac McCarthy, whose work deliberately violates conventional rules of clarity and structure to achieve unique effects. I think of architects like Antoni Gaudí, whose organic, seemingly impractical designs defied the engineering efficiency standards of his day. I wonder if such revolutionary creators would find themselves increasingly constrained in a world where AI tools establish and reinforce standards of "good" creative work.
In chess, we've already seen how engines have rendered certain openings and approaches obsolete; not because they cannot succeed in practice against human opponents, but because they fail to meet the new standard of objective correctness established by computational analysis. I worry that similar narrowing might occur across creative fields, with approaches that don't satisfy algorithmic metrics of success being gradually abandoned, regardless of their subjective value or potential for innovation.
Other posts you should check out
Remembering Chess's Creative Heritage
Those who would dismiss these concerns might argue that chess is fundamentally different from truly creative endeavors; that it's a closed system with finite variables, more amenable to computation than arts requiring genuine creativity or emotional expression. Chess, they might say, was always destined to become the province of machines, while human creativity will remain sovereign in domains like literature, music, or visual art.
This argument, however, reveals what I see as a historical blindness, one that carries important warnings for other fields. Chess was not always perceived as a computational problem. For most of its 1500-year history, chess was considered an art form, an expression of human creativity, psychology, and aesthetic sensibility.
Consider Mikhail Tal, world champion in 1960-61, celebrated not for computational accuracy but for bewildering sacrifices and psychological warfare over the board. His contemporary, David Bronstein, explicitly rejected the idea of chess as calculation, writing: "Chess is not a science, or an art—it lies somewhere in the interim zone. Chess can and should be aesthetic... an expression of harmony and beauty."
From the romantic chess of the 19th century to the hypermodern revolution of the early 20th, chess was a battlefield of ideas and aesthetic visions rather than a search for objectively perfect moves. Even into the 1990s, distinct styles flourished. Kasparov's aggressive dynamism contrasted with Karpov's subtle positional mastery, both valid expressions of chess artistry rather than deviations from some computed ideal.
It was only with the rise of engines that chess transformed in our perception from an art to a science, from a realm of creative expression to a domain of computational optimization. What we now take for granted—that there is an objectively "best" move in any position, discoverable through sufficient calculation—would have seemed alien to chess philosophers of earlier eras.
This transformation offers a sobering parallel for other creative fields. Just as chess was reframed from art to science as a result of computational advances, other domains we currently consider immune to the stockfish swap may undergo similar transformations. The territory of "true creativity" may continue to shrink not because machines become more human, but because our understanding of human creativity becomes more mechanical.
I wonder if we might look back on early 21st century debates about AI creativity with the same sense of historical curiosity with which we now view pre-computer chess theories. Will future generations find it quaint that we once considered writing, visual art, or music as uniquely human domains rather than computational problems with optimal solutions?
These questions haunt me not because I believe human creativity will disappear, but because I recognize how profoundly our conception of creativity itself can change when machine capabilities establish new standards. The stockfish swap in chess wasn't just about who (or what) plays better; it was about how we understand the game itself. Similar reconceptualizations may await other domains we currently consider safe from computational redefinition.
Human Meaning in a Post-Swap World
As we move forward amidst this shifting landscape, the challenge becomes not how to compete with machines on their terms—a battle humans will increasingly lose across domains—but how to redefine success and meaning in a world where machine capabilities establish new standards of performance.
In chess, I've observed with interest the emergent responses to the stockfish swap. Some tournaments now ban electronic preparation entirely, requiring players to rely solely on their natural memory and understanding. Variants like Chess960, which randomizes the starting position, reduce the advantage of memorized computer analysis and emphasize real-time problem-solving. These approaches don't reject computer assistance entirely, but they create space for distinctly human engagement with the game.
I wonder if similar adaptations might emerge in other fields as they undergo their own versions of the stockfish swap. Perhaps we'll see literary movements that deliberately embrace aspects of writing that algorithms struggle with: ambiguity, cultural context, lived experience, moral complexity. Maybe visual artists will increasingly focus on the embodied, sensory aspects of their work that cannot be fully captured in digital reproduction, no matter how sophisticated.
What I find most promising are approaches that neither reject AI capabilities nor submit entirely to them as the defining standard. The most thoughtful chess players today use engines as tools for exploration rather than as prescriptive authorities, allowing computer analysis to expand their understanding while maintaining their own judgment about which positions suit their strengths and preferences. Similarly, writers might use AI assistants to identify potential weaknesses in their prose while preserving their unique voice and perspective.
The philosopher Bernard Williams once observed that "Life has to be somewhat opaque to be bearable." This resonates deeply with me as I consider the implications of the stockfish swap. Perhaps there's wisdom in maintaining some domain of human activity that resists complete computational transparency. Not because machines can't match or exceed human performance, but because human meaning derives partly from the journey of discovery, the struggle with limitation, and the expression of embodied experience.
I think of the difference between Garry Kasparov's reaction to losing to Deep Blue in 1997 and Lee Sedol's response to being defeated by AlphaGo in 2016. Kasparov, caught in the early stages of the stockfish swap, responded with suspicion and frustration, struggling against the implication that a machine could match human brilliance in chess. Sedol, by contrast, spoke of AlphaGo's play as revealing new beauty and possibility in the ancient game of Go. "I have grown through this experience," he said. His response hints at a more mature relationship with machine intelligence—not as a rival that threatens human significance, but as a different kind of entity that can expand our understanding of domains we thought we knew completely.
This perspective suggests a potential path forward as the stockfish swap unfolds across evermore fields. Perhaps the question isn't whether humans can match machine performance, but how machine capabilities can expand human understanding and creativity. Perhaps the measure of success isn't how closely we approximate computational ideals, but how effectively we integrate computational insights into distinctly human approaches that reflect our embodied experience, emotional resonance, and conscious values.
As AI capabilities continue to expand across domains, I believe we'll need to become more thoughtful about both what we value in human activity and why. The stockfish swap that has played out in chess offers both a warning and an invitation: a warning about how quickly and unquestioningly we can adopt machine standards as our own, and an invitation to articulate more clearly what makes human creativity meaningful beyond technical performance.
The stockfish swap reveals how complicated our relationship with technology can be: we create tools to extend our capabilities, only to find ourselves measured by the capabilities of our tools. As this pattern repeats across more and more domains once considered uniquely human, we face not just practical adjustments but existential questions about the source of meaning in human activity.
The ultimate lesson from chess is not that we should try to play like computers, but that we should remember why we play at all—for the joy of the struggle, the beauty of the ideas, and the distinctly human experience of reaching beyond our limitations, even if imperfectly. In a world increasingly defined by machine standards, the most revolutionary act may be to value the human element, mistakes and all. Not because it's superior in technical performance, but because it's the expression of conscious beings making meaning where none existed before.
The challenge for us is to navigate these transitions with awareness and intention, shaping rather than merely reacting to the new standards that emerge as AI capabilities continue to expand.
Thank you for reading. If you enjoyed this essay, please consider sharing it or subscribing.
References and Further Reading
Quotes and References
Claude Shannon’s original paper has been preserved online. It’s interesting to see how close it came in some regards, as well as the innovations he couldn’t have imagined.
The Elo rating system used in chess adjusts each player’s score based on their respective strength and how likely one is to beat the other.
Stockfish’s analysis of world championship games can be a bit dense if you aren’t familiar with chess, but is worth the effort required.
Magnus Carlsen has spoken on multiple occasions on reducing time allotments during championship games. The over-reliance on computers among leading players is frequently cited as the reason he has stepped back from the game.
Carlsen has advocated for Chess 960 quite vocally on numerous occasions, offering it as a potential solution to the hyper-optimized game that chess has become, a return to human creativity and ingenuity.
Wesley So’s quote about chess shifting from creativity to memorization comes from a 2022 interview he gave in The Atlantic.
David Bronstein's quote about chess as an expression of harmony and beauty appears in his book The Sorcerer's Apprentice, co-authored with Tom Fürstenberg, which explores his philosophy of chess as more art than science.
Bernard Williams' observation about opacity appears in his philosophical writings on ethical life, particularly in Ethics and the Limits of Philosophy.
Lee Sedol's comment about growing through his experience with AlphaGo was made in press conferences following the historic match in 2016.
Additional Perspectives
Those interested in the technical evolution of chess engines, might check out Matthew Sadler and Natasha Regan's "Game Changer: AlphaZero's Groundbreaking Chess Strategies and the Promise of AI," which explores how neural network approaches revolutionized computer chess beyond traditional calculation-based engines. I haven’t read this yet, but it came highly recommended to me and I plan on getting to it in the next couple months.
The Turk: The Life and Times of the Famous Eighteenth-Century Chess-Playing Machine, by Tom Standage is a fascinating history of the ingenious and eponymous hoax. Even knowing how it would turn out, this book kept me engaged right to the end.
Garry Kasparov's "Deep Thinking: Where Machine Intelligence Ends and Human Creativity Begins" offers fascinating perspective from the first world champion to lose to a computer, including his evolving thoughts on human-machine collaboration.
For a broader exploration of how AI is reshaping creative fields, Brian Christian's "The Alignment Problem" provides thoughtful analysis of the challenges of aligning AI systems with human values and intentions across domains.
Douglas Hofstadter's "Gödel, Escher, Bach: An Eternal Golden Braid," though written before the current AI revolution, remains profoundly relevant in its exploration of meaning, pattern, and the essence of human thought versus formal systems.
Those interested in the philosophical dimensions of AI and creativity might enjoy Margaret Boden's "AI: Its Nature and Future," which explores the question of whether machines can truly be creative and how we might understand creativity itself in computational terms.
Demis Hassabis and David Silver's paper "Mastering the Game of Go without Human Knowledge" documents how AlphaGo Zero developed superhuman capabilities without human examples.
Hannah Fry's "Hello World: Being Human in the Age of Algorithms" offers accessible analysis of how algorithmic systems are changing various domains of human activity and what this means for our future.
This chess observation you made in this piece is wild, the reversal where now humans are judged by how closely they match machine moves is just really craaaaazy.
It’s such a great point to raise right now. We really should dig deeper into understanding our human traits and explore the limits we should set in our human-machine collaboration.