Paper T5 — Why Play When AI Can Win: The Non-Transferability of Participation and the Visibility of Dignity

Preprint version: 3
Canonical DOI: https://doi.org/10.31235/osf.io/593ah_v3

Kengo Tomita
Institute of Technology, Shimizu Corporation, Tokyo, Japan

Abstract

Knowledge work is learning to hand itself over. Drafting, analysis, computation, and even the role of a practice opponent can now be delegated to artificial intelligence, often with better and faster results. This paper asks what does not move in that handover. Drawing on Bernard Suits's analysis of games, I distinguish an achieved state from participation in the activity that produces it and defend an agent-relative non-transferability theorem: another agent may realise the goal state on my behalf, but cannot thereby make it true that I pursued that goal under constitutive constraints I accepted as reasons. AI can deliver the state of affairs; it cannot deliver the fact of participation. The theorem is independent of whether machines can achieve, play, or matter morally. It leaves a human question: under what cultural conditions can people continue activities they know machines can perform better? I use an explicitly extended Kantian contrast between price and dignity to propose that AI, by supplying functionally equivalent outputs at scale, makes newly visible the non-equivalent fact of participation. Japanese professional shogi, which publicly encountered machine superiority and continued, is examined as a completed historical case in which recognition for serious play persisted after technology made its instrumental outputs substitutable. The case illustrates a condition analysis centred on dignity apparatuses: cultural forms that prevent appraisal by comparative performance from exhausting recognition of participants' standing.

Keywords

play, games, Bernard Suits, artificial intelligence, delegation, achievement gap, meaningful work, shogi, dignity, recognition respect

1. Introduction

In November 2019, Lee Sedol retired from professional Go. He was the only human ever to win a formal game against AlphaGo, and when he explained the decision he did not appeal to age or fatigue: he told Yonhap News that even if he became the number one player, there was an entity that could not be defeated (Vincent 2019). The premise of the remark was and remains correct. The machine was stronger, and would stay stronger.

Seven years earlier, in January 2012, Japanese shogi had faced the same arrival, and its response ran in the opposite direction. Yonenaga Kunio — sixty-eight years old, long retired from active competition, and serving as president of the Japan Shogi Association — sat down against the program Bonkras in a public match that he had himself designed, and lost in 113 moves. Within a month he published a book whose title translates as I Was Defeated, containing his own analysis of the game, the full transcript of his press conference, and the reflections of the program's developers (Yonenaga 2012). He died before the year was out. The event he created became an annual series in which professionals continued to face machines until the question of superiority was no longer open; and when it was no longer open, the professionals went on playing, the public went on watching, and the game's public stature in Japan did not collapse — by several ordinary measures it flourished (the evidence, and the causal hedges it requires, are given in Section 5). Two of the deepest board-game cultures on earth met the same event — the arrival of superhuman performance at their games — within a single decade, and produced two different responses. In one, one of the game's greatest players concluded that something essential was gone and left the board. In the other, an institution absorbed the defeat, reorganised around it, and stayed. The units are deliberately asymmetric — one player's exit against one institution's absorption — and the Go world itself did not follow its champion out the door: professional play continued, and by one population-level measure improved (Section 5 returns to the evidence). The contrast this paper needs is therefore not between the fates of two cultures but between two responses that the same event made available.

What did the machine take from Lee Sedol that it failed to take from Japanese shogi? The vocabulary in which the question is usually discussed — jobs, skills, productivity, achievement, meaning — does not separate the cases cleanly, because along most of those axes they are nearly identical: in both, the machine became decisively stronger; in both, human competitive supremacy ended, permanently and in public. What differed is something that vocabulary does not name, and this paper is an attempt to name it. Because the thing to be named is a property of play rather than a property of machines, the central claim of the paper is not a claim about AI. AI functions here as a developer functions in photography: it has brought out, sharply and at scale, a structure that was present all along. Play is chosen here not as the paper's scope but as its purest specimen: the activity in which the participant-relative, non-substitutable dimension of participation is least obscured by additional ends (Section 4.4 says why its recognition matters). The structure is this: the value of participating in a game is agent-relative and does not transfer. Parts of the activity can be handed over — preparation, analysis, record-keeping, even the role of one's opponent — but no one, and nothing, can take over one's participation itself, because "participating in my place" does not pick out a well-defined service awaiting a capable enough performer; it does not pick out a well-defined service at all.

The question has pressed itself on public attention in two waves. The first came in 2016–2018, when machine superiority arrived in the games themselves — most visibly in Go — and a public discourse arose that sought to protect a human remainder by distinguishing games, at which machines now excelled, from play, creativity and enjoyment, which were declared essentially human. That discourse conceded more than it needed to: it treated machine victory in games as machine entry into play, a misrecognition for which Sections 2 and 3 supply the diagnosis, using a concept that is Suits's own. The second wave began around 2024, when generative systems started absorbing the domain of work, and the question returned from the opposite side: if machines can do the work, what remains is — hopefully on some accounts, despairingly on others — play. The present academic debates belong to this second wave. Beneath both waves, Section 4 will argue, lies a third and older layer, named by Suits before either wave existed: the suspicion that most people cannot regard a life without necessity as worth living — so that the deepest resistance to AI may be a defence not of employment but of meaning.

Within the second wave, the literature has consolidated along two lines. The first concerns achievement. Danaher and Nyholm (2021) argue that widespread AI threatens an "achievement gap": on their analysis, an achievement requires a valuable product, a non-lucky causal contribution, costly involvement and voluntariness, and automation erodes the middle conditions at scale. The ensuing debate has refined and contested the diagnosis: Tigard (2021) doubts that the gap is as damaging as it appears; Scripter (2022, 2025, 2026) sorts gaps from swaps and maps the adjacent losses; Karlan (2023) takes Lee Sedol's retirement as the debate's anchoring case, and argues that little in the technology itself warrants the despair. At the debate's far end stands the question whether machines themselves achieve: Kieval (2024) argues that a system like AlphaGo can satisfy the conditions of achievement, while Vacek (2025) objects that achievement is bound to creditworthiness and responsibility. I return to this exchange in Section 3, where the theorem defended here proves independent of its outcome.

The second line concerns whether games can carry the meaning that work has carried. Its centre is Danaher's Automation and Utopia (2019a), which welcomes the automation of work and defends a ludic utopia — a post-work future organised around games — as the best future available to us (the essay-length version of the defence is Danaher 2019b). The critics press from two sides. O'Brien (2025) argues that where AI can do everything better and faster, self-directed work collapses into game-playing whether or not one consents, and counts the transformation a serious loss — the despairing formulation of this paper's title question. Knell and Rüther (2024) argue that a life of games is an insufficient substitute for meaning, which requires goods — care among them — that games do not supply. Adjacent to both stands the literature on meaningful work, which asks what makes work worth doing and what automation would remove (Smids, Nyholm and Berkers 2020; Parmer 2024; Nyholm and Rüther 2023). The deep engagements come later: O'Brien, and Knell and Rüther, in Section 4.

This paper stands at the intersection of the two lines and places there a claim that neither has stated. Against the first line, the claim concerns participation rather than achievement: I concede — indeed insist — that achievement, as that debate defines it, is genuinely damaged by AI, and argue that the value of participation was never on that axis. Against the second, the claim concerns non-transfer rather than substitution: I do not argue that games can replace work as a source of meaning, and nothing in what follows requires them to win any such competition; I argue that one specific value was never transferable in the first place, and that the operative question — the one on which the two responses above diverged — is under what cultural conditions people can accept this. The contribution is accordingly threefold: an agent-relative theorem of non-transferability (Section 3); a condition analysis of its acceptance (Section 4); a completed historical case (Section 5). Within the condition analysis, Section 4.4 proposes that AI's production of functional equivalents makes newly visible the distinction between exchangeable outcomes and non-equivalent participation. It is not a general account of meaning-loss after AI, not a further entry in the ledger of what AI takes away, and not a position on whether machines can achieve, play, or matter morally.

The claim also has a genealogy, and an interrupted one. Bernard Suits opened the underlying question in 1967, in a paper asking whether life is a game we are playing, and laid out the fork along which everything since has travelled: if a life can have the structure of a game, what follows for the one who comes to see it so (Suits 1967)? The Grasshopper recovered a rigorous definition of game-playing against Wittgensteinian scepticism about definition and pressed the life-as-game thesis to its utopian limit (Suits [1978] 2014).¹ The negative branch of the fork then received its sharpest formulation not in English but in Japanese: Kawatani (2012) argued that absorbed play and the lucid recognition of a game as a game are close to incompatible — the one who fully awakens steps down from the board — and that a culture in which the recognition spreads therefore faces the collapse of its games. Kawatani died young, and his argument has, to my knowledge, not previously been engaged in the Anglophone literature. The present paper brings it in, gives its author a seat at the table rather than a line in a bibliography, and answers him. What the genealogy leaves vacant is the positive branch: an account of the conditions under which one can see the game as a game and keep playing — awake, and in earnest. Supplying that account is the work of this paper.

Why was that account never written — why, indeed, was the theorem at the centre of this paper never stated, in a literature on play that runs from antiquity to the present? Because for almost all of that history it had no work to do. A property becomes visible when it begins to bear load. So long as no one could delegate intellectual participation — so long as there was no agent to whom one's thinking, one's moves, one's working-out could be handed — the non-transferability of participation was a wall against which nothing ever pressed. What cannot be attempted does not need to be shown impossible. Machine delegation made the attempt real, cheap and general; and a distinction that had slept inside the grammar of play for twenty-five centuries became, within a few years, load-bearing.

Suits, once more, came closest — not by stating the theorem but by running the scenario. The Grasshopper opens with an argument that the ant's virtue is self-undermining: prudence, the logic of work, aims at the elimination of sacrifice, and a logic that aims at the elimination of sacrifice aims at its own retirement. Read now, the argument acquires a referent its author did not live to see: AI is the completion device of industriousness — diligence engineering its own unnecessity — so that the ant's ethic flows, by its own internal logic, into the grasshopper's question. And in the book's final chapter Suits executed, in 1978, the scenario we now discuss under the name of AI: a utopia in which all instrumental activity has been mechanised, in which the traditional occupations of a human life — inquiry, morality, art, mutual aid — have run out of work to do, in which even the arts of pure form could be assigned to computing machines, and in which the only candidate left for filling a life is the playing of games. His Seekers, equipped with a memory bank from which any answer can be retrieved and choosing instead to work the answers out for themselves, read today less like allegory than like a description of an ordinary weekday desk, with a model open in the adjacent window. This paper does not apply Suits to artificial intelligence. It observes that artificial intelligence has caught up with Suits, and asks what the experiment, now running at scale, shows that his text did not say.

This is also why the question has stopped being a philosopher's luxury. Kawatani's problem turned on a recognition — seeing the activity one lives inside as a game — that historically arrived rarely, privately, one person at a time. AI is a forced mass distribution of that recognition: every time a system delivers, in seconds, the goal state of an activity that someone had made the substance of a working life, it shows that person what Kawatani's awakened player saw — that the outcome does not need them. Lee Sedol and Yonenaga Kunio are the two branches of Kawatani's fork, lived out in public within a single decade: one stepped down; the other, knowing everything there was to know about the machine's strength, played deeper. On the argument of this paper, the availability of the second response is neither a psychological accident nor a cultural essence: it depends on specifiable conditions in how a community treats those who play. Making those conditions specifiable is what the remaining sections do.

The paper proceeds as follows. Section 2 introduces the Suitsian apparatus and separates two things that the current debates run together: the achieved state of an activity and the value of participating in it. Section 3 states the non-transferability theorem, specifies the criterion it runs on, and shows its independence from every live question about the status of AI, including the question of whether machines might one day play. Section 4 turns from the theorem to its reception: it takes Suits's own vision of collapse as the strongest objection, engages the critics of the ludic future, develops the condition analysis — dignity apparatuses — that an affirmative answer to the title question requires, and relates the apparatus (Section 4.4) to the distinction between price and dignity. Section 5 presents professional shogi as a completed historical case: a domain that staged its own public encounter with machine superiority and kept its play. A short postscript closes the paper.

2. The Separation: Achieved State and Participation

2.1 The apparatus

Two traditions have tried to say what play is. The older one defines it by its autotelic character: play is activity whose end is internal to it, done for its own sake and not for a product beyond it. The younger one, culminating in Suits, defines a narrower object — the playing of games — by its structure. This paper runs on the second definition, and borrows exactly one thing from the first at the point where the two meet.

Suits's analysis has four elements (Suits [1978] 2014). A game has, first, a prelusory goal: a state of affairs that can be described, and in principle brought about, independently of the game — a ball resting in a hole, a person across a line before the others. It has, second, lusory means: the means the rules permit for bringing that state about. Third, constitutive rules: rules whose characteristic work is to prohibit the more efficient means in favour of less efficient ones — one may not place the ball in the hole by hand, or cut across the infield to the finish. Fourth, the lusory attitude: the acceptance of the constitutive rules just because they make possible the activity they constitute. In the portable version Suits offers for carrying around, to play a game is voluntarily to attempt to overcome unnecessary obstacles.

The fourth element is the hinge on which this paper turns, and it is worth pausing on what it does. The older tradition located the autotelic character of play in an inner quality — a spirit, a mood, a way of being absorbed. The lusory attitude relocates it in a structure of reasons: what makes an activity a case of game-playing is that the constraints are accepted because they make the activity possible — that the prohibition of the efficient route functions, for the agent, as a reason rather than as an obstruction. Nothing in this requires inspecting anyone's inner life; it is a matter of what stands as a reason for what. The importance of that relocation for an argument conducted in the age of artificial agents will become clear in Section 3.

One more piece of housekeeping, borrowed from Suits himself. In the book's fourteenth chapter Suits marks his own dichotomy of work and play as stipulative — a proposal about the use of words, not a discovery about their essence. I follow the precedent one level up. "Play" in the broad sense is used stipulatively throughout this paper; no real definition of play-in-general is offered, and no position is taken in the long dispute over whether one exists. The game apparatus, by contrast, is used descriptively: it captures a real structure in a delimited class of activities, and the argument runs entirely on that class. Nothing below requires that all play be games, that all valuable activity be play, or that the word "play" name one thing. Wittgenstein's family-resemblance challenge, to which Suits's definition was itself the answer for the narrow term, is left standing for the broad one (Wittgenstein 1953). (For the contemporary reception of the Suitsian apparatus in the philosophy of games, see Nguyen 2020.)

2.2 What delegation delivers, and what it dissolves

With the apparatus in hand, delegation can be defined without leaving Suits's vocabulary. In the instrumental form relevant here, delegation assigns the realisation of a describable state to another agent in order to reduce the delegator's own expenditure: one hands the task to whatever agent — employee, contractor, machine — will bring the state about at the least cost in one's own scarce resources. Delegation undertaken for other ends — to teach, to distribute work fairly, to place a matter with those it properly concerns — is not this paper's subject. Instrumental delegation is an efficiency operation, and its object is always a describable state of the world: the report written, the position analysed, the ball, as it were, in the hole.

Game-playing, in the same vocabulary, is the voluntary prohibition of the most efficient means, together with the undertaking of that prohibition as one's reason for acting. The two operations are not merely different; they are opposed operations on the same variable. Delegation minimises precisely the inefficiency that play institutes.

It follows that complete delegation of the realisation of the goal state, on my behalf, removes my participation from the activity: applied to a game, it delivers the achieved state while dissolving the game. Whatever agent realises the prelusory goal on my behalf gives me the winning position; it cannot give me the having-played, because the having-played was never a feature of the position. It was a feature of a pursuit — mine — conducted under constraints I had accepted as reasons. The position can be shipped. The pursuit is not the kind of thing that ships.

This is the separation the current debates need, and it can now be stated in one breath. Every game presents two distinct objects of desire: (i) an achieved state — the prelusory goal, describable and in principle producible without the game — and (ii) a participation — the agent's pursuit of that state under self-accepted constraint. The first is deliverable, by anyone or anything capable enough. The second is not delivered at all, by anyone, ever; it is enacted. A sentence such as "the machine could play it for me" will turn out, in the next section, to owe its air of sense entirely to sliding between (i) and (ii).

2.3 Two clarifications

Professionals. An objection arises at the definitional stage and, if sound, would wreck Section 5 before it begins: professional players play for money, and an activity pursued for money is not pursued for its own sake, so on this definition professionals are not playing at all. Suits's own reading of "just because" blocks the inference (Suits [1978] 2014, ch. 13). The lusory attitude requires that the constitutive rules be accepted because they make the game possible; it does not require that this be the agent's only reason for being at the board. Further ends — prize money, a livelihood, reputation among one's peers — are compatible with the attitude, provided they are ends attainable only by way of the playing itself. What distinguishes the amateur from the professional is their attitude toward the game — why they are at the board at all. Their attitude toward the rules, which is what the lusory attitude names, is the same. The professional is playing, in the full Suitsian sense, and Section 5 may proceed.

Two kinds of obstacles. Efficiency, in Suits's analysis, is always relative to scarce resources; an obstacle is whatever exacts expenditure of them. This yields a distinction on which much of the current anxiety about AI turns. Some obstacles merely sat there: nobody chose them, they constitute nothing, and they stood between people and outcomes only because time, skill and attention were scarce. Other obstacles were chosen: they are the constitutive constraints of an activity, and their difficulty is not a tax on the activity but its fabric. Generative systems have made the resources behind the first kind cheap, and obstacles of that kind are falling in rows — which is delegation working exactly as intended. The familiar lament that AI is taking the craft away characteristically fuses the two kinds. Where the vanished obstacle was unchosen, the lament mistakes relief for loss. Where the obstacle was chosen, the lament errs in the other direction: nothing has been taken, because a constitutive constraint is not something another agent can remove from an activity — removing it does not improve the activity but replaces it with a different one. What is genuinely new is not the loss of craft but the arrival of a decision: for the first time, one must determine, obstacle by obstacle, which kind one is facing. While delegation was impossible, the question could not arise; now it arises everywhere. The next section states what happens at the second kind of obstacle when delegation is nonetheless attempted.

3. The Non-Transferability Theorem

3.1 Statement and scope

The theorem can now be stated. Let G be a game with prelusory goal p and constitutive rules R. If any agent other than A — a colleague, a contractor, a machine; nothing turns on which — realises p, then p obtains. But the fact that A pursued p under R, with the constraints accepted as reasons, does not thereby obtain; and no further performance by the other agent can make it obtain, because A's having played is not a causal consequence of p that a sufficiently capable agent could also bring about. It is a property of a pursuit, and the pursuit did not occur. Participation, in short, does not transfer. Parts of the activity are delegable — preparation, analysis, record-keeping, the opponent's seat, even collaboration in mid-game — but the participation itself has no possible substitute, because "take over my participation, so that I will have played" does not specify a task at which an agent could succeed or fail.

Following Parfit's terminology (1984), Nagel (1986, pp. 152–153) calls a reason agent-relative when its general form includes an essential reference to the person who has it. The participation-based reason at issue here has that form: its content refers essentially to A's own participation, not merely to the realisation of p. I use agent-relative derivatively for the theorem's non-substitutability: the corresponding participation fact must identify A, because replacing A with B yields a different fact.

The sentence that seems to say otherwise — "the machine could play it for me" — owes its air of sense, as promised, to a slide between the two objects separated in Section 2. Read as "the machine could bring about p for me", it is true, and describes delegation: p arrives, and no game occurs. Read as "the machine could win for me", it is also intelligible, and describes a proxy player: a game occurs, and the winning is the proxy's. Read as "the machine could play in such a way that I will have played", it describes nothing. There is no third task, between the first two, for a better machine to perform.

Notice that the theorem nowhere mentions machines. People did not stop running when engines began to outrun them, and the world's amateurs play on beneath the level of the professionals without their play being pre-empted from above — a point Danaher (2019b) himself registers in defending the ludic life. What AI has changed is not the theorem but its address: for the first time, the offer of delegation arrives at every desk, for every activity, at negligible cost. A truth that never mattered has begun to matter everywhere at once.

One objection should be met where it arises. The theorem is not ontologically surprising. What is new is that this grammatical fact has not been treated as an independent value variable in the current debates over achievement, meaning, and substitution — and Section 1 gave the historical reason: a property that bears no load gets no name. Being obvious and being theoretically dispensable are not the same thing.

3.2 The criterion: undertaking, not provenance

What separates the player from the agent that merely realises p? A tempting first answer is the causal provenance of the goal: the machine's objective was written by its designers, whereas the player's is his own. The answer fails, and it is important to see why. The professional player also receives the winning condition and the rules from outside — from a culture, an institution, teachers, a federation. Nobody invents shogi privately before sitting down to it. If external provenance disqualified, no one would ever have played anything.

The dividing line is not where the goal came from but what happened to it on arrival: whether the goal and its constraints are undertaken by the agent as the agent's own reasons — and, on the side that can be observed, whether the agent is treated by the practice as a participant who has so undertaken them. This yields the formulation on which the rest of the paper runs: AI systems as currently structured are configured to realise goal states; in the use-contexts at issue, they are not treated as participants who undertake those goals and constraints as their own reasons. Provenance is thereby demoted, not discarded: for current systems, the design documents are a convenient observable proxy — the objective function is written down, and no one in the practice treats the system as having taken it up — but the proxy is not the criterion, and the criterion requires no inspection of an inner life on either side. It is, throughout, a fact about reasons and treatment.

Suits supplied the concept this criterion needs, in the thirteenth chapter of a book published long before the systems existed. His quasi-game player achieves the goal state and moves, outwardly, within the rules — Suits imagines a man tearing around a race's course because he must reach and defuse a bomb — while standing to the game not as one who has taken its goal and constraints as his reasons but as one executing an external necessity through it (Suits [1978] 2014, ch. 13). The figure was, in 1978, a philosopher's curiosity. It is now an industrial standard. The contribution claimed here is accordingly modest: not a new category, but the observation that the category has been realised at scale — together with what that realisation forces into view, namely behavioural indiscernibility. At the level of moves, the quasi-game player and the player can be identical. Nothing on the board distinguishes them; everything in the structure of reasons and treatment does. This, retrospectively, is what the first wave of public discourse was missing. Faced in 2016–2018 with machine victory, it could only choose between "the machines now play" and "the games were never the point". The accurate description had been available for four decades: what had arrived was quasi-game play — indistinguishable from play move by move, and distinct from it in everything the moves do not show.

3.3 Independence from the status of AI

Behavioural indiscernibility invites a behaviourist reply, and the strongest version belongs to Danaher (2020): if an entity is behaviourally equivalent to one that merits a given treatment, it merits that treatment — no inspection of inner lives required. The reply to this has two independent stages, and the theorem survives on either alone.

The first stage separates the fronts. Ethical behaviourism answers a question about how an entity is to be treated as a partner in relations: whether it can merit moral consideration, whether it could be a friend, whether it could be a worthy opponent. That is not this paper's question, and nothing here bears against an affirmative answer to it; for the second-person relation, behavioural equivalence may well suffice. This paper's question is whether participation transfers — whether another's playing, however genuine, can amount to my having played. The two positions belong, note, to the same camp: both refuse to ground anything in private inner states. The dispute is intramural, and it dissolves once the questions are held apart: behavioural equivalence is offered as sufficient for treatment; participation turns on the relevant agent's undertaking, for which behavioural equivalence was never even a candidate ground.

The second stage grants the strongest case outright. Suppose a future system is a genuine player, with complete moral status. Then his play is his play. Habu Yoshiharu — by common consent among the strongest shogi players in the game's recorded history — cannot play my game for me either; a capacity that full human status has never conferred on anyone will not be conferred on a machine by its acquiring status. The theorem is agent-relative, and agent-relativity is untouched by every possible outcome of the status debate. The same bracket closes the achievement dispute flagged in Section 1: grant Kieval (2024) entirely — let systems like AlphaGo satisfy the conditions of achievement — and those achievements are the system's; nothing in anyone else's ledger changes. Vacek's (2025) rejoinder, that achievement travels with creditworthiness and responsibility, runs adjacent to the answerability line of this series — the claim that standing to answer does not transfer without legitimate conferral (Tomita 2026a) — and is noted without being needed: even if AI has achievements and play of its own, that does not make my playing done.

One corollary deserves its own sentence, because it certifies the spirit of the whole. Nothing in this argument denies robot-versus-robot shogi; it may well become an attractive genre in its own right. That the affirmation costs the theory nothing is the proof that nothing protectionist is going on: the human seat is kept without demeaning the machine's game, because the seat was never held by superior ability in the first place.

3.4 Two roots and two world-images

The theorem, so defended, has a deeper basis, and stating it pays two debts at once: it connects this paper to the series it closes, and it gives Kawatani (2012) the theoretical seat promised in Section 1.

What remains on the human side of delegation has two independent roots. The first is normative, and was isolated earlier in this series: the standing to answer for a judgement does not transfer without legitimate conferral — the answering seat (Tomita 2026a). The second is grammatical, and is isolated here: delegation presupposes an end external to the activity; where the end is internal, the object "a complete proxy for my participation" is undefined — the playing seat. The roots are independent, and a pair of opposite cases shows it: the executive who signs a consequential decision answers for it but is not playing; the solitary crossword solver is playing and answers to no one. Each seat can be occupied without the other. They connect, all the same, at one working joint: answerability can survive where passage has quietly gone missing — an explanation can be owned after the fact, polished and presented as one's path — and it is summoned verification, the sustained follow-up questioning that Section 5 will show institutionalised in shogi, that peels an owned explanation off an unwalked path. This paper thus does not sit beneath the earlier one; it supplies one reason the earlier one's verification can work.

The deeper basis itself is Kawatani's. His analysis turns on a doubling of world-images. In the first image, players stand side by side in a shared space of comparison; every ranking, every capability claim, every delegation sentence is at home here. In the second, my life is the sole stage on which everything occurs; others appear as characters within it, and there is no co-equal player — so that "a substitute for me" is not a role that is hard to fill but a phrase with nothing to refer to. The two images contradict each other and require each other, and the space we actually live in is their unstable fusion, the third image — whose instability explains, better than any psychology, how the present anxiety and the present fact are true at once: that one is being overtaken is true in the first image; that one's participation cannot be taken is true in the second. And the basis owes nothing to machine capability: the entire first image can be conceded to AI without remainder.

The concession is exactly where the achievement debate lives. The four conditions of Danaher and Nyholm's (2021) analysis — a valuable product, a non-lucky causal contribution, costly involvement, voluntariness — are, without exception, conditions internal to the first image. That is why the concession of Section 1 was structural rather than diplomatic: the achievement gap is real, and real exactly there; participation lives in the second image, where the gap's variables do not apply. Their own observation that arbitrarily added difficulty does not increase praiseworthiness marks the same seam from their side — true within the efficiency norms of the first image, and beside the point in play, where the added difficulty was never for praise. Their anchor case, finally, is Lee Sedol, and a diagnosis anchored in one exit invites a contrasting case; Section 5 supplies it.

Kawatani's analysis also yields the qualification for the limiting game, and with it the first of two locks. The qualification is structural, not biological — nothing in it requires membership in a species. It asks for an existence that knows, through language, that it will die; that does not know when; that could at any moment end the game itself, so that its going on counts as play rather than mere persistence; and that holds the primal conviction — which cannot be checked, only held — that when the named player dies, this I dies with him.² Systems as currently built are copied, checkpointed and restored as routine engineering; they lack the one-time death, and so cannot constitute the primal conviction. They possess language, the epistemic condition of the qualification, without death, its existential condition: the means of knowing are in place, and the thing to be known is absent. That is the first lock, and it belongs to the limiting level only. The second lock is the theorem itself: were an artificial being to satisfy the qualification, his game would be his game. The levels must be kept apart, and the paper claims at each level only what that level's lock secures. In ordinary games, only the second lock is engaged — the opponent's seat is open to machines, as affirmed above. In the limiting case Kawatani analysed — life itself taken as a game — both locks engage.

One element of his analysis is left open here deliberately. In the second image, Kawatani observes, the game's internal goal itself doubles: victory over co-players, and an absolute victory with no one to face (Kawatani 2012; Matsumiya 2023). What absolute victory would consist in remains an open question. One candidate is to locate it in the way a life crystallises as one's own — but that lies beyond the scope of this paper.

3.5 A first-person corollary

The theorem has an operational form that any reader can apply without instruments, and it is worth stating because it translates the lusory attitude into the vocabulary of delegation. Ask, of any activity: if only the result were delivered, would I be satisfied? If yes, the activity is instrumental, and delegation is exactly what delegation is for. If no, the activity is autotelic, and the offer of substitution is not a bargain to be weighed but a category mistake — indeed, a persistent wish that the obstacle be removed would itself take the activity out of the definition. The test reads the constitutive desire, not the momentary one: "not today", said in fatigue, triggers nothing; a settled preference for the result alone does.

What the theorem removes, finally, should be named, because it is the load the rest of the paper carries. The last layer of the present anxiety is not the loss of employment. It is the announced substitutability of the activities a person lives inside — heard, when it lands, as a denial of the person. The theorem removes exactly this. Participation is not endangered property, because it was never property: what cannot be transferred cannot be taken. The fear of being overtaken is real, and remains real, in the first image; the existence it seemed to threaten was never in the first image at all. What remains — and it is the whole remaining difficulty — is that a truth of this kind is not automatically a comfort. Whether it can be lived with, and under what conditions, is the question of the next section.

4. The Conditions of Acceptance

4.1 The vision of collapse and its heirs

The strongest objection to everything so far is Suits's own. In the closing pages of The Grasshopper, having built the utopia in which only games remain, he turns on it. Most people, he suspects, could not regard a life spent on games as worth living; stripped of the belief that their doings are necessary, they would find existence unbearable; and rather than admit that they are playing, they would fabricate the necessity — and, in a detail that reads differently half a century on, they would turn on the computers, with enmity and with law. The prediction is from 1978. Section 1 promised a third layer beneath the two waves of public reaction to AI; this is it. If Suits is right, the deepest resistance to AI was never about employment. It is a defence of meaning, conducted by people who would rather have necessity back than face what its absence reveals. Note what kind of objection this is: it does not touch the theorem. It says, instead, that the theorem's consolation cannot be received — that the collapse is existential, not logical, and that people will destroy the frame rather than live inside it.

Kawatani gave the vision its mechanism, and it is the sharpest on record: to play absorbedly and to recognise the game lucidly as a game are, he argued, close to incompatible; the one who fully awakens finds the game absurd and steps down; and a culture in which the recognition spreads faces the collapse of its games from the inside. Section 1 called AI a forced mass distribution of exactly this recognition. If the mechanism is right, the vision is not a mood but a forecast.

The vision has contemporary heirs, and the nearest stands at this paper's title. O'Brien (2025) asks the title question from the side of despair. Where AI can do everything instantly, he argues, self-directed work becomes the voluntary erection of artificial obstacles — a game by Suits's own definition — and the conversion cannot be refused: "Once work becomes a game, it is impossible not to play" (O'Brien 2025). There is no individual opt-out and, he argues, no communal one either, since the availability of the machine's route converts every refusal of it into a chosen constraint. The sentence independently re-derives, for the AI era, the transcendental structure Kawatani found in life itself, and the present paper takes it as simply correct. The disagreements are four, and each has already been prepared. First, his accounting of the loss runs entirely along the axis of product value — of two otherwise identical strivings, the one yielding valuable outcomes makes for the better life — which is to say, entirely inside the first image; the loss is conceded there in full, and participation was never on that axis. Second, his charge that the post-AI worker's self-imposed constraints amount to an act is answered by a distinction the apparatus already contains — call it lusory lucidity: to disguise a chosen constraint as necessity is indeed to act, and is precisely the fabrication the vision predicted; to undertake the constraint knowing it is chosen is to play. The difference lies not in the moves but in the lucidity of the undertaking. Third, from inside the first image the only visible lever is speed, and his conclusion is accordingly a prima facie case for slowing AI down; this paper adds a second lever, which operates at any speed — the conditions of dignity, below. Fourth, his case leans at one point on machines having no goods of their own, and he concedes in a footnote that sentience would change the calculus (O'Brien 2025, fn. 23); the locks of Section 3 hold after that concession exactly as before it.

Danaher's defence of the ludic future is exposed to the same weather. Meeting the objection of triviality, his account asks of utopian games that they be worth the devotion of a life — and thereby writes the awakened perspective into the definition itself: the utopian plays games known to be games and known to be unnecessary (Danaher 2019b). That is precisely the position Kawatani's mechanism targets, and the defence specifies no mechanism that would hold the lucid player at the board. The claim of this section is that candidate mechanisms exist, have names, and have taken durable cultural forms; the dignity apparatus names what Danaher's utopia leaves institutionally unspecified. His own logic, meanwhile, raises the stakes. Pressing an analogy with automated driving, he entertains the thought that where machines do a thing better, continued human involvement becomes morally questionable. In play, the efficiency norm behind that thought lapses — my slow lap harms no one — and once it lapses, play is the last ground of human activity immune to the argument. If that is where the logic leads, the conditions under which people can bear to stand on that ground stop being a boutique question. His discussion of quasi-games — smithing and woodwork pursued past their necessity — conserves the same demarcation this paper draws, from the other side.

Knell and Rüther (2024), finally, press Danaher from a different direction: a life of games is an insufficient substitute for meaning, which requires among other things care and receptive goods that games do not supply. The criticism lands on the substitution thesis, and this paper does not hold the substitution thesis. Nothing here claims that games can replace work as meaning's source; the claim is that one value — participation — was never transferable, together with an account of the conditions of accepting this. A reader who takes their point in full can take this paper's in full; the two do not compete.

4.2 Diagnosis and the third response

The dropping-out mechanism has a hidden premise, and the premise is false. "Awake" and "absorbed" contradict each other only if what one wakes to is the game's triviality — if recognising the game as a game means recognising it as beneath seriousness. Remove that premise and the contradiction disappears; and the removal is not a philosopher's manoeuvre but an observation. Professional players are awake and absorbed for a living. No one at the top of shogi believes the game is a necessity of nature; everyone at the top of shogi is inside it to the hilt. Lucid absorption is not a paradox. What the dropping-out problem tracks, then, is not a structural impossibility but a conditional one: awakening leads to departure especially where playing seriously costs standing; where the surrounding culture files games under the unserious, so that to see one's own activity as a game is to see oneself demoted. That is not a tragedy of structure. It may instead reflect a deficit of institutions, and such deficits are in principle open to repair.

Suits's vision leaves its people two paths: fabricate necessity, or admit the truth and wither. There is a third — understand the truth correctly, and accept it — and the understanding has predecessors. Schlick (1927/1979) placed the meaning of life on the side of play half a century before Suits: work, in his analysis, is activity whose purpose lies outside it; play is activity that carries its purpose within itself; a life has meaning in the measure that it contains the second kind, and his figure for that measure was youth. The essay stands at one corner of a triangle whose other corners this paper has already visited — Wittgenstein, whose scepticism about the definability of Spiel is what Suits's definition answers, and Suits himself. The Schlick–Wittgenstein edge of the triangle has only recently drawn scholarly attention (Vrahimis 2024); the Schlick–Suits edge, the one that runs through the meaning of life, has to my knowledge not been drawn at all. It is drawn here because the third response needs it: Schlick supplies the established claim that play is not meaning's opposite but its address.

Understanding, however, is not yet acceptance, and the vision's force was never merely intellectual. The condition under which the third response actually operates appears to be undertaking: the vision fits those who regard the question from outside; those who have made the problem their own — who passed through it as pain rather than surveying it as topic — accept the conclusion with something like naturalness.³ Yonenaga, whom Section 5 treats at length, challenged the machine in earnest, lost in public, and accepted what the loss did and did not touch. The author's own research practice supplied the pain in the present case, and the sequence was the same. And in the author's observation, practitioners in unrelated fields have independently arrived at the same acceptance by the same route. These are not data; they are specimens — enough to show that the response exists, not how it distributes.

4.3 Dignity apparatuses

A dignity apparatus is a cultural form that makes autotelic activity socially recognisable as worth taking seriously — that blocks the default downgrade of such activity to "mere" play. The vision treats the unbearability of a game-filled life as a constant of human nature; the existence of such forms supports treating it as at least partly contingent on cultural form. Four candidate mechanisms can be distinguished, each illustrated in a continuing practice or tradition.

An ordering of value. The Analects ranks three relations to a thing — knowing it, loving it, delighting in it — and ranks them upward: the one who delights stands above the one who merely knows (Analects 6.20; Confucius 2003). The ordering is explicit and canonical, and a culture that carries the sentence carries an enduring reason not to file delight under the unserious. Its third term also names what Section 4.2 observed: delight, in the triad, is lucid absorption — being inside the activity with open eyes.

A bridge into working life. An ordering that lived only in classical texts would decorate, not carry. Shibusawa Eiichi, the Meiji industrialist, cited the triad in 1916 as the culmination of shumi, in a chapter that defines shumi as bringing one's own ideals and desires into the work entrusted to one, inside a book addressed to businessmen (Shibusawa 1916). The gesture is the mechanism: the ordering is carried across the boundary into the world of work by someone whose status in that world is beyond question. Suits's utopia, at its limit, contains Shibusawa's figure: the striver, who goes on building houses nobody needs, is shumi pursued past all necessity — the utopia's own answer and Shibusawa's prescription have the same shape.

A role with the highest respect. A culture can attach its most serious honours to those who stake the most on play. Japanese shogi does exactly this: the professional player is, structurally, a person who has bet a life on a game, and the institution's respect scales with the bet rather than with any usefulness. Section 5 examines this candidate mechanism under pressure.

A practice form. The kansōsen — the post-game review in which both players jointly reconstruct the game and answer for their choices — attaches an answering seat directly to the playing seat. A game one answers for is not "mere" anything; the review's weight is an enduring proof that seriousness does not depend on scarcity or necessity. Section 5 returns to it.

The apparatus has a negative edge, and the analysis is incomplete without it. Shame operates only where there is face to lose. An institution that uses shame as a brake against hollow play or unearned claims must first recognise a participant's standing and attach an interactional face to how that standing is occupied. The kansōsen gives both participants a standing from which reasons may be asked and then appraises how that standing is occupied; failure may cost face without cancelling the recognition floor. Distribution and edge are two sides of one form, and the design order follows from the logic itself: not stronger punishment, but the prior issuance of something worth losing.

None of this refutes Suits's scenario, and none of it is offered as refutation. The East Asian materials concretise the form of social recognition that the scenario lacks: they show what it looks like when a culture has already built, and run for generations, the thing whose absence the vision correctly fears.

4.4 Price and dignity: what the apparatus protects

It remains to say, in one vocabulary, what the apparatus protects. In the kingdom of ends, Kant wrote, everything has either a price or a dignity: what has a price can be exchanged for something else as its equivalent; what stands above all price, so that nothing can serve as its equivalent, has dignity (Kant 1785/1996, AA 4:434–435). Kant distinguishes two kinds of price — a market price for what serves a need, and a fancy price for what pleases without serving one, his example being delight in the purposeless play of the mental powers — while dignity, in his immediate claim, is reserved for morality, and for humanity insofar as it is capable of morality. I borrow one formal feature of his distinction — the contrast between exchangeability and non-equivalence — and extend it beyond that original application. This is a constructive use of Kant's vocabulary, not an interpretation of Kant's own account of what has dignity; it needs marking, since neither the absence of a market nor the fact of being play by itself moves an activity out of Kant's price column. For the present argument, I use the ideal-bearing aspect of Shibusawa's shumi (Section 4.3) as an analogue of undertaking, rather than treating the term as mere taste. This is a selection within Shibusawa's usage, not a claim that shumi as a whole maps onto a single Kantian category.

Throughout this section, a functional equivalent means an outcome treated as equivalent relative to a specified task and evaluative criterion. The sentence-substitution test is expository rather than an independent metaphysical criterion: its result depends on the task description and evaluative criterion fixed in advance. Take a sentence describing the putative fact, put another agent in the subject's place, and ask whether the same fact remains. Three answers sort the cases needed here. (i) Where the outcome description intentionally abstracts from provenance, the same functionally specified state may be realised by different agents; representations can be copied and outcomes delegated, while capabilities are acquired or exercised rather than transferred in the same sense. (ii) Institutional authority can be reassigned under rules of succession: a transfer of office, not the production of an equivalent history. (iii) A different fact results, and no operation by any agent can rewrite the subject: the fact of having participated is of this kind. The theorem of Section 3 is a theorem about (iii); the abstract's compression of it into one sentence places the state of affairs under (i) and the fact of participation under (iii). What the theorem establishes is an agent-relative form of non-substitutability: replacing A with B changes the participation fact. Kant's contrast concerns equivalence in value. I treat the former as a reason to invoke the latter vocabulary, not as showing that the two criteria are identical.

Passing from the theorem to this vocabulary is therefore a further step. The theorem establishes that the participation fact is agent-relative and non-substitutable. It does not by itself establish dignity. The further normative claim is that, once an agent's undertaking is treated as reason-bearing — the criterion of Section 3.2 — functionally equivalent outcomes cannot exhaust what is owed in recognising that agent. Non-transferability is agent-relative; standing is relational. The theorem identifies whose participation fact obtains, while a recognition apparatus concerns how that participant is addressed and treated by others. Neither asks what the substituted agent is; the status of AI stays where Section 3.3 left it.

Four words must now be held fixed. Dignity names the basic normative status of persons and of any other beings independently established to possess such status; this paper does not decide whether any artificial system qualifies. Standing names the institutionally recognised position of a participant or office-holder; face, the social value sustained or damaged in interaction; appraisal respect, a graded positive assessment of excellence, closely related to but narrower than ordinary esteem. A dignity apparatus does not confer or revoke Kantian dignity; it makes recognition of a participant's standing socially operative. The distinction between appraisal and recognition is Darwall's (1977). Appraisal respect is a positive assessment of a person or performance against standards of excellence; it may vary in degree and may or may not be merited. Recognition respect consists in giving a relevant fact or standing its proper weight in deliberation and conduct. Equal recognition respect for persons as persons does not vary with excellence; the requirements attached to offices and roles depend on the object in question. Recognition respect for a social office is accordingly a different attitude from respect for the dignity of persons, which Darwall (2006) later analysed as the second-personal standing of equals to make and acknowledge claims and demands within a community of mutual accountability.

The apparatus institutionalises practices of recognition without displacing appraisal: Section 4.3 itself contains appraisal — honour that scales with the stake, a brake against hollow play. The apparatus secures a recognition floor while leaving performance open to appraisal.⁴ Where the practice includes an answering relation, the participant is someone to whom reasons are owed and from whom reasons may be asked; participation as such does not entail such a relation. The quality of moves and performance remains open to appraisal. The distinction between appraisal and recognition fixes the reach of what AI does. AI may alter the comparative value of some capabilities and outputs, and institutions may respond by revising roles and forms of esteem. What does not follow from comparative performance alone is the loss of recognition owed to a person or participant: repricing performance does not by itself determine a person's or participant's standing. It also fixes what Section 4.6 prohibits. To use a performance score as a measure of a person's standing — or to score the authenticity of undertaking, delight, or participation — is to answer a recognition question with an appraisal metric. This does not prohibit scoring outcomes or performances.

I propose face as an interactional medium through which recognition becomes socially legible. Darwall includes the presented self among the objects of recognition respect, citing Goffman's The Presentation of Self in Everyday Life (Darwall 1977, n. 7; Goffman 1959). Goffman's earlier account of face supplies a distinct interactional vocabulary for how such standing is sustained and threatened (Goffman 1955). On the constructive synthesis proposed here, Kant supplies a vocabulary for non-equivalence, Darwall a distinction between appraisal and recognition, and Goffman an interactional account of face; a dignity apparatus operates in the third register and protects the second. The kansōsen is the worked example: by requiring both players to reconstruct the game together, the practice creates a setting in which the loser's good moves can be acknowledged after the result has been settled, and the recognition it sustains attaches to participation rather than to the result alone. Matsumoto (2026) argues that fully realised — or successful — second-personal respect requires mutuality and may require an effort to draw the other party into it. I extend this insight by treating an apparatus as that effort institutionalised — a practice that operates independently of any participant's momentary inclination to sustain it.

This is the place to say why the paper is about play. The formal theorem in Section 3.1 concerns games. Section 3.5 states a first-person corollary that generalises its structure to activities in which the agent's end includes the agent's own performance. Many activities, especially work, combine functionally specified outcome requirements with a participant-relative dimension. Being owed to others does not itself make an outcome substitutable, and delegability must be assessed task by task. Play is methodologically useful not because it lacks markets, audiences, or external obligations, but because participation is conceptually constitutive and cannot be exhausted by delivery of the result. Professional play may still carry income and institutional duties, as Section 2.3 established; play is simply the activity in which participation itself is least obscured by additional ends. That is the specimen the introduction promised. Nothing here places play above other activities; and the corollary's conceptual scope is broader than the practical scope of the application proposed here, which remains limited by the boundary conditions stated in Section 4.6.

Two theses remain; their logical status comes first. The theorem entails neither of the following; I offer them at a different epistemic level. The visibility thesis is an interpretive thesis: where AI makes functionally equivalent outputs newly available at scale, the distinction between outcome-equivalence and participant-relative history becomes salient. The two waves described in Section 1 may be read as one trace of that salience. The apparatus thesis is a sociological hypothesis: where standing has been strongly tied to scarce excellence, the resulting public exposure of one's output as substitutable may be experienced as humiliation unless institutions provide forms of recognition not exhausted by comparative performance. Section 5 stands to both as follows: the historical case demonstrates compatibility and offers a candidate mechanism; it does not establish the apparatus's necessity, sufficiency, or causal effect.

4.5 Two further pillars

Two further observations bear on the vision's empirical weight; neither is a proof, and the first is offered explicitly as an analogue rather than a counterexample.

The first is Wittgenstein. The concept of the language-game gave an entire intellectual culture the recognition that our talk — confessing, promising, testifying, praying — has the structure of rule-constituted games. On the vision's mechanism, that recognition should have lightened everything it touched. It did not. Confession did not become optional; testimony did not lose its weight; no court adjourned on the ground that its proceedings were, after all, a game. The experiment the vision fears — mass recognition of rule-constitution — has in one domain already run, at scale, with the opposite result. The analogue must be handled honestly: language-games are not Suits's games — participation in language is not voluntary, and its constraints are not chosen — so the case refutes nothing. What it suggests is narrower, and enough: the recognition that an activity is constituted by rules does not, by itself, make the activity light. Where seriousness survives the recognition, it was never resting on necessity in the first place.

The second is the ubiquity of local acceptance. The crossword puzzle was born post-utopian: the answer has always existed — in tomorrow's paper, in the dictionary, one telephone call away — and the activity is constituted by the choice not to retrieve it. "Don't tell me the answer" is the characteristic sentence of an established practice, and the practice shows no sign of collapse. Puzzles, gardening, fishing, weekend carpentry: ordinary people accepted autotelic activity in the local regions of life long ago, and the activities did not dissolve on being seen for what they are. What generative AI has done is generalise the seeker's situation — retrievable answers, everywhere — to the whole of knowledge work; it has made nothing new. Two questions carry the observation to a first-person point, and they are diagnostic instruments, not measures. The seeker's question: when the answer can be retrieved, do you retrieve it, or work it out? — retrieval and working-out being different activities, the question locates, for the asker, the line between collaboration and substitution. The striver's question: if AI made this unnecessary, would you still do it? Asked across a life, the two questions show that nearly everyone already answers "work it out" and "yes" somewhere. The vision, in short, may overestimate human fragility. The live problem is not whether local acceptance is possible — examples are widespread — but whether practices of recognition that sustain it locally can be extended into the domain of work. Shibusawa attempted such an extension in 1916; Section 5 examines a case compatible with its institutional realisation under maximal pressure.

4.6 What this account does not license

Two prohibitions bound the account, and they belong to it rather than qualify it. First, the apparatus described here is an analysis of cultural conditions, not a management technique, and its concepts do not license the scoring of standing. To use a performance score as a measure of a person's standing — or to score the authenticity of undertaking, delight, or participation — is to answer a recognition question with an appraisal metric (Section 4.4). This does not prohibit scoring outcomes or performances. The moment an organisation grades individuals on the authenticity of their play — measures delight, audits undertaking — it has rebuilt, inside this paper's vocabulary, exactly the instrumental capture the vocabulary was built to diagnose; the diagnostic questions of Section 4.5 are first-person instruments, and turned outward as metrics they self-destruct. Second, nothing here prescribes play. A culture can make serious play recognisable; it cannot command the undertaking itself. An externally imposed constraint may still be undertaken, but the command alone cannot constitute that uptake, and the framework's own criterion says so.⁵ Whether any particular person should treat any particular activity as play is not a question this paper answers, or takes to be answerable from outside. What Section 5 shows is a culture in which the proposed conditions and continued play coexisted under maximal pressure.

5. A Completed Historical Case

5.1 The case

Section 1 introduced the match; here is the case in full. In January 2012 the president of the Japan Shogi Association sat down against a machine in an event of his own design — he had conceived the Den'ō-sen, named it, and put himself first on its card. Yonenaga was sixty-eight, long retired from active play, and in treatment for cancer; by every measure of self-interest the game meant nothing to him, and he prepared for it as if it meant everything. The preparation is the detail that matters most for this paper: he installed his opponent on a consumer machine and studied the machine with the machine — probing its weaknesses, building an opening specifically against it, and discovering in passing that he needed a physical board under his hands to think in his own way rather than in the machine's. Collaboration ran as deep as collaboration can run, and the seat did not move an inch: the analysis was delegated everywhere, the participation nowhere. He lost in 113 moves. Within a month, as Section 1 recorded, the loss was answered in print, by the loser, under his own name (Yonenaga 2012).

Five years before the match there had been an exchange that gives the case its semantic depth. Recruiting the professional Satō Yasumitsu for an earlier man–machine match, Yonenaga had described it as mere play; Satō had made him sit formally and asked whether play at a professional board could be called mere. The collision is exactly the one this paper's vocabulary resolves: both men were reaching for the deflationary sense of play, and what Satō was defending was the other sense — the full-stake, autotelic sense that Schlick's vocabulary would have let him assert directly: it is because it is play that everything is at stake. Five years later, the man who had been made to sit formally was himself at the board, staking what remained of his strength on a game that could do nothing for him. One life performed the semantic turn this paper has spent four sections arguing for.

5.2 The assay: three axes

What exactly was lost that day, and what was not? The event decomposes along three axes, and the decomposition is an assay executed at the scale of an entire field. On the capability axis the loss was real, total and permanent: the 2012–2017 Den'ō-sen period for shogi, 2016–17 for Go, and from 2022 onward — writing, drawing, analysing — for very nearly everyone. On the play axis there was nothing to lose, because no contest of play occurred there: the machine stood to the game as Section 3's quasi-game player — realising the goal state, unmatched on the board, and not treated, anywhere in the practice surrounding the match, as a participant who had undertaken the game's goal and constraints as its own reasons. On the seat axis there was likewise nothing to lose: the kansōsen's chairs were occupied, that evening and every evening since, by humans answering to humans. The defeat happened on one axis and can be read as rendering the other two newly visible within the practice — value that superior human capability had previously obscured became newly legible when capability fell. That is the sense in which the case functions here as an assay.

Two instruments sharpen the reading. The first is Suits's account of defeat: in a game, losing leaves intact the achievement of having genuinely carried the attempt through (Suits [1978] 2014, ch. 7) — the trying is not annulled by the losing. I Was Defeated is that account with a name on the cover: the title records the capability loss, and the book's existence — an answer, given — records what the loss never touched. The second is the evaluation number. Since engines became authoritative, every broadcast position carries the machine's assessment, and the effect on the human player has been the opposite of erasure: the number works as a counter-stain works in histology, rendering the player's choices visible against the background of the optimal line. Where the professional departs from the engine, and why, and what they say about it afterwards, is now legible to any spectator. The machine settled into the instrument layer; the playing seat and the answering seat remained human, and more visible than before.

5.3 What survived, and the record

Suits distinguishes, within the attitude of contest, the parties' agreement to make a contest of it at all from each party's desire that its own side prevail (Suits [1978] 2014, ch. 7). The kansōsen is anatomically legible in those terms: it is the seat where the agreement to make a contest of it is explicitly reopened, once the question of prevailing has been settled. The two players, one of whom has just lost everything the evening had to offer, sit back down and jointly reconstruct the contest they made together. Nothing in the practice ever measured itself against third-party strength, and so nothing in it had anything to lose to the machine; it continued, untouched, through the years in which the capability question was opened, contested and closed.

The wider record must be stated with the hedges Section 1 promised. The exhibition Yonenaga designed became an annual series; professionals kept entering it until the superiority question was no longer open; and the decade that followed — the decade of Fujii Sōta — saw the professional game's public stature rise rather than fall. No causal claim is made about the rise: flourishing has many parents, a generational star among them, and this paper asserts coexistence, not mechanism. But coexistence is what the vision said was impossible, and coexistence is on the record: total, public, permanent machine superiority, side by side for a decade with a full tournament calendar, a broad audience, and professionals whose cultural stature did not diminish. Adjacent evidence points the same way, with a domain caveat: in Go — not shogi — the quality and novelty of professional human decisions rose after the arrival of superhuman AI (Shin et al. 2023); the finding is population-level, belongs to the capability axis, and is cited as an adjacent anchor rather than extrapolated.

One more feature of the case belongs to Suits. He closed The Grasshopper with an assignment: since the machinery of plenty may arrive before we know what to do with ourselves, begin now to devise the games a life without necessity will need. The Den'ō-sen reads, in retrospect, as an execution of that assignment — an institution designing, before the crisis peaked, the very game in which the crisis would be played out, with its own president going first.

5.4 Conclusion

The case yields a model, and the model has three parts. The full stake: nothing was held back — a retired president in cancer treatment prepared for months and played to win. The public answer: the loss was analysed by the loser, in print, within a month. The redesign of the arena: the encounter was not suffered but staged, serialised, and run to its conclusion — the field controlled the manner of its own defeat, and had rebuilt around the human seats before the last game was lost. The rest of knowledge work arrived at the same board a decade later; the difference is that the manner of sitting there has already been modelled.

The contrast with the leading philosophical proposal can now be drawn in one sentence. What Danaher (2019a) conceived as a utopia of retreat into the virtual, the shogi world partially implemented without retreating from reality: the kansōsen continued not in a virtual world but on the tatami of the shogi hall in Sendagaya, Tokyo. The seats this paper has described were kept in the same rooms, in the same reality in which the machines won — which is what the theorem said from the beginning was possible: the achieved states went over to the machines, and the participation never moved, because it was never movable.

AI can assist play. It cannot take one's place in it.

Postscript

In the spring of 1951, Ludwig Wittgenstein was dying of cancer and knew it (Malcolm 1984). The most obvious external reasons for continuing had fallen away: he had resigned his professorship years before, he would found no school, and the book he had laboured over for decades he had chosen not to publish in his lifetime. He wrote anyway — on certainty, on knowledge, on doubt — and the dated entries run to two days before his death (Wittgenstein 1969). I do not claim to know what the work was to him; a postscript is not the place to smuggle in a biography. But the scene can be read, and I read it so: as work from which everything instrumental had drained away and which continued because the activity itself remained — the third response of Section 4, lived to the end of a life. His last words, by the account of the friend who recorded them, were "Tell them I've had a wonderful life" (Malcolm 1984). That too can be read — not as a verdict on achievements, for there was no one left to outrank and nothing left to win, but as the view from inside delight: the completion report of what Section 3 could only leave open, a life crystallised as one's own.

He had addressed his Investigations, in its preface, to the few the book might reach and stir to thoughts of their own (Wittgenstein 1953), and he once noted that he was writing for friends scattered in the corners of the world (Wittgenstein 1980, 9). Those friends existed then, in the corners of a darkening world, and they exist now. This paper, too, was written in the hope of reaching a few of them.

Acknowledgements

AI systems (Claude, Anthropic; ChatGPT, OpenAI) were used in drafting, source verification, and revision, under the author's direction. The author alone determined the paper's claims and answers for them. Neither system is listed as an author, consistent with COPE and ICMJE guidelines.

References

Confucius. (2003). Analects: With selections from traditional commentaries (E. Slingerland, Trans.). Hackett.

Danaher, J. (2019a). Automation and utopia: Human flourishing in a world without work. Harvard University Press.

Danaher, J. (2019b). In defense of the post-work future: Withdrawal and the ludic life. In M. Cholbi & M. Weber (Eds.), The future of work, technology, and basic income (pp. 99–116). Routledge. https://doi.org/10.4324/9780429455902-8

Danaher, J. (2020). Welcoming robots into the moral circle: A defence of ethical behaviourism. Science and Engineering Ethics, 26(4), 2023–2049. https://doi.org/10.1007/s11948-019-00119-x

Danaher, J., & Nyholm, S. (2021). Automation, work and the achievement gap. AI and Ethics, 1(3), 227–237. https://doi.org/10.1007/s43681-020-00028-x

Darwall, S. (1977). Two kinds of respect. Ethics, 88(1), 36–49. https://doi.org/10.1086/292054

Darwall, S. (2006). The second-person standpoint: Morality, respect, and accountability. Harvard University Press.

Goffman, E. (1955). On face-work: An analysis of ritual elements in social interaction. Psychiatry, 18(3), 213–231. https://doi.org/10.1080/00332747.1955.11023008

Goffman, E. (1959). The presentation of self in everyday life. Doubleday.

Kant, I. (1996). Groundwork of the metaphysics of morals (M. J. Gregor, Trans.). In M. J. Gregor (Ed.), Practical philosophy (pp. 37–108). Cambridge University Press. (Original work published 1785)

Karlan, B. (2023). Human achievement and artificial intelligence. Ethics and Information Technology, 25, Article 40. https://doi.org/10.1007/s10676-023-09713-x

Kawatani, S. (2012). "Jinsei" ga gēmu de aru to iu kanōsei ni tsuite [On the possibility that life is a game]. Hokkai Gakuen Daigaku Gakuen Ronshū, 151, 1–23. https://hokuga.repo.nii.ac.jp/records/2000248 (In Japanese.)

Kieval, P. H. (2024). Artificial achievements. Analysis, 84(1), 32–41. https://doi.org/10.1093/analys/anad052

Knell, S., & Rüther, M. (2024). Artificial intelligence, superefficiency and the end of work: A humanistic perspective on meaning in life. AI and Ethics, 4(2), 363–373. https://doi.org/10.1007/s43681-023-00273-w

Malcolm, N. (1984). Ludwig Wittgenstein: A memoir (2nd ed.). Oxford University Press.

Matsumiya, T. (2023). [Book review: B. Suits, The Grasshopper, Japanese translation]. Journal of the Philosophy of Sport and Physical Education, 45(2), 127–130. https://doi.org/10.9772/jpspe.45.2_127

Matsumoto, D. (2026). Dāworu no nininshōteki sonkei to sōgoteki sonkei ni tsuite [Second-personal and mutual respect in Darwall's philosophy]. Yamagata Daigaku Kiyō (Jinbun Kagaku), 21(1), 1–15. https://doi.org/10.15022/0002001699 (In Japanese.)

Nagel, T. (1979). Moral luck. In Mortal questions (pp. 24–38). Cambridge University Press.

Nagel, T. (1986). The view from nowhere. Oxford University Press.

Nguyen, C. T. (2020). Games: Agency as art. Oxford University Press.

Nyholm, S., & Rüther, M. (2023). Meaning in life in AI ethics—Some trends and perspectives. Philosophy & Technology, 36, Article 20. https://doi.org/10.1007/s13347-023-00620-z

O'Brien, G. (2025). All play and no work? AI and existential unemployment. The Journal of Ethics, 29(4), 747–771. https://doi.org/10.1007/s10892-025-09522-y

Parfit, D. (1984). Reasons and persons. Clarendon Press.

Parmer, W. J. (2024). Meaningful work and achievement in increasingly automated workplaces. The Journal of Ethics, 28(3), 527–551. https://doi.org/10.1007/s10892-023-09434-9

Schlick, M. (1927/1979). On the meaning of life (P. Heath, Trans.). In H. L. Mulder & B. F. B. van de Velde-Schlick (Eds.), Moritz Schlick: Philosophical papers, Vol. II (1925–1936) (pp. 112–129). Dordrecht: D. Reidel.

Scripter, L. (2022). Meaningful lives in an age of artificial intelligence: A reply to Danaher. Science and Engineering Ethics, 28, Article 7. https://doi.org/10.1007/s11948-021-00349-y

Scripter, L. (2025). The achievement gap thesis reconsidered: Artificial intelligence, automation, and meaningful work. AI & Society, 40(1), 89–102. https://doi.org/10.1007/s00146-023-01828-5

Scripter, L. (2026). Is artificial intelligence a threat to meaningful work and living? Technological unemployment and the existential challenges of a transitional era. AI & Society, 41(7), 6555–6572. https://doi.org/10.1007/s00146-026-02941-x

Shibusawa, E. (1916). Rongo to soroban [The Analects and the abacus] (A. Kajiyama, Ed.). Tōadō Shobō. (In Japanese.)

Shin, M., Kim, J., van Opheusden, B., & Griffiths, T. L. (2023). Superhuman artificial intelligence can improve human decision-making by increasing novelty. Proceedings of the National Academy of Sciences, 120(12), e2214840120. https://doi.org/10.1073/pnas.2214840120

Smids, J., Nyholm, S., & Berkers, H. (2020). Robots in the workplace: A threat to—or opportunity for—meaningful work? Philosophy & Technology, 33(3), 503–522. https://doi.org/10.1007/s13347-019-00377-4

Suits, B. (1967). Is life a game we are playing? Ethics, 77(3), 209–213. https://doi.org/10.1086/291634

Suits, B. (2014). The Grasshopper: Games, life and utopia (3rd ed.). Broadview Press. (Original work published 1978)

Suits, B. (2022). Return of the Grasshopper (C. C. Yorke & F. J. López Frías, Eds.). Routledge. https://doi.org/10.4324/9781003262398

Tigard, D. W. (2021). Workplace automation without achievement gaps: A reply to Danaher and Nyholm. AI and Ethics, 1(4), 611–617. https://doi.org/10.1007/s43681-021-00064-1

Tomita, K. (2026a). Decomposing agency, isolating answerability: Cultivating what cannot be delegated in AI-assisted learning. EdArXiv. https://doi.org/10.35542/osf.io/hvbfe_v2

Tomita, K. (2026b). Soshiki ni okeru "jibungoto-ka" no adohiaransu-teki saikijutsu [An adherence-based redescription of jibungoto-ka: What AI-accelerated knowledge work requires of organizations]. SocArXiv. https://doi.org/10.31235/osf.io/495wg_v1 (In Japanese, with English abstract.)

Vacek, D. (2025). Against artificial achievements. Analysis, 85(3), 690–698. https://doi.org/10.1093/analys/anaf034

Vincent, J. (2019, November 27). Former Go champion beaten by DeepMind retires after declaring AI invincible. The Verge. https://www.theverge.com/2019/11/27/20985260/ai-go-alphago-lee-se-dol-retired-deepmind-defeat

Vrahimis, A. (2024). Schlick and Wittgenstein on games and ethics. Philosophical Investigations, 47(1), 76–100. https://doi.org/10.1111/phin.12401

Williams, B. (1981). Moral luck. In Moral luck: Philosophical papers 1973–1980 (pp. 20–39). Cambridge University Press.

Wittgenstein, L. (1953). Philosophical Investigations (G. E. M. Anscombe, Trans.). Blackwell.

Wittgenstein, L. (1969). On certainty (G. E. M. Anscombe & G. H. von Wright, Eds.). Blackwell.

Wittgenstein, L. (1980). Culture and Value (G. H. von Wright & H. Nyman, Eds.; P. Winch, Trans.). Blackwell.

Yonenaga, K. (2012). Ware yaburetari: Konpyūta kisen no subete o kataru [I was defeated: Everything about the computer match]. Chūōkōron-shinsha. (In Japanese.)