Play

Eight essays and a few hundred citations into this project, I have arrived at a conclusion that was available, free of charge, to any sufficiently honest nine-year-old: nothing is in charge. Not the universe, which runs on thermodynamics and does not have a complaints department. Not biology, which is an arms race wearing a nature documentary as a disguise. Not institutions, nor the people who appear to be running them. Order exists locally, temporarily, and wherever someone is paying for it. Everywhere else it’s weather.

I expected this conclusion to feel like a defeat. It felt like a receipt. Fine. Everything is chaos, nobody is coming, and being upset about it has — I checked, repeatedly, under varied conditions — no measurable effect on the chaos. Now what?

The “now what” turned out to be the interesting part. When I made an inventory of what still mattered after the audit, the list was short and faintly embarrassing. Being with the few people I love. A body that isn’t actively betraying me. Understanding something today that I didn’t understand yesterday. And, running through all three, a thing I had spent most of my adult life filing under “break from the point.” The only item that survived the chaos audit was the one I’d been treating as optional.

Then I looked up the scientific definition of play, and discovered that according to it, this collection should not exist.

Gordon Burghardt, the ethologist who spent a career trying to pin the concept down, proposed five criteria in 2005. Play is not fully functional in the context in which it appears. It is voluntary and apparently pleasurable, done for its own sake. It differs structurally from the “serious” version of the same behavior — exaggerated, incomplete, reordered. It is repeated but not stereotyped. And — this is the one that matters — it occurs only when the animal is in what he called a “relaxed field”: fed, safe, healthy, not under stress. A stressed, hungry, or injured animal does not play. Play is the first thing to go when resources drop, like a feature that gets disabled below minimum specs.

I have not been in a relaxed field for most of this year. The reasons are in Fragility, and I won’t repeat them here, partly out of respect for the reader and partly because repeating them would violate at least two of the criteria. Yet here is the ninth essay in a series that nobody commissioned, nobody pays for, and nobody — I have checked the analytics — reads in numbers that would justify the hours. By the definition, this is play. By the definition, it should have been disabled months ago.

So either the definition is wrong, or I am, or there is a kind of play that survives the conditions that are supposed to kill it. I went looking for the science to settle the question. What I found was a definition, several hypotheses, a great deal of philosophy, and one extremely well-studied rat.

The evolutionary puzzle is real, and in one place the numbers are specific. Robert Harcourt spent hundreds of hours watching South American fur seal pups at Punta San Juan, in Peru, and published the accounting in 1991. Over the season, sea lions attacked the colony 102 times and killed 26 pups. Twenty-two of those 26 were playing in the shallow tidal pools at the moment of the attack, apparently oblivious to the other animals fleeing around them. Play took up a small slice of the pups’ day and accounted for eighty-five percent of their deaths. In the Taï forest of Côte d’Ivoire, respiratory outbreaks kill chimpanzee infants in cycles, and when Kuehl and colleagues went through twenty-two years of records in 2008, the cycles tracked the age at which infants play the most — because play is where young chimpanzees touch each other. Observational, not experimental, but the pattern is there. Play costs energy, invites injury, spreads infection, and makes you a worse lookout. And yet it persists in nearly every mammal, in parrots and crows, in octopuses, and — in a controlled experiment published in 2022 — in bumblebees, forty-five of which rolled small wooden balls, some of them over a hundred times, for no reward, when they could have been foraging.

The fur seals taught me something I hadn’t expected from a mortality table. The attacks went on all season. The relaxed field, in the wild, is not a place and not a period. It is the gap between attacks, and the pups played in the gap, because the alternative — waiting for the sea lions to leave — was not on offer. I wrote an entire essay about why nothing is coming. I had not noticed that the animals had already priced this in.

So play must do something. The problem is that nobody has rigorously demonstrated what.

The literature on play is one of those places where confident claims float freely above thin evidence. Burghardt’s criteria are a definition, not a finding. The most cited hypothesis — Špinka, Newberry, and Bekoff’s 2001 proposal that play is “training for the unexpected,” the animal deliberately putting itself into loss-of-control situations so it learns to recover from surprise — is a hypothesis paper. Elegant, consistent with observation, heavily cited, and unproven. Bekoff’s observation that stronger dogs handicap themselves so the game can continue is naturalistic observation, and the word “fairness” he attaches to it is his.

What does exist, and is causal, is the rat.

Heather Bell, Sergio Pellis, and Bryan Kolb published a controlled experiment in 2010 that remains the strongest single piece of evidence in the field. Juvenile rats raised without the opportunity to play with peers develop an abnormal medial prefrontal cortex — and the abnormality runs in the direction you would not guess. Their neurons have longer dendrites. Play, it turns out, prunes. The design even separated play from mere company: the orbitofrontal cortex responded to how many peers were present, whether they played or not, but the medial prefrontal cortex responded specifically to play. And a broader body of rat work, from several labs over three decades, says what the unpruned cortex does in adulthood: play-deprived rats fail to signal submission to a larger opponent, fail to adjust when the situation changes, and get hurt more as a result. Those are deprivation experiments too, which is more than most of this field can say.

This is the entire causal foundation, and it is a foundation about rats. Everything above it — including everything I’m about to say about myself — is inference. I find this oddly comforting. The behavior evolution refused to eliminate is one that science has largely failed to explain. There is something fitting in an activity that resists justification also resisting explanation.

My own field has a formal version of this problem, and it is one of the few places where the mathematics is clean.

An agent learning by trial and error faces a dilemma: exploit what it already knows yields reward, or explore actions whose value is unknown. Pure exploitation locks into the first decent strategy and never finds the better one two moves away. Pure exploration never cashes in. Every reinforcement learning system that works manages this trade-off.

The best explorers, it turns out, look a lot like curiosity. Pathak and colleagues showed in 2017 that an agent rewarded simply for being surprised — for encountering states its internal model failed to predict — would explore a video game level and make real progress through it with no external reward at all. Schmidhuber had formalized the principle decades earlier: reward the agent not for novelty but for learning progress, for being in situations where its model of the world improves fastest. And AlphaGo Zero, in 2017, reached superhuman Go with zero human games in its training — entirely by playing against itself. The best Go player in history learned the game the way a puppy learns to fight: by doing the thing with a partner, repeatedly, without instruction, with no immediate stakes.

I share a conviction with some of the researchers I most respect: that real intelligence comes from interacting with a world and building a model of how it behaves, not from reading about it. If that’s right, then play is not an accessory to learning. It is the mechanism in its purest form — exploration without a task, model-building without a deadline.

Two caveats. First, “curiosity” in these papers is a named reward term, not an experience; the resemblance to a child is structural, not literal. Second — the part that stayed with me — curiosity-driven agents have a documented failure mode, one the field called the noisy-TV problem years before anyone tested it: put a television showing random static in the environment, and an agent rewarded for surprise should be drawn to it, because static is maximally unpredictable and the model never gets better at predicting it. Burda and colleagues built the television in 2018 — a maze, a TV, a button to change the channel — and found that the agent didn’t stare forever, as the thought experiment feared. Its learning slowed drastically. It got to the goal eventually. Much later, and by way of the television.

I am not going to pretend this is a scientific finding about humans, because it isn’t. But I have, this year, under stress, spent more time than I’d like scrolling through feeds that deliver an endless supply of surprising, unpredictable, and completely unlearnable material. The agent got there eventually, by way of the television. So, as far as I can tell, did I. The analogy is mine, not the literature’s. I’m keeping it.

Here is where I expected the strongest evidence and found the weakest.

The theory is beautiful. Piaget cast play as assimilation — the child bending the world to fit the structures already in her head. Vygotsky went further and wrote the line everyone quotes: in play, a child “always behaves beyond his average age, above his daily behavior; in play it is as though he were a head taller than himself.” His example of two sisters, five and seven, who decide to “play sisters” — and who, in the game, dress alike and speak alike and display a sisterhood they never bother to display in real life — is one of the most quietly devastating things ever written about human motivation. Both are theories from the first half of the twentieth century. Both are influential. Neither is “demonstrated.”

When Angeline Lillard and five colleagues reviewed the research on pretend play for Psychological Bulletin in 2013, their conclusion was blunt: no good evidence that pretending makes children more creative, more intelligent, or better at solving problems. Some evidence for language, storytelling, social skills, and self-regulation — but from studies too weak to say whether play causes those things, travels alongside them, or is simply what imaginative children happen to do. Others pushed back; the field has not settled it.

The single meta-analysis in the area — Skene and colleagues, Child Development, 2022, seventeen studies and nearly four thousand children — found that guided play beats direct instruction for early maths skills, with a small effect, and for shape knowledge, with a medium one. No detectable difference for literacy. None for social or emotional development. The funnel plot leaned the way funnel plots lean when a literature has been generous with itself, and thirty-eight of the thirty-nine studies were rated at high risk of bias — which the authors note, fairly, is hard to avoid when you can’t blind a four-year-old to whether she’s playing. So the most rigorous thing anyone can currently say about play and children is that it helps them, modestly, with shapes.

I contribute, lightly, to a friend’s children’s book project. The books are built on the intuition that children learn by playing with stories. I believe the intuition. I also now know that the strongest evidence for it is a rat and a shape-sorting effect size of 0.63. The gap between what we believe about play and what we can prove is roughly the width of a childhood. We proceed anyway. Everyone does. That is either faith or play, and I’m no longer sure the distinction holds.

Gregory Bateson noticed in 1955 that play requires a particular kind of message, and that the message is logically strange. When two dogs play-fight, the nip is not a bite. But the nip stands for a bite — it is a bite-shaped action that does not mean what a bite means. For the game to work, both animals must send and receive a signal that says “this is play,” a frame that denotes something which does not denote what it would otherwise denote. Whatever else it is, it is a structural requirement: play only exists when both parties know the frame.

Adult humans have a form of social play that fits every one of Burghardt’s criteria and that almost nobody files under the heading. Voluntary, exaggerated, not functional in the moment, repeated but never the same twice. And it happens only in a relaxed field — try it while stressed and you will produce something that is recognizably not it. We call it flirting, and I have been, this year, attempting to learn it from the peer-reviewed literature.

I read the attraction research the way I read everything — as if it were a training protocol — and the research, to its credit, had something to say about me. Eastwick, Luchies, Finkel, and Hunt’s 2014 review found that people’s stated ideal-partner preferences predict their reactions to hypothetical partners — profiles, photographs, descriptions — quite well, and their attraction to real people they’ve actually met face-to-face, trait by trait, almost not at all. A 2024 follow-up across forty-three countries and ten thousand people found that matching effects do exist, but small, and smaller than the field had assumed. The list in your head is a list for a person who does not exist. It predicts something. Mostly it predicts the list. I have, as I confessed in Projection, a long history of being attracted to the list.

The honest situation is this: social play requires a relaxed field, I haven’t had one, and the thing I’m trying to learn is the one thing that cannot be learned by study — because the moment you are executing a protocol you have exited the frame. Bateson’s dogs don’t consult a paper mid-nip. I’m doing it anyway, because it’s the only method I have, and because even the analytical approach occasionally, by accident, produces a moment that is actually play. The accident is the point. I can only keep showing up where accidents happen.

And I should say what the handshake is for. Flirting is the opening of a frame. The thing I’ve concluded — and this is a conclusion, not a finding; there is no study, and the study that would settle it would fail ethics review in three different ways — is that the long version of the frame is what I was after all along. Two people who both know they’re playing. Who build each other a relaxed field, which is to say make each other fed and safe and unafraid, and then do the least functional thing possible inside it, repeatedly, never the same way twice, for years. Burghardt’s five criteria, held open by two people on purpose. I don’t have a word for this that isn’t embarrassing, so I’ll use the embarrassing one: it’s what love looks like when you describe it in ethology. Of all the things on my short list, it’s the one I have the least evidence for and the most confidence in, which tells you something about either the list or the evidence.

At my workplace, I have become the point where every department’s problems converge. Nobody assigned this. It accreted — a question here, a fix there, until I was the informal integration layer for an organization that has no formal one. The role is pure exploitation in the reinforcement learning sense: every hour goes to known problems with known solutions, with no margin for the exploratory move. It is a field under continuous load.

I have thought seriously about leaving. I’ve decided not to, and the reason, when I examined it, surprised me. The company will not change; I expect that. What keeps me there is a friend — the one with the children’s book project — and the open possibility of building something else with him. The only thing holding me in an environment with no play is the prospect of play. I stay in the serious room because of the door to the unserious one.

Bekoff’s self-handicapping dog is the best description of that kind of friendship I’ve found. The stronger dog pulls its bites, lets itself be pinned, rolls over on purpose — not out of weakness but because the game is worth more than winning it, and the game needs the other dog to stay. Humans do this rarely. Everyone is the center of their own universe, and the center shrinks when something hurts; this year I have been a very small universe, and I’d like to report that I handicapped myself generously for others, but the logs don’t support it. The rare friend who comes down to your level when it costs him something — who keeps the game going when he could win it or leave it — is as close as our species gets to the dog. I have one or two.

I am keeping my involvement in his project minimal right now, and I want to be clear that this is not a failure or a lapse. It is the one clause play has that nothing else in my life has: you can put it down. No postmortem. No exit interview. No one asks what went wrong, because nothing did — the relaxed field closed, and the behavior switched off, exactly as Burghardt’s fifth criterion says it should. I’ll pick it up again when the field reopens. Every other commitment I have would call that abandonment. Play calls it Tuesday.

Johan Huizinga argued in Homo Ludens (1938) that play is older than culture — that law, ritual, poetry, and war all began as games and only later forgot. Friedrich Schiller, in his Letters on the Aesthetic Education of Man (1795), put it in a sentence that has been quoted out of context for two centuries: man only plays when he is in the fullest sense of the word a human being, and he is only fully a human being when he plays. Neither claim is empirical. Both are the kind of thing one wants to be true, which is a reason to be careful with them.

The philosopher I actually find useful is Bernard Suits, who in The Grasshopper (1978) defined playing a game as “the voluntary attempt to overcome unnecessary obstacles.” Golf is the attempt to get a ball into a hole using means deliberately worse than walking over and dropping it in. The obstacles are the point. Remove them and you have an errand.

By that definition, this collection is a game. Nobody needs a cross-disciplinary essay on play with six lenses and a section cataloguing its own gaps. I could have written three paragraphs. I imposed the obstacles myself — the structure, the sourcing standard, the ironic voice, the rule that every claim gets labeled at its real evidential weight — and then spent weekends overcoming them. Suits would say this is the only thing I’ve done all year that qualifies as fully human. The Stoics, whom I cite in every one of these essays like a man checking his pockets for keys, would be less sure; they had a complicated relationship with activity that serves no end. I’ve decided to side with the grasshopper. The ants, I notice, do not write essays.

The opening of this piece is the closing of Entropy, re-read in a better mood: nothing is in charge, and the question is what you do in the gap. Play is the structural opposite of Overfitting: exploration against the local optimum. It is Fragility run in reverse — the margin, the surplus, the training for the unexpected that a lean system cannot afford and a robust one maintains. The noisy TV is Noise with a reward function attached. Play is the only domain in my life that Irreversibility doesn’t touch: the one two-way door, the only commitment you can walk back without a verdict. Bateson’s nip-that-is-not-a-bite is Superposition with fur — real and not-real, held simultaneously, collapsing the moment one party stops signaling the frame. And the attraction research reconnects to Projection: the list in your head predicts your feelings about people who aren’t there.

I haven’t engaged with Roger Caillois, who in 1958 split play into four categories — competition, chance, mimicry, and vertigo — and whose taxonomy might explain why some people’s play looks like chess and other people’s looks like a rollercoaster. I haven’t looked at the anthropology of play across cultures, where the Western assumption that play is for children may itself be a local artifact. I haven’t addressed the dark side: gamification, loot boxes, the deliberate engineering of play’s reward structure into products designed to be impossible to put down. I don’t know whether adult hobbies are the same phenomenon as juvenile play or merely resemble it. I don’t know whether the human prefrontal cortex is pruned by play the way the rat’s is, because nobody can run that experiment, and the rat remains, in a very real sense, the only subject we have. And I don’t know whether describing love as a frame held open by two people is an insight or a category error that happens to be comforting; the two are harder to tell apart than I’d like, and I’m the wrong person to referee.

And I don’t know whether this essay is play or a case for play. Suits’s definition says the former. My need to write it says the latter. The honest answer is probably that I started with the second and, somewhere around the fur seals, it turned into the first.

Here is what survived the audit, in the order it survived.

Nothing is in charge. I’ve stopped waiting for that to change, and I’ve stopped — this took longer — expecting the relaxed field to be issued to me by the universe, like a permit. The field is a gap between attacks. The attacks went on all season, and the pups played anyway. I’ve stopped reading that as recklessness.

Being upset about the chaos is a reasonable response and a useless one — and worse than useless: frustration doesn’t just fail to fix the field, it closes it. A stressed animal doesn’t play. For most of this year I treated my frustration as a form of insight — proof that I was seeing clearly. It was a precondition for not playing. I haven’t stopped. I’ve started hearing it as the sound of the field closing, which is at least a different relationship.

The one thing I’d rank above everything is being functional — a body that isn’t betraying me, which I’ll leave at that one line, since Fragility has the rest. I used to think the relaxed field was something you received. Some of it is. Some of it you can build, with unglamorous materials: sleep I protect, movement I do whether I feel like it or not, the word “no” said to things that only look important. I’ve also watched people who are seriously ill find a good side to it, which I won’t pretend to understand; I just note that the field can be narrower than I thought and still be a field.

Everything that made it through is the simplest possible thing. An evening with the few people I love at which nothing gets accomplished. A walk. A problem I understand on Sunday that I didn’t on Saturday. I spent years assuming the important things would be complicated. The inventory says the opposite. It is, I notice, Burghardt’s fifth criterion translated into a life — fed, safe, healthy, with your people — and the play is what happens next, when those hold.

So, concretely, three kinds of play, and I’ve moved all three from the “later” column into the current one. The long frame with one person — the one I have the least evidence for and the most confidence in; a direction, not a status, and the only item on the list I’ve stopped being ironic about. The handful of friends who handicap themselves for the game, whom I’m trying to deserve by doing the same, imperfectly. And this — the ideas, the essays, the weekend obstacle course — which is the second channel, the one the rat doesn’t have: when the body can’t play, the mind still can. Not a substitute. A separate circuit, and I’m grateful for it in the specific way you’re grateful for a backup you didn’t know you had.

I don’t schedule play, because scheduling it kills it. What I do is refuse to ask what it’s for. That’s the whole discipline. When the question “what’s the purpose of this?” surfaces — and it surfaces constantly, because I’m the kind of mind that generates it — I treat it as the noisy TV: unpredictable, attention-grabbing, and not something that improves the model. I leave the room.

And I’ll say the thing the structure demands, because the structure is one of the obstacles I chose: this essay is a case for play, built by someone who cannot stop building cases. The part of me that needs the case is the part the fur seals didn’t have. I have more vigilance than a fur seal pup and, apparently, less capacity to simply do the thing. The essay is my version of the tidal pool. The actual play happens in the moments I forget I’m making an argument — and I notice that those moments have been getting slightly longer.

Slightly. A rat would not be impressed. But the rat, as far as I know, has never written anything down, so we’re each ahead in one respect. The metadata below says this will change. For once, I believe the metadata.