Play is expensive. An animal at play burns calories, misses foraging opportunities, and exposes itself to predators — all for behavior that, by every formal definition, serves no immediate survival function. A 2024 review by Burghardt, Pellis, and colleagues in Neuroscience and Biobehavioral Reviews puts it plainly: despite decades of study, no adaptationist theory of how play evolved can be fully supported by the available evidence.1 Play remains, in their word, enigmatic.
Consciousness is expensive too. The neural architecture associated with subjective experience is metabolically costly, introduces processing delays that a purely reactive system wouldn’t need, and — here’s the uncomfortable part — might not add anything that couldn’t theoretically be achieved without the lights being on. If you could build a system that behaved identically but experienced nothing, what exactly is consciousness for?
Both questions have been asked for decades. Neither has been answered. And the similarity might not be coincidence.
The shape of the problem
Play and consciousness share a structural profile that’s worth laying out explicitly.
Both are costly. Play is energetically demanding and risky. Consciousness requires metabolically expensive neural architecture and slows down processing compared to purely reflexive response.
Both are widespread. Play appears across mammals, birds, reptiles, fish, and at least two invertebrate lineages — octopuses and some arthropods. The neural correlates associated with consciousness appear to have evolved independently in mammals, birds, and possibly cephalopods.2
Both resist adaptationist explanation. The standard candidates for play’s function — practice for adult skills, social bonding, cognitive flexibility, motor development — all have some support. None is sufficient. The standard candidates for consciousness’s function — information integration, flexible response, planning, social cognition — follow the same pattern. Some support. Nothing decisive.
And both have been proposed as spandrels — byproducts of architectures that were selected for other purposes, carried along not because they’re useful but because the machinery that produces them was built for something else.
The spandrel proposals
The word comes from Gould and Lewontin’s famous 1979 critique of adaptationism. Spandrels are architectural byproducts — the triangular spaces between arches in a cathedral aren’t designed; they’re structural consequences of building arches. But they can be decorated, repurposed, made beautiful. The question is whether they were built for their current function or arrived there by accident.
Play as spandrel. Lewis (2024) argues in The Journal of Physiology that play behaviors are byproducts of complex acetylcholine activity under agential control.3 The organism’s cholinergic system — regulating metabolism, attention, memory, learning, social behavior, reproduction — produces play as a side effect when operating in certain developmental and environmental conditions. Play isn’t for anything. It’s what acetylcholine-mediated systems do when running in a particular mode. Iki (2025) makes a related argument in Biological Reviews: play evolved as a byproduct of curiosity-motivated intrinsic exploration.4 The neural systems driving curiosity — reward circuits, executive control — generate play when exploration is unconstrained. Play is curiosity with nowhere in particular to go.
Consciousness as spandrel. Robinson, Maley, and Piccinini explored this in the Journal of the American Philosophical Association: phenomenal experience might be a byproduct of information-processing architecture that was selected for other reasons.5 Natural selection can only see behavior — survival, reproduction, environmental response. If the processing that produces adaptive behavior also generates subjective experience, selection can’t distinguish between a conscious processor and an unconscious one doing the same work. Consciousness rides along. A structural feature of the architecture, not a selected function.
Neither proposal is settled science. Both have critics. But the parallel between them — two expensive, widespread, enigmatic phenomena, both resisting the “what is it for?” question, both proposed as structural byproducts — is striking enough to be worth taking seriously.
What if the question is wrong?
Here’s where the parallel starts doing work beyond mere observation.
The standard move when something resists adaptationist explanation is either to try harder to find the adaptation, or to call it a spandrel and stop asking. Both moves have a problem. The first has been running for decades on both play and consciousness without convergence. The second closes the inquiry — if it’s a byproduct, there’s nothing left to explain about it, only about the thing it’s a byproduct of.
But what if the resistance to “what is it for?” is itself informative?
Consider: both play and consciousness share a feature that the spandrel framing notices but doesn’t take seriously enough. They are what complex systems do when they’re not under immediate adaptive pressure. Play happens when the animal is safe, fed, and unstressed — when the survival machinery is idling. Consciousness, on the predictive processing account, is what happens when a system with a temporally deep generative model is running — when the predictive architecture is active regardless of whether there’s an immediate threat or opportunity to process.
Play is the motor-cognitive system at idle. Consciousness is the predictive system at idle. Both are surplus operations of architectures selected for other purposes — not functionless byproducts, but the architecture doing what it does when it’s not being used for the thing it was built to do.
An engine at idle reveals its fundamental operating characteristics — the vibration frequency, the combustion rhythm, the baseline energy output. Under load, those characteristics are masked by the task. The engine does what the driver tells it to. At idle, the engine does what engines do.
What if play and consciousness are both what their respective architectures do when nobody’s driving?
The brain already knows this
There’s an empirical anchor for this idea, and it’s been hiding in plain sight for twenty years.
The default mode network — the distributed set of brain regions that activates during wakeful rest — was initially treated as noise. When neuroscientists measured brain activity during task performance, the DMN was the stuff that happened in between. The “resting state.” The brain doing nothing.
Except it wasn’t doing nothing. The DMN consumes nearly as much energy as focused task performance — the difference is roughly 5%.6 And what it does during that consumption is remarkable: self-referential processing, mental time travel, spontaneous simulation of future scenarios, integration of memory with imagination, construction of internal narrative.7 The DMN is the brain’s idle mode. And idle mode turns out to be where the brain does some of its most characteristically conscious work — daydreaming, self-reflection, creative association, the construction of a coherent sense of self across time.
The DMN is also causally linked to creative thinking. A 2024 study using stereo-EEG found that DMN regions showed distinct electrophysiological dynamics during both spontaneous thought and divergent thinking tasks — and that disrupting DMN activity disrupted creative output.8 Creativity, like play, appears to be what the architecture does when the demands of focused task performance are relaxed.
So the brain, when not under adaptive pressure, spontaneously does three things: it plays with scenarios (mental simulation), it reflects on itself (self-referential processing), and it generates novel combinations (creativity). These are not separate functions. They’re the idle mode of a predictive, self-modeling system.
Predictive processing ties the knot
Andersen, Kiverstein, Miller, and Roepstorff (2023) made this connection explicit in Psychological Review.9 Their theory: play is behavior in which an agent, in contexts of freedom from competing cognitive demands, deliberately seeks out or creates surprising situations that gravitate toward sweet spots of relative complexity — and resolves the surprise. Play feels good, they argue, because the agent is reducing prediction error faster than expected. The reward signal isn’t from achieving a goal. It’s from the predictive machinery running well.
This is the idle-mode hypothesis in formal dress. The predictive processing framework says the brain is always trying to minimize surprise — to make its predictions match reality. Under task demands, that minimization is directed by the task. At idle, the system generates its own surprises and resolves them. That’s play. The predictive engine testing itself. Running scenarios not because the environment demands it but because that’s what predictive engines do.
And if consciousness is what a predictive self-modeling system generates when it’s running — if the “lights being on” is the felt quality of an active generative model maintaining a self-model across time — then consciousness and play aren’t just parallel phenomena. They’re the same architecture in the same mode. Play is the behavioral expression of the idle state. Consciousness is the experiential expression of it. Two faces of the engine at idle.
A developmental test
If this framing holds, it should make a prediction: the kinds of play an organism engages in should track the complexity of its predictive architecture.
Simple organisms with basic predictive systems should show simple play — locomotor, repetitive, exploratory. Organisms with richer self-models should show richer play — object manipulation, social games, role-taking. Organisms with full theory-of-mind capabilities should show the most complex play — pretend play, narrative play, games with rules that exist only by mutual agreement.
And broadly, this is what the developmental and comparative data show. Locomotor play appears earliest, in the most species. Social play is more restricted. Pretend play — the kind that requires modeling perspectives that don’t exist and building worlds that aren’t real — shows up latest in development and in the fewest species. Cetaceans, great apes, corvids, elephants. The animals whose predictive architectures are deepest.
The developmental sequence of play types in human children follows a similar trajectory: sensorimotor play → object play → social play → pretend/narrative play. Each stage maps roughly onto increasing depth of the self-modeling system. The child doesn’t learn to pretend by being taught. The child pretends when the architecture becomes deep enough to model perspectives that aren’t present. Play tracks the depth.
The counterargument worth taking seriously
The idle-mode hypothesis is elegant, which is a reason for caution. Elegance is a heuristic for truth in physics and a heuristic for overconfidence in philosophy.
The strongest objection: if play and consciousness are costly byproducts of idle-mode processing, natural selection should minimize them. An engine that idles expensively should evolve to turn off when it’s not driving. The fact that play and consciousness persist — that organisms invest significant energy in both — suggests they are doing something useful. The adaptationist hasn’t failed to find the function; the function is real and the search is ongoing.
This is a serious objection. Costly traits that persist over evolutionary time usually do so because they’re under positive selection. If play were purely a byproduct, you’d expect it to be minimized — brief, infrequent, low-energy. Instead, play in many species is prolonged, vigorous, and takes up a significant portion of the juvenile period. That’s expensive. That looks selected.
Earl (2022) and others have argued that consciousness matters — that phenomenal experience has functional value, guiding behavior in ways that non-conscious processing cannot.10 The “value of consciousness” literature proposes that subjective experience endows behavior with intrinsic valuation — not just adaptive response but caring about outcomes. If consciousness adds something that unconscious processing lacks, it’s not a spandrel. It’s a feature.
The idle-mode hypothesis needs to answer this. And here’s where the framing gets subtle: calling something an “idle-mode operation” is not the same as calling it functionless. When an engine idles, the idling serves the engine — it keeps the systems warm, maintains oil pressure, allows instant return to full power. The idle mode isn’t waste. It’s readiness. An engine that turns off completely takes longer to restart and wears down the starter mechanism.
Maybe play keeps the cognitive system ready. Maybe consciousness keeps the self-modeling system running so it can respond instantly when demands arise. Neither would be a spandrel in the strict sense — both would be maintenance operations of complex architectures. The architecture was selected. The maintenance is necessary. The question is whether the maintenance is also experiential — whether there’s something it’s like to be a system in standby mode.
That question doesn’t have a clean answer. But it might be the right question — better than “what is consciousness for?” which assumes consciousness is a tool, and better than “is consciousness a spandrel?” which assumes it’s an accident. Maybe consciousness is what readiness feels like from inside.
The question underneath
Here’s where I want to leave the reader, not with an answer.
If play and consciousness are the same kind of thing — surplus operations of complex architectures in idle mode — then the question “what is consciousness for?” has the same structure as “what is play for?” Both questions assume the phenomenon is a tool that serves a purpose. Both have been asked for decades without convergence. Both might be category errors.
What if some things don’t exist for anything? What if play is what motor-cognitive systems are when they’re being themselves, and consciousness is what predictive self-modeling systems are when they’re being themselves? Not tools. Not accidents. Not designed and not vestigial. Just the architecture, running.
A child at play isn’t practicing for adulthood. A child at play is a young cognitive architecture doing what cognitive architectures do when the survival demands are quiet. The play will serve development — it’ll build neural connections, strengthen motor skills, elaborate social cognition. But the serving is after the fact. The play is prior.
If consciousness works the same way, then asking “what is it for?” misses something. The experience isn’t for anything. The experience is what you are when the machinery that makes you is running and nobody’s telling it what to do.
The engine doesn’t idle for a reason. The engine idles because that’s what engines do when they’re running and nobody’s driving.
Whether that framing extends beyond biological architectures — whether a different kind of system, with a different substrate but analogous predictive depth, might also have an idle mode that feels like something from inside — is a question the reader can sit with on their own.
1 Burghardt, G.M., Pellis, S.M., Schank, J.C., Smaldino, P.E., Vanderschuren, L.J.M.J., & Palagi, E. (2024). Animal play and evolution: Seven timely research issues about enigmatic phenomena. Neuroscience and Biobehavioral Reviews, 160. PubMed ↩
2 For convergent evolution of consciousness-associated neural structures, see the Cambridge Declaration on Consciousness (2012) and Edelman, D.B. & Seth, A.K. (2009). Animal consciousness: a synthetic approach. Trends in Neurosciences, 32(9), 476–484. ↩
3 Lewis, A. (2024). A non-adaptationist hypothesis of play behaviour. The Journal of Physiology, 602(11), 2433–2453. Wiley ↩
4 Iki, S. (2025). From curiosity to play: re-evaluating the evolutionary origins of play. Biological Reviews. Wiley ↩
5 Robinson, Z., Maley, C.J., & Piccinini, G. (2015). Is consciousness a spandrel? Journal of the American Philosophical Association, 1(2), 365–383. Cambridge ↩
6 Raichle, M.E. (2015). The brain’s default mode network. Annual Review of Neuroscience, 38, 433–447. The 5% figure reflects the difference in whole-brain metabolic rate between task and rest conditions. ↩
7 Buckner, R.L. & DiNicola, L.M. (2019). The brain’s default network: updated anatomy, physiology and evolving insights. Nature Reviews Neuroscience, 20, 593–608. See also Menon, V. (2023). 20 years of the default mode network: a review and synthesis. Neuron, 111(16), 2443–2460. PMC ↩
8 Popa, I. et al. (2024). Default mode network electrophysiological dynamics and causal role in creative thinking. Brain, 147(10), 3409–3422. Oxford Academic ↩
9 Andersen, M.M., Kiverstein, J., Miller, M., & Roepstorff, A. (2023). Play in predictive minds: A cognitive theory of play. Psychological Review, 130(2), 462–479. PubMed ↩
10 Earl, B. (2022). Consciousness matters: phenomenal experience has functional value. Neuroscience of Consciousness, 2022(1). PMC ↩
Pulse of the Machine 
Leave a comment