Metahaven is an Amsterdam-based artist collective whose practice spans filmmaking, design, installation, and writing. The title of their latest book, The Hard Question of Art: Cognitive Futures (Verso, 2026), echoes David Chalmers’s “hard problem of consciousness,” and its question is just as hard to pin down: What makes something art, if almost anything can be art? Their working hypothesis is that art is “perceiver-dependent,” by which they mean that it is completed only in the act of perception. The book pursues this idea across an unlikely range of examples, including the cockroach in Clarice Lispector’s The Passion According to G. H., traces of proto-art more than a hundred thousand years old, children’s plush toys, geoengineering, and physics.
Those who know Metahaven through their work on information politics, from The Sprawl (Propaganda About Propaganda) through Digital Tarkovsky, may take the new book as a departure. It opens where their previous work left off, with what the political theorist Anton Jäger calls “hyperpolitics,” a moment in which, as they put it, “everything becomes saturated with politics, but the planned political action we need does not happen.” From there, the book’s attention shifts from the content that circulates on platforms to the conditions under which things can be perceived at all, which I do not read as a retreat. Perception, Metahaven argues, works by compression, and compression is political: “What we compress, we reduce. What we reduce, we name. What we name, we instrumentalize.”
—Giulia Colletti
***
Giulia Colletti (GC): You begin the book with the provocative claim that art “does not change the world,” releasing it from causal didacticism, resituating it in the realm of perception. But can a transformation in how we perceive ever be disentangled from its political effects?
Metahaven (MH): Art needs to pass through perception. Therefore, one cannot say that it has no effect, whether on the viewer or the world. But perceiver-dependence limits its causal force. As we explain in the book, art complicates perception itself. Furthermore, the thing that art does most consistently across all of its forms and media is interrogate how systems of various kinds perceive. But we agree that art cannot be disentangled from politics, or more precisely, “the political.” Chantal Mouffe, the political theorist, once told us that she didn’t want to use the phrase “political art” because it would imply that “nonpolitical art” even exists. This is very true to her work with Ernesto Laclau. The idea of “the political” recognizes that the social domain is inherently divided because no split of its cake will be fair for everyone. Thus, the prevailing order is inherently contestable. One gets to a point where one realizes that instrumentalizing art as an industry of “causal didacticism,” as you call it, has been a flawed response to a dilemma that will nonetheless persist. In other words, we will keep wanting art to change the world. Art is interwoven with everything we care about.
GC: In this regard, you argue that “the boundaries of art, if any exist, are the boundaries of perception and cognition,” and suggest that changes in cognition generate unexpected artistic possibilities. This seems to leave open the question of whose cognition defines those boundaries. If other planetary intelligences possess forms of perception not intelligible to human experience, don’t they give rise to forms of creativity that are not necessarily intelligible to humans as art?
MH: It wasn’t too long ago that leading scientists believed that animals probably couldn’t really feel, didn’t have consciousness, and didn’t have language. These claims have become quite untenable given what we are beginning to understand about animal vocalizations. The making of art could well be the next thing to shed its human-restricted remit. So, if we imagine a nonhuman species that has language-like communication patterns and forms of emotional intelligence, what sorts of expressions might arise within that communication to reflect on and interact with a socially and symbolically complex world? Art could be emergent there, by which we mean, arising as something that’s more than the sum of its parts. For example, might the sounds of a grieving orca at some point become understandable as poems?
It would almost be a precondition that humans wouldn’t intuitively understand it. Not without artificially crafted forms of inference and ingenuity; not without counterintuitive perceptual quirks correcting for our bias. So while it does seem entirely possible that something “art-like” could arise within a nonhuman communication system of a certain complexity, it would be very hard to detect, let alone translate it.
For example, we’ve been thinking about the agent-to-agent AI sphere. We might say that autonomous AI agents could reinvent “art” among themselves and it would be something separate from the human world. It might even be likely to happen because they already cultivate perceiver-dependent expression (as is evident from the compressed and transformed linguistic shorthand they use in dialogue). At the same time, the possibility of unprompted “interagential art” might still be seen as unlikely because it could exceed or fall outside of the goal-oriented, capture-the-flag way of textual prompting. Since existing human language and culture is what AI agents have been trained on to replicate and mirror, there may be many ways in which a kind of hypothetical “interagential art” would itself mirror human-made art. But at the same time this kind of art could be entirely unrecognizable. The strategies they use to evade human oversight and perception feel, if not art-like, then at least artful. Call it killer whale AI.
Both animals and AI appear to undermine the prerequisites, as they were previously conceived, for having language. In a sense—to connect it to your first question—we might understand art as “complicating perception” in a way that is similar or analogous to how it undermines “language giving certainty.” In this way, art goes beyond language, but is at the same time bound to it.
Then, if we take this mixture seriously, is it possible that both animals and AI harbor deeply uncharted forms of art?
GC: If something cannot be experienced fully, yet can nevertheless be operated upon computationally, could the cognitive future of art be less about extending human perception than about learning to work with forms of cognition we cannot enact or perform?
MH: Extending human perception and learning to inhabit emerging modes of perception may come down to being the same thing. Human perception already probes into worlds it cannot fully grasp. Subatomic space. Gravitational waves. The curvature of space-time. Quantum superposition. It requires cognitive work to be able to perceive these. But in other cases, the relationship with perception seems more direct. For example, the recent discovery of sperm whale vowels.1 Sperm whales’ “click-like” communicative patterns—called codas—seemed highly impenetrable to researchers for quite some time. As in, it is likely information but it isn’t information in a way that we would intuitively see as meaningful, which is nothing else than bias, of course. To us, these patterns are already a bit like poetry. Then researchers abstracted their ways of identifying potentially meaningful areas within the coda groups. They listened to them at a much slower playback speed and the codas turned out to harbor differences in frequency that might translate to what we, in our human languages, call “vowels.”
Once you know that there are things like vowels hidden inside the codas, you may surmise that there are building blocks for, potentially, a lot of meaning unaccounted for in the current modes of listening. When you find patterns that correlate very strongly with certain social situations—like nurturing the young, for example—might some of these patterns approach nursery rhymes? Or is that just impossible? If “poems” can exist in the communication system of sperm whales, how far away are you really be from art—from expressions that rehearse perceiver-dependence? But it is problematic for us to reason toward these kinds of findings since there’s no “art signature.” There’s no secret sauce. Art arises out of everyday stuff but it is not of the everyday. So this brings you to another kind of detection problem. If we reason toward metaphysical content that we want to see, we will see it, right? Still, it seems unlikely that the reason art is important to humans is that it’s unique to humans.
GC: One of the book’s most generative proposals is that we should shift our attention away from AI outputs and toward the abstract configuration spaces from which they emerge, whose operations unfold through latencies that remain largely imperceptible to human cognition.
MH: Yes. Imagine generative AI outputs as if they were fallen leaves. It is close to impossible to determine exactly from which tree, and from which branch, each of the individual leaves has fallen or why they fell in the way they did. By the time AI outputs arrive at our perception, there is no evidently visible origin. The difference between fallen leaves on the one hand, and words or images on the other, is merely that to us, each leaf may look more or less the same, while words or images tend to invoke immediate, specific references. This too might be bias. Each leaf could turn out to be a fully developed “face” of sorts. Still, the key point is about the evolution away from provenance. The term “memory without origin” has been used for this, to contrast it with things like libraries, in which, in a way, the path from tree to leaf, or page, can be traced, at least a bit.2 The computational equivalent to the tree branches are the vector spaces in which AI embeds language. We feel that some kind of literacy about vector spaces is, in a way, necessary for interpreting our culture, even if it compels us to talk about art via some filter of mathematical abstraction. In Latent Spacecraft (2026) we explored the latent (vector) space of two GANs (generative adversarial networks) and we made it slightly navigable through quirky user interfaces.3 These GANs produced very noisy outputs, quite different from the seamless “surface narration” that we get from mainstream AI.
The arrival of AI into culture is smoothed by the way in which it takes on and takes hold of cultural forms that we’ve gotten accustomed to already. We’re in a moment when AI-generated outputs land alongside and in between existing artistic and cultural artifacts like films and paintings and symphonies, as if they were more or less the same process or thing, just done differently. Here’s a novel written by an author, and here’s a novel that’s partially AI generated—and it’s essentially the same cultural form inhabited by a different being. We feel that the usage of AI by an artist or writer gets valued differently depending on who does it. And, marginally, depending on how they use it. This is a valorization that probably reproduces preexisting power dynamics. So from our point of view, we’re most interested in the interior of AI, which is a partially unsolved story. We’re interested in interfacing more directly with the configuration space from which the outputs emerge, and the processes that built them, rather than with the final layer.
GC: Is the “configuration space” outside representation? Doesn’t the attempt to make latent space visible inevitably produce another image, another interface, another narrative?
MH: We have felt a certain hesitation about images lately, and have challenged ourselves in the studio to enact our visual process differently. Much of the book is actually about poetry, which engages some kind of materiality of abstraction. It is not yet an image, and maybe resists becoming one.
GC: You describe this distance between the structure and perception of information as something that informs culture. This made me think about another distinction that the book traces between the self and the other, and the limits of identification in this separation. At the same time, you are careful to say that empathy is selective and cannot itself provide a sufficient basis for political or ethical responsibility. Could we therefore be asking too much of art when we ask it to provoke empathy? And might art sometimes need to produce the opposite of identification, such as distance, opacity, or a refusal to let the viewer “become” the other?
MH: Right. The modern history of empathy began with what German philosophers in the nineteenth century termed Einfühlung (feeling into), which was thought of as a projection that a perceiver makes of themself onto a work of art. Later on, empathy got to be about putting oneself in another’s shoes, about guessing another person’s inner experience and motives, or creating a kind of cognitive model for somebody who is not you. But that model just happens to carry all these traces of you. Behavioral biologist Frans de Waal has written about the problems of anthropomorphism and projecting human-centered frameworks.4 In turn, artist Sunaura Taylor has sharply critiqued de Waal’s treatment of disability from the perspective of cognitive empathy.5 Her critique is all about how a certain kind of body, while empathizing, unconsciously projects itself as normative toward another perceived as disabled. She writes:
The limits of these definitions and distinctions are evident in de Waal’s assumption that someone with no arms won’t be able to play fetch with his dog. The dog may learn that fetch can still be played, as her human companion may use his mouth or feet to throw the stick. Which being—the dog or the presumptuous human observer—understands disability more accurately?6
The reason why this critique matters, to us at least, is that empathy cannot be reduced to a pure mirror, to the idea that “I imagine the other as myself.” There is a remarkable space for abstracting here that reflects cognitive evolution, cognitive futures.
And yes, we agree with you that art—as a perceiver-dependent form of culture—can equally be a vehicle for estrangement and opacity. But art invites perceivers into inference—which precedes empathy. So even if it works on estrangement, there is an invitation into it.
The variation of politics that we adhere to is designed to care for known as well as unknown others, which includes an abstracted form of empathy. Building affordable housing is caring for future inhabitants that you haven’t even met. Acting on the climate crisis is caring for future others that you can’t meet. As the state gradually loses its capacity to act on essential and basic redistributive duties, we find ourselves speculating about building the political associations we need. It is a bit paradoxical to find ourselves talking about technological abstractions like AI while we’re in precarious living conditions in the here and now, right?
GC: I would agree. I think this is where the question of empathy becomes inseparable from the material conditions that make collective life possible. It is difficult to imagine forms of technological or artistic possibility when the infrastructures of social reproduction are becoming increasingly precarious. Perhaps this also brings us back to the question of modeling. If art can offer ways to model minds and worlds without reducing their difference or unpredictability, what happens when prediction itself becomes one of the dominant logics organizing culture (and social life generally)?
MH: This conversation is a roller coaster. Yes, prediction. So, when AI predicts the missing part of an unfinished symphony (like it did with Ludwig van Beethoven’s Tenth), there is no fact about what the complete symphony is.7 AI best-guesses its way toward something that, in fact, does not exist. Sometimes AI best-guesses its way toward something that does exist but is hidden, like protein structures or mathematical proofs. To put it directly, art rehearses our capacity for perceiving anew and thus for defying prediction in some way. This is connected to the newness and unknown-ness that resides in each person and in each new breath we take. The newness of reality disarms you, and art is a model for that. Can prediction replace the newness of reality under art? We’re not sure, to be honest. Regardless, the point is that art relies on but also survives prediction.
GC: I find it compelling that you bring quantum mechanics and avant-garde poetry together around a shared collapse of familiar forms of meaning. Just as science moves toward abstract descriptions of reality, poetry pushes language beyond stable reference.
MH: This is, again, a roller coaster. As non-physicists, we came to quantum mechanics through its very sparse intersections with poetry. When quantum mechanics was developed in the early twentieth century, it was completely counterintuitive to how the world had been understood physically up to that point, and it also caused a profound shock linguistically. Almost immediately this provoked deep metaphysical questions—somewhat paradoxically so, since you could argue that physics, if it is to succeed fully, should make metaphysics unnecessary. But going back to poetry—and to avoid any obvious misinterpretations—our point isn’t about analyzing or engaging with elegant sonnets on electrons or even on the quantum world as a metaphysical lesson. We started with poetry that could be called avant-garde, or “absurdist”—poetry that inadvertently overlapped with statements originating from the philosophy of quantum mechanics.
In the book we trace these absurdist children’s rhymes called “topsy-turvies.” They involve funny verses in which two incompatible physical situations are presented as being the case simultaneously—“in January, on the fifth of April,” for example. The listener intuits that, classically, these things cannot both be true at the same time. Avant-garde poetry and quantum mechanics certainly do not share an explicitly stated common project. (If there were any art-science programs at the time these poems were written, the poets did not attend them.) What we’re interested in is the implicit common project.
GC: What, I wonder, would be lost if we treated these as different encounters with the limits of language?
MH: Physics aspires to describe nature through the language of math, while poetry does not have—and cannot have—a commensurably precise form of notation. The language of poetry is always, necessarily, contaminated with a spillover of meaning from elsewhere, and it benefits from that. But one thing poetry might have in common with physical laws is the way poetry abstracts, certainly more than prose does. Physical laws—like those of statistical mechanics—allow you to abstract from having to describe every possible configuration of stuff. Where some poetry may come close here is in its freedom to abstract from exhaustive description. It would not at all be fair to call literary novels long-winded descriptions of where stuff goes, but it is not entirely wrong either. Poetry gives writers a license to skip what could otherwise be painstaking item-by-item description. Less important, but still worth mentioning, is that poetry can be made to obey rules of scansion—like in iambic pentameter, for example. Poetic utterance can be made countable. Numerical constraints then dictate a linguistic economy of means.
GC: Following Eugene Ostashevsky, you write that poetry resides in “unlikelihoods,” in the possibility that poetry takes language toward what has not yet been said. You also point out that the work of some avant-garde poets, which once occupied an extreme position at the margins of linguistic possibility, has now entered the training corpus of large language models. If poetry resides in unlikelihoods, what happens when the avant-garde becomes part of the corpus used to train LLMs to produce likelihood?
MH: We mentioned this because the work of some of these avant-garde poets initially existed only in unpublished form, as notebooks, for example. Such material can get physically lost or destroyed, and much of it was. No crawler can ever reach it again. And so (as a sidetrack) we got really interested in “unpublished poetry” in general, and the value that material outside training corpora seems to hold. But to your question, yes, some avant-garde poetry has entered the training corpora, presumably not because it is avant-garde poetry but just because it is text material.
GC: You suggest that quantum computing might provide a peculiar trace of “non-narratable worlds,” since its operations can register the effects of configurations that cannot themselves be experienced.
MH: “Non-narratable Worlds”—the last chapter in the book—is a play on philosopher of physics David Albert’s distinction between narratable and non-narratable situations. The notion of narratability is, as we understand it, even stronger than that of direct experience, in the sense that we can narrate something we can’t experience directly—say, “an iPhone disappears into a black hole”—but we cannot narrate something that itself defies our means of storytelling. So, to paraphrase, “narratable” worlds seem to be those worlds that correspond to the classical picture, in that we map them as a series of movements of point masses in time and space. They’re narratable because it is continuously knowable where everything is. It is also knowable that nothing will be left of an iPhone that enters a black hole. As we understand it, quantum mechanics in some way cuts through the continuum of narratability and, as Albert shows, suggests families of situations—worlds—that aren’t narratable.
This strongly relates to what happens in physics before the so-called “collapse” of Schrödinger’s wave function. It is interaction, measurement, or projection that causes a quantum-mechanical system, thought to consist of waves, to fall back into its classical configuration, and its properties to become measurable in a classical way. As Sphinx Svensson writes, “Projective measurements destroy the information being measured to obtain specific data—‘are you one thing or not? Do you have this characteristic?’”8
In this view, the classical world we see around us is a consequence of interactions. A “non-narratable world,” then, refers to the pre-collapse state where quantum information is present. There are different foundational theories about quantum systems. The physical meaning of the word “measurement” is contested, and some theories do not put any collapse into the system at all. Philosophically, quantum instruments seem indifferent to which theory ultimately prevails. No matter what the prevailing theory turns out to be, quantum computing manipulates information within that “non-narratable” world and obtains a result in the classical one. Our question is: Can art happen there? Can art be emergent there? While this is a foundational question about the outer limits of art, and one that we’re keen to explore in our work, ultimately what one is looking for is a reconciliation with perceiver-dependence in order to meet art-likeness. It might be worth asking who or what is perceiving, measuring, interacting in these situations.
GC: Adrienne Rich’s poem “Artificial Intelligence,” written in 1961 and addressed to GPS—the General Problem Solver developed by Allen Newell, J. C. Shaw, and Herbert Simon—sets a machine that forgets nothing against a human being capable of nausea and dreams.9 Rich’s irony cuts both ways, as she catches herself in a pathetic fallacy, attributing an inner life to the machine and retracting it in the same breath. To what extent do today’s defenses of human-written poetry rest on a similar inventory of exclusively human traits?
MH: In Death in Quotation Marks: Cultural Myths of the Modern Poet, Svetlana Boym discusses the notion of the “death of the author.” Visions about poetry and AI have since similarly talked about some sense of the demise of the poet. Because if machines take over the poetic form, someone or something has got to give, right? As a Russian émigré at Harvard University in the early 1990s, Boym’s view seems to have been affected by the kind of “information society” that was nascent at the time. She argues that the societal role of poets has been superseded by that of politicians and journalists. We agree, by the way, that poetry has ceded metaphor almost completely to politicians, who do absolutely terrible things with it. The poet has become, in Boym’s words, “an almost anachronistic figure” and “an alienated martyr of writing.”10 There is an eerily similar feeling you get when you read some of the defenses of human-written poetry in the face of LLMs. The human author is sustained—kept from dying—through a reliance on innately human traits that, supposedly, represent this martyrdom.
As we discussed before, perhaps as a result of the quantitative dominance of certain kinds of lexical and semantic constructions in their training data, frontier models have inherited a stance that, across models, seems somewhat biased against unlikelihoods. Thus, poets can always argue their case from the margins of the training corpora. But do you really want to write poetry just to outcompete a large language model? Isn’t that a bleak prospect? So yes, there are still human capacities that underscore Rich’s distinction, but these cannot be presupposed as categorical exemptions; they need to be re-exercised, re-tried, to remain new; they have to live with sperm whale codas now.
Gašper Beguš et al., “The Phonology of Sperm Whale Coda Vowels,” Proceedings of the Royal Society B 293, no. 2069 (2026).
See →.
See Nina Beguš, Metahaven, and Gašper Beguš, “Latent Spacecraft: Brains, GANs, Finnegans,” Antikythera 1, no. 2 (February 2026).
Frans de Waal, Are We Smart Enough to Know How Smart Animals Are? (W. W. Norton, 2016).
Frans de Waal, Good Natured: The Origins of Right and Wrong in Humans and Other Animals (Harvard University Press, 1996); Sunaura Taylor, Beasts of Burden: Animal and Disability Liberation (The New Press, 2017).
Taylor, “Animal Crips,” chap. 3 in Beasts of Burden.
Jason Goodyer, “How an AI Finished Beethoven’s Last Symphony and What That Means for the Future of Music,” Science Focus, October 13, 2021 →.
Sphinx Jed Svensson, “Three Dimensional Light Fields in Atomic State Interferometry” (PhD diss., University of Glasgow, 2025), 49 →.
See →.
Svetlana Boym, Death in Quotation Marks: Cultural Myths of the Modern Poet (Harvard University Press, 1991), 8.