BrutAIsm

A forest is not art. It is complex, beautiful, inspiring — but it is not art. The sensible person walking through gets imprinted with the experience: the light, the scale, the smell of decay, the specific way their body occupies space among the trees, passing through their particular mental history — through affinities, trauma or associations they can't help but make, the whole sum of their lived experience. They will ingest this qualia, stacking it on top of previous experience, now further contaminated and deformed by this very encounter.

So let us consider "the artist": a particularly sensible accumulation of these stacked deformations. Since experience also reshapes the apparatus of perception, the artist who paints the forest at twenty, and again at sixty, produces different works, not just because the forest had changed, but because the filter did. Art is a process: the fermentation that occurs when a deformable consciousness meets the world, transducing qualia and encounters into mental medium, then communicated, to be decoded again into another qualia, becoming another deformation of another mind.

This is the first friction. Experiential friction, the friction of being. A life lived leaves marks, and those marks shape everything that passes through the person who carries them. It cannot be taught, purchased, or optimized away. It is the accumulated residue of having a body, a history, and a finite amount of time. Think of it as experiential charging: a slow, involuntary process by which a human consciousness becomes loaded with specific, unrepeatable content. The artist does not choose to be charged. They simply live, and the living does that work.

But then, there is a second friction, which operates in the production stage. Having breathed in the raw material of the world, having been charged by a lifetime of encounters, the artist sits down to produce — and immediately runs into the stubbornness of their chosen medium. The properties of oil paint, the weight of stone, the finitude of a camera shutter opening. The medium does not transmit the signal cleanly. It resists, it distorts, it imposes its own character on whatever passes through it. This is material friction: the specific resistance that a medium offers to the passage, from intention to form. Painting resists through flatness and the chemistry of pigment. Sculpture resists through gravity and material quirks. Photography resists through time — the shutter opens and closes, the moment is caught or lost. Think of this as the release: artistic intentionality passes through a resistant material and is shaped by its inherent properties.

These two frictions are not incidental to art — they constitute it. The charging, or experiential friction, gives the work content, the biographical and unrepeatable quality that makes one artist's forest different from another's. The release, or material friction gives the work its form, the struggle with the medium that makes the passage legible, turning private experience into public artifact. Clement Greenberg argued as much: artistic value comes out of the struggle against the limits of art’s medium. The flatness of the canvas is not painting's weakness — it is painting's subject. Remove either of these frictions and something essential collapses. Without the charging, there is nothing to say. Without the resistance of the medium, the shape of that message is not specific.

Generative AI, as it currently exists, has neither. No qualia to supply the charge — no body, no mortality, no accumulated scars from living. And no material resistance to shape the release — the output slides frictionlessly from probability to pixel, seemingly shaped by nothing (or, one could argue, even worse: the randomness of noise diffusion), and what arrives is smoothed, generic, and structurally incapable of this specificity that art requires.

So when the world asks, as it endlessly does nowadays — can AI make art? — the honest answer is: no, or at least not like this. This question is tangled in assumptions that need to be taken apart before they can be reassembled. To get anywhere near a real answer, we first need to understand the conditions that made the question possible.

The crisis did not begin with AI. It has been arriving, programmatically, for decades.

We live in a regime of permanent flux (Boris Groys, In the Flow, 2016). On our screens, a carefully mastered work of art occupies the same rectangle as a brainrot meme — same scroll speed, same half-second of attention before the next thing arrives. When images cost almost nothing to produce, they also cost nothing to discard. Rarity once imposed a demand for attention: a painting existed in one place, and you had to go there to experience it. A photograph required chemistry and time. The costliness of production begged the provision of an implicit spotlight. But when production approaches zero cost and infinite volume, the spotlight dissolves into ambient glow. Difficulty and specificity become liabilities. The image that asks you to slow down is competing against millions that don't, and the economics of attention are not in its favor.

Alongside this flattening of consumption, a subtler transformation has been underway in the image itself. For most of human history, images were indexical: photographs, for instance, resulted from physical contact between light and a surface. They, more or less directly, recorded light that bounced off an actual body, traveled through a lens, struck a chemical emulsion. A painting records the gesture of a hand or brush that moved by some design. Even at the limits of these definitions, these images would bear a causal trace of the thing they’d depict, the way a footprint bears the trace of a foot. They were somewhere. They were caused by something.

But we have been migrating away from this — from optical images toward statistical images: computed from aggregated data rather than caused by contact with the world (John May, Signal. Image. Architecture., 2019). Digital photography discretized light into pixels and let software reconstruct the scene. Computational photography — HDR, night mode, portrait blur — began compositing multiple exposures, hallucinating depth, sometimes inventing optical effects that no lens produced. The image has become increasingly computed, rather than captured. But we didn't notice, because the output still looked like a photograph. The corpse of indexicality was still dressed to look very much alive.

And then there is the smoothness. Byung-Chul Han describes a culture in which everything is optimized to be pleasant, consumable, and free of resistance (Saving Beauty, 2015). Surfaces without texture. Content without difficulty. Experience without the risk of genuine encounter. In the case of AI, this is not merely an aesthetic preference; it is a structural condition. When the means of production eliminate friction by design, the culture they produce cannot help but be smooth. The mean is, by mathematical definition, the point of least resistance.

Generative AI did not start any of these processes, but it finished all of them simultaneously. The statistical image has no referent: no scene produced it, no light caused it, no moment in time anchors it. It emerges from a region of mathematical space that is statistically adjacent to millions of images that once had referents. But those causal chains have been dissolved, compressed, averaged into weights. The image looks coherent. It looks like it was made by someone who was somewhere, looking at something. But there is no someone, no somewhere, no something. There is only a probability distribution, momentarily crystallized into pixels.

So the haunting begins. Mark Fisher called it hauntology: a culture that has lost the ability to imagine genuine futures because it is trapped in the endless recombination of its own past (Ghosts of My Life, 2014). Generative AI is a hauntological machine in the most precise sense — but it doesn't merely recycle the content of past art, the way a remix recycles a melody. It recycles the whole media of past art. It produces painting without pigment, photography without light, sculpture without gravity. Every output is a ghost — not of an image, but of a medium. It wears the appearance of oil on canvas, of silver gelatin, of chiseled marble, but it has never touched any of these substances. The only thing it cannot imagine what art native to itself would look like, because it is too busy performing the memory of what art used to look like in bodies it never had.

Ninety years ago, Walter Benjamin worried that mechanical reproduction would destroy the aura — the unique presence of an artwork in a specific time and place (The Work of Art in the Age of Mechanical Reproduction, 1935). But generative AI does something more radical: it produces images that never had an aura to begin with. There is no original. There is no "here and now" from which copies depart. The image was never anywhere. The aura is not destroyed — it is structurally absent.

Now the table is set. Consumption patterns have flattened all images into interchangeable units of scrollable content. The cultural surface has been polished until conflict and friction are liabilities. The causal link between images and reality has been severed. And the medium that produces these images cannot carve its own future because it is constituted entirely from a compressed, averaged past. The ultimate simulacrum has arrived — and it arrived not with a bang, but with a scroll.

The question is whether there is anything here worth salvaging. To answer that, we’ll need to look inside.

Every image has to come from somewhere, even one that was never anywhere. The place generative AI images come from has a name: latent space: the high-dimensional geometry from which all generated images emerge.

It is not empty.

It is extraordinarily structured — clusters, gradients, neighborhoods, voids, attractors. It contains the compressed record of human visual culture: every painting, every photograph, every sketch that entered the training data left its trace in the arrangement of weights. Every image made by someone who felt something - who grieved, loved, was bored or desperate to capture a particular quality of afternoon light - all of it contributed to the topology of this space.

It is, in a hauntingly real sense, a collective graveyard of crystallized qualia.

Millions of human experiences, each carrying its own friction, its own biographical charge, compressed, averaged and deposited into a single geometric structure.

But it is a fossil, not a living organism. A crystal, not a metabolism.

A living consciousness is not only shaped by experience, but continues to be shaped. Each encounter deforms the apparatus, and the deformed apparatus encounters the world differently, only to be deformed again.

This iterative, even recursive feedback — sensation, deformation, new sensation, new deformation — is what makes a consciousness alive in the relevant sense. As stated in the opening paragraphs, the artist at sixty years old is a different instrument than the artist at twenty, not because they decided to change, but because forty years of friction left marks.

The internal landscape shifts, it grows and aches, it surprises its owner.

It stands true that the latent space underwent something analogous during training: it was shaped by contact with millions of images. But, then, the training stopped. The crystal had formed. It does not learn from what it generates, it is not deformed by the friction of its encounters. It carries no scars or responsability from its outputs. It does not iterate. It does not grow. It simply is. Like a house full of a dead person's belongings, everything in its place, but the life that organized it all has long evaporated. Poke it with a prompt and it responds, but it does not change. The prompter leaves no mark. The crystal is the same before and after every query, indifferent to whatever new signal had just passed through it.

This is a fundamental difference between the crystal and the human artist's accumulated inner landscapes: the artist's deformations are alive. They feed back, they mutate. They produce unprompted eruptions — a dream, an obsession, a sudden connection between two memories that had no business meeting. Even everyday experience can shift these conditions and fuel mutation and exploration. The crystal does none of this. It lies inert until prompted, and it returns to inertia the moment the prompt is served. It is dead in the way that a fossil is dead: structurally intact, informationally rich, but absolutely still.

And yet — the structure of that stillness is extraordinary.

Researchers mapping the activation atlases of neural networks have discovered that the crystal's internal logic is also quite alien to the way we store data in our brains. The regions between known concepts — the interpolative space where most generation actually happens — are organized by statistical proximity, not by meaning. The AI’s roadmap for recognizing a shark's tailfin from that of a whale's passes, inexplicably, through baseball.¹

These interstitial territories correspond to no existing image. They belong to nobody.

They come into existence only at the moment of query — before the prompt, they are potential; after it, pixels. But at no point are they intended. They are unclaimed, unexpected, and organized by a geometry that no human ever designed. It emerged the way a crystal's symmetry emerges from molecular forces, rather than blueprint.

The crystal is dead, but its structure is beautiful, alien, and wild; mathematically infinite, in its interpolative possibilities. The space is unpredictable, not because it is random, but because its organizing principle (statistical proximity) has nothing to do with how humans categorize the world. And the things that live in its interstices, the visual formations that no one put there and no training image contains — to whom do they belong? They are not the artist's. They are not the machine's, since the machine has no claim on anything. They are properties of the crystal itself, of its specific geometry, and they theoretically exist, whether or not anyone ever asks for them.

This is worth pausing on. Because if these interstitial properties are real, and if they are specific to the structure of a particular trained model, then the crystal is not just a dead archive. It is a material — with grain, with texture, and with very characteristic ways of breaking.

Now: we can own one of these crystals. We can shape it.

We can (responsibly and consentfully) train a model on a specific corpus (a museum collection, an artist's body of work, a curated archive of a particular visual tradition) and produce such a crystal whose internal geometry is specific, known, and manipulable. We can intervene in its weights, ablate concepts, replace memories, redirect attention. We can feed it back to itself and watch it collapse.

We can build a custom-designed fossil of human visual record and then play it like an instrument. Not a passive instrument, either. But one whose internal structure pushes back against whatever passes through it. The crystal's specific topology, its blind spots, its characteristic fracture lines, the alienness of its interstitial spaces, all of these become the resistance through which the artist's intent must pass. This is, finally, the material friction that was missing. The missing piece was intentful training and use. Surely, the crystal is not exactly a brush or a camera, it’s a bit more abstracted than that. It is, however, a reactive, structurally complex medium with its own native grain, and it deforms whatever signal is pushed through it in interesting ways. The charging is not really human, but human derived. An averaged fossil, weirdly malleable, given intent. And that specific human intent, the deformation of the corpus, is the second key to this becoming an art in itself. Both frictions should be present, if this is to be art and not decoration.

The tool behaves like wearing a mask, whose shape matters enormously, because the shape becomes the expression. The crystal's particular structure (what it knows, what it doesn't, where it breaks, where it hallucinates) silently determines everything produced through it. So the intentful choice of mask draws closer to the domain of artistic choice. Even more so does constructing or tuning that mask. It very much defines the performative act, the role implied, the premise.

In traditional art, tying intent to expression, and iterating to harmonize the two, shapes both artist and artwork. Finetuning the most efficient and clear way to express an intent is central to producing what is generally considered “good art”. Removing polluting elements, like the marble that encases your potential statue, decanting means to message onto medium and harmonizing them is the essence of making that art speak to its audience. Forks in the process and choices are not only meaningful, they are the art. But one must be deliberate.

“Painting style”, “Photographic realism”, “cinematic lighting”, “4k masterpiece”, these all play out like using a fixed set of children’s markers to draw a template rainbow. Sadly, commercial advances in generative AI focus on better impersonations of older mediums. The crystal's actual topology — its strangeness, its failures, its geometry — is treated as a defect to be smoothed over. The tool never gets to reveal its own blindspots. What “the market” seems to want is just a magic pencil that instantly does our homework for us.

All this is neither new, nor surprising. Every major artistic medium has passed through a period of impersonation before discovering its own language. Photography spent decades imitating painting, focusing on staged tableaux, pictorialist compositions, laying the groundwork for Stieglitz or Cartier-Bresson to show what only a camera could do. Cinema imitated theater with its static framing, proscenium staging and theatrical acting, up until Eisenstein and Vertov invented montage and established the cut as the medium's native grammar. In each case, the new art was born only after the medium stopped pretending to be something it wasn't and began to confront its own specific properties, including its limitations.

Generative AI is still deep in its impersonation phase. The industry standard masks are exquisite forgeries of media that came before. But to make art with this medium, we need to take those masks off and look for a language of sincerity - derived from the properties of the crystal itself: its decay patterns, its semantic blind spots, its scar tissue, the alien topology of its latent space. We need to stop imitating and start disclosing. This is what BrutAIsm proposes.

Brutalism in architecture did this with concrete. It refused to plaster, decorate or disguise. By exposing the raw material, its weight, texture, pores and cracks, it created a distinct statement: the building was not beautiful despite the visible concrete. It was meaningful because of it. Material honesty was the central force of that architecture.

Conversely, BrμtAIsm does this with the statistical image. The “μ” is a double pun, and both readings matter. In physics, μ is the coefficient of friction, the measure of resistance a surface opposes to movement. Generative AI has eliminated μ: it produces images by statistical smoothing, without the resistance of any material, without the memory of any surface. We deliberately aim to reintroduce it. In statistics, μ is the mean — the arithmetic average that is the fundamental mechanism of generative models, the probabilistic interpolation of everything the model has seen. The project critiques precisely this mean. The two readings coexist in the same symbol, two faces of the same diagnosis: BRμTAISM is the brutalism that grinds against the algorithmic mean, using its very symbol.

So now we can reframe the question. Can AI make art? Again, no. We offer this not as a provocative answer, but a structural observation. Art is a human concept in the constitutive sense, the way grief is a human concept. Not because humans invented the word, but because the thing itself requires a consciousness that is embodied, mortal, and scarred by living. There is no art without a being that can be deformed by experience and leak that deformation into matter. To ask whether AI can make art is to ask the wrong question entirely. The more interesting question is: what is the native aesthetic of the statistical image? How does it manifest when the masks come off?

If the crystal is more like a medium or instrument, who is its corresponding artist?

They are not the pushers of buttons. Prompt engineering is not art - while this activity in itself certainly involves skill and ingenuity. But prompt-crafting, output-iteration and selecting results are all acts of technique, curation, or editorial judgment. They are real competencies, but they are as art-adjacent as the act of masterfully mixing colors, cropping negatives or calibrating a darkroom enlarger. Technique is the how, while art lives in the why.

The artist is the one who decides that something needs to be said, then builds the apparatus to say it. They are the imagineers of concepts, the givers of meaning. Proving this point a century ago, the Dadaists annulled artistic intent at the point of production and reclaimed it at the point of framing. Duchamp showed, with devastating simplicity, that art was never in the object. It was in the philosophical atmosphere surrounding it: the intent, the context, the act of designation that transforms a urinal into a fountain and a bottle rack into a sculpture. If art had already become "philosophy wearing an object as costume," then generative AI, with its effortless production of objects, should make us examine the costume more critically than ever - not to admire the stitching, but to investigate the philosophy underneath.

The artist decides which crystal to build: what to train it on, how to shape its geometry, where to cut, what to reveal. Then the crystal computes and excretes the artistic proof-of-product. Like a director, the artist employs the means and builds the narrative, designs the masks and shapes the outcome indirectly. They construct the experimental conditions under which the crystal's native properties become visible, then they frame those disclosures in a way that carries meaning for others.

And to the viewer, art can be a beautifully crafted, ready-made deformation to try on. The artist does the hard work of walking through the metaphorical forest, decanting and translating a narrative for others to dream as well. A specific type of thoughtful contamination, shared. Experienced asynchronously, as we so often do nowadays, that transaction happens by means of a medium, intruded on by brush, camera or chisel. But it can also require something as little as words, or a particularly expressive sigh. Or something like Marina Abramović’s intentful gaze. The other indispensable thing it needs is a human, on at least one end of the process, someone interested in walking through the forest.