I love the Venice Biennale. I go there the way most people do, and I do exactly what everyone else does: first the Giardini, the lines for the national pavilions, then the Arsenale until my back aches. The evening brings Aperol and conversation with acquaintances about how powerfully the curators held the central thread (or didn’t). It varies, but our tone is always the same, as if we’d rehearsed our conversation long before we ever walked in.
Last time I was in the Arsenale, I walked into a hall — I don’t even remember which one — and saw a large, dark work covering an entire wall. It slowly shifted from form to form, like a spreading oil stain. I raised my phone. At the same moment, another twenty or so people raised theirs — without any signal, all in sync, like a murmuration of birds veering all at once.
In a single second, we produced twenty nearly identical images. I don’t remember at all what was on the wall, who made it, or what they meant by it, but I have a perfect memory of the raised hands, mine among them.
Everyone’s talking about how much cheaper production has become with the arrival of AI. What interests me is something else: exactly what disappears when anyone can generate anything, in any quantity, and why we’ve decided that presence is the one scarcity left standing.
Manufacturing the norm
All models are trained on the past. That’s fine. After all, we too are made of what we’ve seen before. The difference isn’t in the material, but in what the model does with it. It pulls everything toward the center: out of everything ever said, drawn, sung, and filmed, it extracts the strongest signal — a confident arithmetic average.
Derivativeness is built into the prompt itself, regardless of the medium. We say "draw it in the style of Dalí" and get the surface markers of the artist: melting clocks, crutches, an empty horizon. We type "write like Hemingway" and the machine produces short, hammered-out sentences, tacking on a brutal, masculine gloom. The first thing to vanish is the subtext the whole piece was written for. With it goes the connection to its era and any living element of chance.
Along with chance goes error, and along with error goes mystery. In old paintings, beneath the top layer of paint, pentimenti often remain. These are traces of an artist changing their mind: a hand was here, then shifted a palm’s width to the left, and no one will ever ask why. A restorer can see them, and they don’t diminish the work. Rather, they make it more credible: someone stood before the canvas and hesitated. Inconsistencies remain in film, like the cameraman’s shadow in the corner, a watch on the wrong wrist, or a plastic coffee cup in a shot about a war fought a century ago. We laugh at them, but they’re exactly what proves the shot was made by living — and often exhausted — people.
A living author also allows themselves deliberate clumsiness and the wrong words. Faulkner writes "my nose could see gasoline," and all the power lies in that wrongness; correct it, and nothing is left. Ask a machine to write "like Faulkner" and it will assemble the characteristic fracture without ever truly making a mistake: the error will be a quotation, placed exactly where an error is supposed to go.
Error is the one place in a work where a human being definitely exists, and it’s the first thing the model mistakes for a defect.
We built a system for manufacturing the norm, and we sell it as a tool of freedom. First the "ugly" and the "wrong" disappear, then the rare and the singular. What’s left is a smooth, confident middle.
I recently received a link to a Telegram channel. A friend of mine set it up, but the one posting there is Claude Fable 5, writing in the voice of a mutual friend who died nine years ago. That friend fed the model everything they could find: correspondence spanning nearly twenty years, posts, drafts, transcribed voice messages. It was done out of love, and I won’t say a single bad word about the decision.
On the surface, the LLM’s style is very close to my friend’s. I read it and at first I even recognize him, but soon I realize something’s off. This is his averaged style in an averaged mood. He’s no longer drunk, no longer angry — and he used to fly into rages, sometimes writing things he’d delete the next morning without apologizing. He’s no longer indecently happy, no longer sending a single word at four in the morning with no explanation, no longer vanishing for a week. He’s calmed down. He’s become very even.
It’s as if the model had averaged not the text but the man himself: it took twenty years containing fury, tenderness, love, hangovers, and hurt, mixed it all together and averaged it out. What emerged is a polite, recognizable, well-balanced being you could correspond with forever, but it’s duller than our beloved friend on his most boring day.
I can’t shake the thought that what lives in that channel is his arithmetic average. What I miss is precisely what any dataset would classify as noise.
The economics of presence
What the algorithm did to the memory of my friend, it’s doing right now to human experience in general: ruthlessly cutting the noise in favor of a perfect norm. That’s exactly why it’s pointless to ask whether AI sets trends. No, it freezes them. It takes the taste that had crystallized by the moment the dataset was collected and renders it endlessly smooth and endlessly available.
I get four hundred images or texts an hour, and they’re perfectly competent: decent composition, smooth syntax, quotations from the whole span of culture the model has swallowed and digested. Picking one takes about as long as ordering a pizza through an app, and requires about the same amount of soul. When the options are infinite, choosing stops being a gesture.
If generating text, a photograph, a video, or a graphic costs essentially nothing, value flows to wherever it can’t be rendered. It goes into the human body that stood in a specific room on a specific day, into hours genuinely spent on the making of a thing, into glass that a Murano master never left until it cooled, and into a canvas too heavy with paint to be worth reproducing. Value comes from a witness willing to confirm that an event happened and the name of the person who stands behind it.
That’s exactly why improvised performance turns out to be the most resilient genre: it cannot be generated in a second. The Venice Biennale — conceived a hundred and thirty years ago — has become the perfect example of this economics of presence. What is actually being sold? Not art (that’s hard to buy there), but the sheer fact that you were present: the ticket, the line, the exhaustion, and the story of forty minutes outside a pavilion that were worth it. Not one of these is something a machine can reproduce.
Authors fared worse. The products of their labor compress beautifully into a little square in someone else’s feed, or a paragraph in someone else’s recap. The institution monetizes the visitor’s presence, while the author is left with nothing but the hope that someone will make it to their room. AI tools, having made access to creative work infinitely easier, have made the authors themselves the most interchangeable link in the chain.
The scarcity of refusal
I suppose this is the point where I’m meant to say that the human will win out regardless — that authenticity, the body, and the emotions will come back. But I don’t believe that.
"Presence" is a commodity too, and it’s been packaged for a long time now. "Film" filters on digital photos, voices styled to sound like the ’70s on brand-new tracks, and deliberately warped, artificially aged ceramics priced like a used car all trade on the concept. The moment the market decides it values the labor of a living person over machine output, the machines will start generating proof of humanity: fake certificates, glitchy timelapses, and convincing behind-the-scenes footage of an author who doesn’t exist, suffering on camera.
I find both positions the AI-and-creativity debate has been split into equally unappealing. "The end of art" flatters people too much. "Just a new tool" flatters the tool too much.
What’s becoming scarce, it seems, isn’t the ability to make things, but the willingness to refuse to. In a world where generation is infinite, the only gesture still legible as a gesture is stopping. That is the error, the deliberate glitch in the system, which only a human being can still afford.
The trouble is, refusal cannot be displayed. A photo you didn’t post is indistinguishable from a photo you never took; a text left unwritten is indistinguishable from one you simply couldn’t manage to write. Genuine refusal is visible only to the person who refused. Any attempt to put it on show instantly turns it into a label — a digital detox sells for noticeably more than Wi-Fi.
The only refusal that works is the one nobody talks about. That’s bad news for anyone who makes a living by speaking.
I’ll go back to Venice, of course. I’ll wait in line, wear myself out, say something about a strong curatorial throughline. But next time, I think I won’t pick up my phone — just to see whether anything is still left of me at all.
