Técni painted a wall outside a middle-school cafeteria. A kid named Gina read it and decided to try — and in canon, that changed the rest of her life. So here is the question this lab exists to ask: if a machine had painted that wall, and Gina had read it, and she still walked into the room — did the mural do the same thing?
Gina read it and decided to try. That mural changed a kid’s next ten minutes and, in canon, the rest of her life. Now hold that fact still and ask the machine question against it. It is not rhetorical. It has an empirical part that has been tested, a philosophical part that has not been settled, and a hinge between them most people never notice they’re standing on.
Técni’s wall is not signed with a name — where Técni signs at all, it’s a single initial, “— T.” You know it’s theirs the way you know a tag: by the style, not the signature. And the canon is explicit that the wall was contested: “teachers argued about whether it was vandalism; kids took pictures of it like it was treasure.” Same paint, two verdicts — and the split is about authorization, not about the image. It worked on Gina either way.
That is the whole lab in one wall. Its author is knowable. Its authorization is disputed. And its effect was real regardless of both. Hold onto that when the AI question arrives — because the AI question is not new. It’s the oldest question in art, arriving at a scale where it can no longer be treated as an edge case about auction houses.
“Same can of paint. On a gallery wall it’s a commission; on a memorial it’s a crime; on a school wall it’s Técni. I’ve been the kid they arrested and the man they hired — for the same move. So don’t ask me whether the machine’s mural is real art. Ask who’s standing next to it when they decide — and what they already believed before they looked. That’s the whole game, and it was the game long before the machine showed up.”
● sourced — a fact or a finding with a citation you can check (Tab 5 lists them; two were verified at primary for this build).
◐ argument — the lab’s own move, not a finding. Flagged as argument so you can attack it. The strongest claim here is deliberately the most attackable one.
◯ contested — genuinely open in the field. The lab names these and does not pretend to resolve them.
Walk these in order. At each step, guess: does art survive this one? Then open it. Four times someone alive for the change called it the end. Then the fifth arrives — and you have to explain why this one is different, using the tools the first four just handed you. ● sourced the sequence is uncontroversial art history.
This is the origin. Nothing prior to be threatened.
The work becomes durable and findable by strangers — the first time a maker’s mark could outlive the maker and be read by someone who never met them.
The work becomes copyable without the maker. The first real panic — and it came from inside the scriptorium.
The work becomes executable by a machine — the picture arrives without a brush. The camera was the first shock; the plotter, Photoshop and the CRT are its descendants.
Here is the one that’s different, and this tab’s job is to make you say why. The press changed the copy. The machine changed the tool. The model changes the origin — it generates candidate originals from a compression of millions of prior ones. That’s the actual discontinuity candidate.
Nelson Goodman (Languages of Art, 1968) drew the line the press revealed. Autographic: the object’s physical history is the work — a perfect forgery of a painting is not the painting. Allographic: any correct instance is the work — a printed Hamlet is not a lesser Hamlet; tonight’s performance is the symphony. The press didn’t kill art; it moved text decisively into the allographic, and Walter Benjamin (1935) argued the same shift stripped the “aura” — and thought that was partly liberating. Keep this split in your hand. The next tab tries to break it. ● sourced
The cleanest case against AI art is the origin move: a model output has no single maker, and that absence is the loss. Here is the counter that kills the romantic form of it — and it’s a move worth making against your own argument before someone else does.
Hominin 2427 walks through White Sands, sees marks on a rock wall, walks for months to western Arizona, and scrapes his own version from memory. No name survives. No chain of authorship you could ever trace. Has origination loss happened there too?
If untraceable origin disqualified art, most art ever made is disqualified. Anonymity is the normal condition of art history, not the exception: unsigned cathedral carvings, icon painters, guild workshops, the entire Paleolithic corpus. The named individual artist is a late European invention. ● sourced — and motif recurrence across continents is real, though ◯ contested whether by diffusion or independent invention, so the lab states the pattern and not the cause.
But watch what the counter actually collapses — two different things wearing one word:
| Origin unknown | Origin absent | |
|---|---|---|
| Hominin 2427 | ✓ | — |
| Model output | — | ✓ |
| What it is | One first maker existed. He was somewhere. The record is lost — the chain is real. | Not a chain that was lost. One that was never singular. A weighted average over millions of hands. |
Anonymity ≠ authorlessness.
2427 didn’t copy. He saw it, walked for months, and what came out was deformed by his hands, his rock, his memory, and whatever he needed it to do. The thing passed through a life and came out changed. A model is not changed by what it processes — weights are frozen at inference. This survives White Sands where the origin argument doesn’t. It is also more vulnerable to the charge of romanticism — which is exactly why it’s the right thesis to build on. ◐ argument — “weights frozen at inference” is technically true of standard deployed models; the aesthetic conclusion drawn from it is the lab’s, not a finding.
To the viewer at the rock wall, the difference between unknown and absent origin is zero — you can never access either. So why should the mere existence of a lost origin matter to your experience? That is the forgery question in its purest form, and the lab must not pretend to answer it. It must name it, because it’s the hinge. The next tab is where you feel it on yourself.
Below are two panels — the same image, twice. Give each an honest score. Then hit reveal. Do the rating first: if you read the studies before you rate, you’ll perform the answer instead of producing it.
If your score for B dropped after the reveal — nothing about the image changed. Only what you were told about it. You just replicated, on yourself, the finding four labs have measured. The load-bearing claim rests on the two behavioral studies, not the brain scans.
The Mona-Lisa thought experiment, run for real across five experiments. The gap between an original and a perfect duplicate runs on two mechanisms: performance (the object judged as a unique creative act) and contagion (physical contact with the maker). The gap comes from people’s lay theories about the domain of art itself — not from any particular art type or known forgery case.
They randomly stuck “Human-created” or “AI-created” labels on paintings that were all actually AI-made. Study 1: human-labelled art scored higher on every criterion — Liking, Beauty, Profundity, Worth. Participants were generally incapable of telling AI from human. The proposed engine: the perceived imbuement of human experience, and artistic effort read as a marker of quality.
Study 1 found the anti-computer-art bias. Study 2 reversed it — not merely shrank it — when observers watched robotic artists in action, driven by inferred anthropomorphism and the effort heuristic (art that looks like it took effort scores higher). The anti-AI penalty is not fixed. ◐ one caveat the study itself states: they couldn’t contrast watching a live human artist against the robot, and the anthropomorphism effect is correlational — so: reversed when the machine was seen working, not a clean causal law.
Randomly label identical images “gallery” vs “computer-generated” and aesthetic ratings and mOFC/prefrontal response rise for “gallery” (Kirk). Label Rembrandts “copy” vs “authentic” independent of the truth, and the brain tracks the told status, not the real one (Huang). Caveat, stated not buried: single-study fMRI from a field with real replication problems. These are illustrations. Delete both and the thesis must still stand on Newman & Bloom and Bellaiche alone — it does. ◯ contested
Everything so far establishes one thing: the drop you felt when B became “the copy” is real, measurable, and older than the technology that provoked this argument. What it means is the fork — and honest people stand on both prongs.
The reaction comes from lay theories about the domain of art (Newman & Bloom) — which is compatible with it being right. Maybe art is partly testimony that a specific life was here, and “it produced the same feeling” genuinely doesn’t settle it. The performance and contagion you feel may be information, not noise.
If you can never access the origin, its mere existence is doing no work in your actual experience — the difference at the rock wall is zero. The valuation drop may be a forgery-era status reflex, inherited from auction houses, that we should notice and set down. Same evidence, opposite conclusion.
The five-step spine already primed the honest answer: continuity. The AI objection may not be a new philosophical problem at all. It may be the oldest one in new clothes — the forgery question, arriving at a scale where it can no longer be filed under “edge case.” This lab names the fork. It does not pick a prong. ◯ open — Park Bench form, not a verdict.
This lab was built by someone who has made 100+ tracks with a generative model. Every question on these five tabs — origin, effort, “did it pass through a life,” does the provenance drop track something real — is one the builder cannot ask from the outside. He’s standing on the wall he’s pointing at.
That is included on purpose. It makes the lab honest and it makes it vulnerable at the same time — which is the same trade the whole lab argues is the right one: build on the version of the claim that can actually be attacked. If the drop you felt in Tab 4 is real information, some of what he made is worth less than it feels. If it’s a status reflex, it isn’t. He doesn’t get to know which either. Neither do you. That’s the hinge, and he’s on it too.
What’s real, what’s mine. The historical spine, Goodman, Benjamin, and the four studies are ● sourced. Técni, the mural, Gina and Steel City are MPC-Universe canon. The “passed through a life / weights frozen at inference” thesis is ◐ the lab’s argument, flagged so you can attack it. The motif-diffusion mechanism and the meaning of the provenance reaction are ◯ contested / open and left that way.