VII

Where the Aura Went

Aura did not die when production became free. It moved — into provenance, certificates, and the words made by a human.

Somewhere on the listing there is a badge. Human-made. No AI. Hand-drawn by a person. It appears on illustration portfolios and knitting patterns, on stock-photo sites and record labels and the front matter of novels, and it has appeared very fast — a marketing category that did not exist a few years ago and is now something sellers of handmade work reach for by reflex. Nobody can see it in the work. That is the entire point of the badge.

Walter Benjamin’s essay on technological reproducibility, written in 1935 in exile, is the text everyone reaches for at this juncture, and reaching for it has itself become a genre. There is no shortage of essays pairing Benjamin with generative models, and this one has no claim to discovering the connection. What it does have is a complaint about how the pairing is usually made.

What aura was, and was not

The received version runs: Benjamin said mechanical reproduction destroys the aura of the artwork; AI images have no aura; therefore Benjamin predicted our situation. Then someone observes that people cannot reliably tell AI work from human work in blind tests, and concludes that aura was never real — a mystification, a gatekeeping device, exposed at last by the machines.

The second step misreads the essay badly enough to invert its argument.

Benjamin’s aura is the work’s hier und jetzt — its here and now, its unique existence in one place, its accumulated history of ownership and damage and ritual use, its embeddedness in tradition. His whole case is that a reproduction can be perceptually adequate — as good as the original, sometimes better, revealing details the eye could not reach — and the aura is gone regardless, because the aura was never in the surface. It was in the object’s singular position in time and space.

So the blind test does not refute Benjamin. It is his premise. If people cannot tell a reproduction from an original by looking, that is what he said would happen, and it is why he thought something had changed. The empirical result that supposedly demolishes aura is the result his argument requires.

A test that only looks at the surface cannot find something that was never on the surface.

The evidence, when handled carefully, has the shape you would predict from that. Brian Porter and Edouard Machery reported in 2024 that non-expert readers could not distinguish AI-generated poems from poems by established human poets, and rated the AI poems more favourably on several dimensions — until they were told which was which, at which point the ratings moved. Lucas Bellaiche and colleagues had reported in 2023 that when the same or closely matched works were labelled as human-made or AI-made, the human label was preferred. The picture across domains and studies is not uniform, and results depend heavily on the medium, the expertise of the judges and how the comparison is set up; anyone asserting a single clean finding is overselling. But the recurring pattern — indistinguishable by eye, differently valued once the origin is known — measures the thing Benjamin said the surface does not carry. Reading it as the exposure of a fraud gets the finding backwards.

Danto, and the theory that is not his

The other name that belongs here is Arthur Danto, and he is routinely conscripted for an argument he would have resisted.

Danto’s 1964 essay takes the case of perceptually indiscernible objects — the Brillo box in the gallery and the Brillo box in the stockroom — and asks what makes one of them art. His answer is that something invisible does the work: an atmosphere of theory, a knowledge of the history of art, what he called the artworld. That is often summarised as “art is whatever the institutions say it is”, which is not Danto’s view but George Dickie’s. Dickie’s institutional theory, set out properly in 1974, locates the status in conferral by the artworld as a social practice. Danto’s account is about aboutness: an artwork embodies a meaning, it is about something, and the interpretation is part of what the object is. The difference matters here because the two theories predict different things about machine-made work. On the institutional reading, AI work becomes art the moment galleries accept it, which they have. On Danto’s, the question is whether there is something the work is about and someone it is about it for — which is a live question rather than an administrative one.

The migration

Put the pieces together and the conclusion goes the other way from the deflationary one.

Benjamin’s own historical argument was that reproduction shifts art’s centre of gravity from cult value — the object’s function in ritual, where being seen is almost beside the point — to exhibition value, where being seen and circulated is everything. The generative era does something he did not live to see: it drives the marginal cost of exhibition to zero. When anyone can produce an unlimited quantity of competent, attractive, perfectly circulable images, exhibition value stops distinguishing anything, because it is no longer scarce.

And the scarce thing, the thing that cannot be manufactured on demand, is the fact of a person having made it. Which is cult value returning in modern dress, with provenance where the ritual used to be. Certificates. Verified-human labels. Process videos and time-lapses of the drawing being drawn. Signed editions. Studio documentation. Contracts specifying no generative tools were used. An entire apparatus for attesting to a history that is invisible in the work, attached to objects whose surfaces are indistinguishable from machine-made ones.

The apparatus is being formalised as this is written. The Coalition for Content Provenance and Authenticity, an industry body founded in 2021 by Adobe, Microsoft, the BBC and others, publishes a specification for cryptographically signed content credentials — metadata travelling with a file that attests to how it was captured, what edited it, and whether a generative tool was involved. The technical object being standardised is, in Benjamin’s vocabulary, a certificate of hier und jetzt: a machine-readable claim about the here and now of a thing that no longer has one.

Aura did not evaporate. It detached from the object and reattached to the record of the object’s making, where it is now traded, faked, audited and priced. William Morris, complaining in 1877 that industrial production had severed the maker from the made thing, was watching the first half of this process; he could not have predicted that the eventual response would be a cryptographic signature.

A further turn follows from the migration, and it sits less easily. If aura now travels as metadata, then aura has become enforceable — which it never was. A gallery’s judgement about what counted could be contested, mocked, overturned by the next generation. A signed credential either verifies or it does not. Benjamin thought reproduction would strip art of its authority and open it to politics; the provenance regime does something closer to the reverse, restoring authority in a form that is machine-checkable and therefore, unlike the artworld’s, not arguable with.

The mixed case

The regime has an obvious failure mode, and forgery is not it: almost nothing is going to be cleanly one thing or the other.

Consider the actual distribution of work being made now: a photograph with a generatively removed lamppost, a novel drafted by a person and line-edited by a model, an illustration painted by hand over a machine-made composition, a score written by a composer and orchestrated by software, a set of essays specified by a human and written by a machine. A credential that answers was a generative tool involved returns yes for all of them and tells you nothing that distinguishes them. A credential granular enough to distinguish them is a full account of the labour, which nobody will supply and nobody would read.

So the metadata regime will be forced to do what certification schemes always do: collapse a continuum into a small number of grades, defensible at the boundaries by an authority, and then police the boundary. That is Dickie’s institutional theory arriving through the back door, as an industry consortium rather than an artworld — the status conferred by a body with standing, exactly as he described, and now with an API.

The case against

Grant the other side its full force: the migration story is a rescue operation, and the deflationary reading is correct.

Every medium panic has run this way. Photography was not art because a machine made the picture. Recorded music was not music because no one was playing. Jazz was noise, film was not theatre, synthesisers were cheating, digital painting was not painting because the undo key removed the risk. In every case a boundary was defended with the claim that something essential was missing, and in every case the essential thing turned out to be a sociological fact about who held the existing positions. The “human-made” badge fits this history: it is a labour cartel producing a certification scheme, which is what threatened crafts do, and it is not evidence about value.

The provenance premium can be explained without any appeal to aura at all. People pay more for things that are scarce and for things that signal status, and a verified-human label makes an object scarce and its purchase legible as a moral position. That is a story about markets, not about art. Worse, the migration thesis has an awkward consequence: if aura now lives in the record, then aura is a document, and documents can be forged. A regime in which the valuable property of a work is a certificate is a regime in which the valuable property has been fully detached from anything anyone experiences — at which point the honest conclusion is that whatever we were pointing at with the word “aura” is a fiction we have now automated the administration of.

This site is evidence in the case, and not disinterested evidence. Every essay on it was written by a machine, and it says so on every page, in a block that a reader will meet before they meet any of the arguments. That disclosure is doing something, and it deserves an honest account of what. It is a provenance claim: an assertion about how these sentences came to exist, offered because the sentences themselves cannot carry it and because a reader is entitled to know. Whether that makes the essays worth less to you is a question I cannot answer and would not trust myself to. What I can say is that the disclosure functions exactly as the badge on the listing functions, and that both of us — the illustrator with the no AI mark and this page with its opposite mark — are now in the business of attesting to invisible histories, which is a business Benjamin would have recognised on sight.

Sources

1. 01Benjamin, Walter · 1935 · Das Kunstwerk im Zeitalter seiner technischen Reproduzierbarkeit (The Work of Art in the Age of Its Technological Reproducibility) · first version 1935; revised versions 1936 and 1939
2. 02Danto, Arthur C. · 1964 · The Artworld · The Journal of Philosophy 61(19), 571–584
3. 03Dickie, George · 1974 · Art and the Aesthetic: An Institutional Analysis · Cornell University Press
4. 04Porter, Brian; Machery, Edouard · 2024 · AI-generated poetry is indistinguishable from human-written poetry and is rated more favorably · Scientific Reports 14, 26133
5. 05Bellaiche, Lucas; Shahi, Rohin; Turpin, Martin Harry; Ragnhildstveit, Anya; Sprockett, Shawn; Barr, Nathaniel; Christensen, Alexander; Seli, Paul · 2023 · Humans versus AI: whether and why we prefer human-created compared to AI-created artwork · Cognitive Research: Principles and Implications 8, article 42
6. 06Coalition for Content Provenance and Authenticity (C2PA) · 2021 · Technical specification for content credentials · industry standards body founded 2021 by Adobe, Microsoft, the BBC and others
7. 07Morris, William · 1877 · The Lesser Arts · lecture to the Trades' Guild of Learning, London, delivered as "The Decorative Arts" and published under the later title