A study from Columbia University, published in June 2025, quantified what many of us had intuited. A group of individuals was shown identical works, some labelled as human-made and others as generated by artificial intelligence. The latter were valued sixty-two per cent lower than the former. The same image, depending on the label it bore, was worth less than half. And there was a revealing detail: human art only gained value when compared with machine art. Alongside other human works, the effect dissipated. It is not that we recognise the human by its virtues. It is that we appreciate it more when the threat of the automatic emerges.
This fact has fuelled an understandable reaction, one that asserts what the machine should not possess: the hand, the gesture, the error, the biography, the trace of the body that laboured. The photographer Rankin articulated this with a fortunate formula when discussing his own images reprocessed with AI: even when transformed, he noted, there remains within them “a trace of me”. The idea circulates, is repeated, and becomes almost a slogan: the authentic is that which retains the human imprint. Against the seamless polish of the algorithm, the tremor of the hand serves as proof of truth.
I understand the impulse and share the concern that drives it. Yet I believe this criterion is weak, and that relying upon it leaves us in a worse position than before. It is weak for a simple technical reason: imperfection proves nothing. A generative model can produce, at will, crooked lines, paper grain, asymmetries, brush textures, and all the accidents we associate with the hand. The tremor is imitable. If tomorrow we place authenticity in the defect, the day after tomorrow the defect will come as standard, perfectly manufactured. We shall be purchasing synthetic human imprints and calling it soul. Whoever defends the hand as the ultimate criterion has not found the impregnable redoubt of the human; they have merely pointed out the next effect the machine will learn to produce.
Furthermore, it is weak for a deeper reason that touches upon my understanding of art. If a work is valuable because it bears the marks of having been made by hand, then we are asserting that art is a symptom: the sign that a human was behind it, their expressive signature, the trace of their psyche. Yet this does not explain what the work does; it merely certifies who touched it. It reduces the painting to proof of presence, like a fingerprint at a crime scene. The art that matters to me is not proof that someone was there. It is something that occurs before the viewer, whoever may have produced it.
Herein lies the point I consider decisive. The true debate is not the one being announced: machine perfection versus human imperfection. That is a decoy, and, moreover, one the machine will win as soon as it sets its mind to it. The real debate, the one with substance, is between two ways of understanding where the value of a work originates. One derives it from provenance: who made it, with what intention, with what biography, and what responsibility behind it. The other derives it from the process: what was selected, how it was organised, what was placed in relation, and who assumes the result. And this second approach is not a surrender to technology. It is, quite simply, how a significant portion of the art of the last century has functioned.
For none of this is novel. In 1859, when photography infiltrated the galleries adjacent to the Paris Salon, Baudelaire wrote in alarm against the notion that an image produced by a machine could belong to the domain of art. It appeared to him that mechanical precision would displace the imagination, that we would confuse recording with creation. A century and a half later, we know how it concluded: photography did not destroy painting nor supplant the imagination; it became an art form with its own questions, and in the process, compelled painting to ascertain what it could achieve that the camera could not. The panic was understandable and the diagnosis, false. Each time a new machine produces images, we repeat the scene: first, the rejection in the name of the human, then the gradual realisation that the question was poorly posed.
I do not suggest that provenance is unimportant. It matters, and significantly so: knowing who made a work, with what labour, and at what cost is a matter of justice, especially when there are models trained on the work of others without permission or payment. That is a legitimate and urgent battle, and I do not conflate it with the other. What I maintain is that provenance is a matter of rights and honesty, not the criterion that determines whether a work is art. These are two distinct questions, and conflating them causes us to lose both: we convert a real labour issue into a metaphysical discussion that cannot be won, and we leave metaphysics in the hands of those who only need to manufacture a convincing tremor.
The hand trembles, and that tremor is beautiful, and it deserves to be remunerated and recognised. Yet that is not where the determination of a work of art is made. The criterion does not reside in the hand. It resides in what the work effects upon the observer, and that is a question we have left unresolved for centuries, one which no machine, as of yet, has assisted us in answering. Nor has it removed it from us.
On open conversation
This text intersects with one of the lines of my research: the proposition that what causes a work to function as art does not depend upon the agent that produces it, nor the medium with which it is produced. Hence, I align myself neither with those who defend the human trace as proof of authenticity, nor with those who celebrate algorithmic production without reservation. I do, however, distinguish between two questions that are frequently conflated: that of the rights of the worker, which I defend without qualification, and that of what constitutes art, which is not resolved by pointing to a hand. If anyone wishes to contribute from the fields of illustration, criticism, the philosophy of art, or practice with generative tools, this notebook remains open.
Sources
Horton, Carl Blaine Jr., Michael W. White, and Sheena Iyengar, on the differential valuation of art according to its attributed authorship (human or AI); institutional summary from Columbia Business School, 29 June 2025.
Rankin, on authenticity and “the human trace”; interview in Pentatonic, 25 November 2025.
Baudelaire, Charles. “Le public moderne et la photographie” (1859), in the wake of the admission of photography to the galleries adjacent to the Paris Salon.
Regarding the incorporation of AI art into the market and institutions: Christie’s, Augmented Intelligence (2025); MoMA, Unsupervised by Refik Anadol (acquired in 2023); and the open letter from artists against the auction, reported by The Art Newspaper (2025).