The Only Avant-Garde Position for AI Art

@mileszim, "Midjourney is getting crazy powerful—none of these are real photos, and none of the people in them exist.", tweet, 13 Jan 2023, 18:26 UTC, photo 3 of 4

"AI is the end of art."1
Depending on who makes this claim, it is either an artist's lament or a tech-enthusiast's triumph over the regime of twentieth-century "old money" which has sustained the division between art and popular culture. This division has long been uncertain—the Pop art movement of the 1960s and 1970s, or the practice of xeroxed and jpeg-compressed détournement which has outlived most contemporary art movements by 20 years directly question this divide between "high" and "low" art, between "Garbage" and "garbage."

Crucial to the impending "victory" of technology over artists (a result that the uninformed in both groups seem to have accepted as inevitable) is an understanding of art as the unending march towards an absolute "naturalism." Art is the pursuit of an Absolute Reality which can only be represented through the mediating force of the artist. The formula of art production, for some artists and for all AI-enthusiasts, is a linear process. There is a reality—affective, multipolar, uncertain, indeterminate, but ultimately sensed and legible. Within this reality sits the artist, who consumes this indeterminacy at all moments through all things. The artist's primary function, within this framework, is, oddly enough, consumption. The artist must reference, must learn, must apply techniques which are applied before and know how their own position relates to the position of all other production where they equally consume. From this interminable consumption, the artist produces a single piece—a synthesis of their consumption, which creates the virtuous cycle of the art market. The artist's next piece consumes their previous and so on, until the details of their consumption become legible in their oeuvre. In this system, the artist is nothing more than a node, transforming flows of data. They simply cut across these distinct flows. "Newness" refers not to creation, but to the unexpected syntheses of "oldness."

Attributed to Banksy, One Nation under CCTV, 2007, Newman Street, London

The construction of the artist "Banksy" is the archetypal example of this theory of art. A "Banksy" is the reconfiguration of consumer aesthetics into a negative commentary—the apotheosis of "this really says a lot about our society 🤔 😔". In One Nation under CCTV, a work attributed to "Banksy", the repurposed phrase from the American Pledge of Allegiance is subverted by replacing "God" with "CCTV" painted by a little child in a red hoodie while watched by a police officer and dog. This says a lot about our society 😔. What does it say? It could be read as a condemnation of the surveillance apparatus as the fundmental aspect of the production of citizen-subjects; it could also lament the departure of the "nation" from "God" to technology—a deeply conservative/reactionary commentary which supposes the nation as a force for good which has become corrupted. This ambiguity is completely disconnected from the value of the "Banksy." Instead, the value lies in the work's references to a categories of cultural consumption and its claims to participate in those categories critically. One Nation under CCTV takes an old concept of indoctrination (the Pledge of Allegiance) and substitutes a new form of power (CCTV) to produce something "new" and "clever." The artist has brought these together in a way that is damning of nothing in particular. This is all the artist can hope to achieve—commentary on flows of consumption.

"Banksy" as a concept is determined by the work and the work is determined by consumption. If the artist only exists as a sort of "Pro" consumer, the fundmental operation of the artist exists as a form of data-collection about reality. CCTV is no longer about surveillance, but a connection between a reality and a data—a transcription device between two forms of knowledge, optical and digital. What is consumption, in this system, if not the transcription of knowledge? The artist, as a pro-consumer, can be simulated through the intense re-combination of data.

Yet, the simulation of the artist is often incorrect. Something of reality is lost in this process of transcription. The aura of the work, certainly, but an understanding of physics and interaction as well. The simulated artist cannot understand hands, or holding, or why an object must be held—it only knows that there are hands and there are objects and they interact in certain ways spread across the dataset. The aim of AI enthusiasts has been to improve this data of interaction, to refine it until this misunderstanding of reality becomes invisible. It will only ever become invisible. The nature of its can only become more improbable to surface.

But there is something else here. AI posits a movement of art which pursues reality. It rejects the Surrealists and the Dadaists as abberations of modernity—failures to pursue the "truth" of being rather than what they were: efforts to expose the perpetual "untruth" of reality. Ironically, AI even posits a rejection of the Futurist aesthetic (but not their ideology). The arc of the universe is long, but it bends towards fascist conceptions of beauty, untrammeled truth, and Absolute Reality.

Stable Diffusion 2 Base, October 2023: "A dog in the style of Picasso."
Pablo Picasso, Still Life: Fruits and Pitcher, 1939, Oil on canvas

There is, however, a strange and cursed beauty in the mangled unreality of the AI, and only in its unreality. PROMPT: A dog in the style of Picasso is as morally reprehensible as Picasso himself, because the AI serves as its own mechanism of obfuscation. The idea is that the data which forms dog and Picasso is subsumed within the creation of their synthesis such that the enactment of that synthesis becomes invisible. The combination of dog and Picasso is supposed to be so "good" and impressive that the viewer forgets that Picasso did not produce this dog. The AI is made to imitate, to produce the worst type of inauthenticity: inauthenticity which seeks profit from its claim to be authentic while taking itself deadly seriously.

At no point should AI be understood as anything except AI. The salvation of AI rests in its non-euclidian generations of hands. These are, like JPEG compression and copy-burn, references to the conditions of production that companies and AI enthusiasts are so eager to obscure. There is nothing real about an image. These fingers refer to the impossibility of AI, to its failures and incompleteness, and, paradoxically, are more true to reality than anything else. When an AI gives a person too many fingers holding an object that upon even the slightest inspection begins to collapse, the image speaks to the distance between reality and the digital synthesis of data. They remind the viewer that they are seeing something fabricated from categories so alienated their own experience as to be virtually meaningless—a construction of an artist at least gives some ground upon which to build an interpretation of intent. The failures of AI, far more than its successes, challenge the view of artist as consumer. What separates artist and AI is not accuracy of an Absolute Reality, but the abstract unpredictability of reality that resists all manner of representation.

AI is a low-labor medium, and should be valued as a low-labor work. It is a deeply alienated form of production, and it will never be able to overcome that alienation through the imitation of less alienated forms of human work. Like photography, it must embrace a different logic—one which still refers to constructedness and failure rather than absolutes. The beauty of photography is in its approximation of the photographer. There is an immediacy which constantly refers to the photographer as a being in the scene. This is enhanced through the limitations of the form: through light leaks, less-than-real colors in film, through lens distortions and flash bulbs. These traces of artifice have become, in a digital age dedicated to their erasure, objects of fetish. Instagram and VSCO filters imitate the failures of film photography and its inaccuracies as a form of Film Chic. Low-pixel blocky JPEG compression itself is becoming an object of fond memories where such compression is either hidden more accurately or unnecessary with bandwidth improvements. These effects, added after the production of a "perfect" high-resolution (HDR, processed, and automatically re-touched) image, serve as anchors of consumable nostalgia. This ironic consumption of effects is only possible through the "perfect" high-resolution image. We can understand AI right now to be striving towards this high-resolution—to be aiming towards the ironic consumption of effects while ignoring the far more interesting failures of the form.

The possibility of AI is not its flexibility, but its error rate, its tendency to hallucinate in ways that are so completely disconnected from what we understand to be reality. Any AI art which considers itself art must engage with the beauty and irrationality of its hallucinations, since only in its hallucinations can something interesting and new exist. The flows of AI must be allowed to move in ways unmoored from aesthetics of human consumption/production towards a logic of irrational and consumption which produces nothing resembling its inputs. The horizon of AI art is a place "where the decoded flows run free, the end of the world".

If there is any sense of avant-garde in AI, it lies in the rejection and mangling of the inputs and logics which perpetuate it. Banksy will be fed to the machine of consumption he gently critiques, and out will spill a tumult of contorted hands and mouths, of unlabeled and misidentified pixel-patterns, of the chaos embedded in the negation of all things recognizable and enjoyable.

It will say nothing about society, despite containing all the data of a society.

Herein is the avant-garde of AI art, and like all weak avant-gardes, it has failed before it can even be conceived.

  1. I've been working on a piece which builds out a theory of AI using Meta's speech model Massively Multilingual Speech (MMS) model and the recordings of the bible in hundreds of languages used to train it. The goal of the piece is to describe the way that data moves through capitalist systems of exchange as a non-commodity.
    In the meantime, I wanted to write something slightly more polemical and dogmatic on AI, particularly AI images. Who knows if I will stand by this piece in even a month.↩︎