What follows is a guest post by G. M. Trujillo, Jr.
Jason Allen recently ignited a firestorm of controversy by winning first place in the Colorado State Fair’s digital art competition. His work, Théâtre D’opéra Spatial, is undeniably beautiful. Its color palette and composition evoke drama. Expressive brush strokes prove care for each element. And the subject—women staring out onto the countryside from a vaguely European but suggestively futuristic ballroom—invites interpretation. Maybe it is a comment on being “kept women”, the anonymous ladies secluded from nature and politics in their artificial home. Or maybe the painting is a comment on class, the wealthy women in the ballroom looking out at rustic people. But Théâtre D’opéra Spatial is a digital painting generated by AI.
Critics have taken little issue with the subject or any interpretation. But they hate that it was generated by an AI art program. To create the painting, Allen fed keywords into Midjourney’s textbox. The AI program then related the queried words to its database of art to synthesize a painting from images related to the keywords. Allen painted nothing, and he didn’t program the engine. He simply typed words into Midjourney. He probably generated many images before choosing the one that fulfilled his vision, but Midjourney assembled the pixels in the image.
Backlash to the victory of the AI-generated art was immediate. Online commenters said things like, “That’s pretty fucking shitty” and “What is this world coming to?” Even reputable sources, like The New York Times, The Atlantic, My Modern Met, and Smithsonian Magazine published articles mixed in their assessments. Some condemned outright; others heralded the birth of a new artistic medium.
Despite the hubbub, scholarly attention to the topic of AI-generated art is scarce. Arthur Miller, Galit Wellner, and Mark Coeckelbergh have written about AI art from the perspective of human and machine creativity. But most of these articles seem to miss the point, especially about pieces like Allen’s. It’s less important to focus on the “mind” of the AI program and more important to compare AI-generated art to works we already laud. We should accept AI-generated art as art for the very same reasons that we accept some of the most famous pieces in art history as art.
Assembling Pre-made Elements in Collages and Readymades
Marcel Duchamp’s Fountain might be the most important artwork of the 20th century. But all Duchamp did was take an already manufactured urinal, put it on a podium, and hastily paint a signature on the side. Readymade or found art asked audiences to consider everyday items in new ways, pointing out the artistic qualities in mundane things.
Artists such as Eduardo Paolozzi pioneered pop art by creating collages of advertisements, magazines, and other pop culture elements. Some music movements have assembled random elements too. Composers like John Cage wrote pieces like “Imaginary Landscape No. 4,” which instructed musicians to manipulate the dials of a dozen radios, allowing the randomness of the radio broadcasts in that location to determine the specific sounds of the piece. The artists of these pieces took elements that were already made (hence “readymade”), placed them into a narrative or artistic context, and constructed something more than the sum of its parts. They created readymade artworks.
In assembling elements from other pieces of art, AI art is not in principle any different from Dada, Pop Art, or other avant-garde movements. Those artists took previously made objects—urinals, periodicals, live radio broadcasts—and used them to create novel artworks. Critics who worry that AI art merely assembles parts of previous artworks would have us excise from history some of the 20th century’s most important artistic works and movements.
Minimalist Art and Concept Work
Both the MoMA and Tate Modern house minimalist artworks that are nothing but simple lines. Onement, I by Barnett Newman and Alpha-Phi by Morris Louis are striking works composed of simple lines on canvas. In typical works of abstract or minimalist art, the concept behind the piece reveals a deeper meaning. Artists rely on their knowledge of art history and on their execution to evoke emotion and meaning from a few lines, shapes, or colors. They create works that do much with little.
The AI art we are currently familiar with comes from programs like Midjourney, Dall-E, and Stable Diffusion. People using these programs feed specific words into an engine. As with traditional media, one can play around with concept, for example by changing the words in the prompt. But the process isn’t always so simple. People using AI art generators must know what descriptions to type in to get the images they envision. And they go through many iterations to find the right effects. People using Midjourney can specify not only details about the subjects (e.g., “group of women standing in front of a window”). They may also specify style or medium (e.g., “film still” or “bokeh” or “art nouveau”).
In fact, generating prompts is so complex that sites such as PromptBase have cropped up, where artists can sell expert queries that generate the desired results. Artists who work with AI even experiment with which generator to use, as they produce slightly different results, even when given the exact same query.
The upshot of this is that the concept is integral to the artwork, even in AI-generated pieces. Without understanding what to query, artists cannot create powerful works of art. Artists using AI generators must be intentional about the works they create. So, if we accept that the concept can carry the weight in minimalist art, we should accept the conceptual work behind AI-generated art.
Photography and Technology
What about the algorithms that power the AI? Isn’t that what creates the artwork, and not the artists? Hearing comments like this, I cannot help but point out that photography faced the very same criticisms.
Critics were concerned that the camera would destroy art or make it too easy. However, one need only examine images such as Alexander Rodchenko’s The Stairs, Aaron Siskind’s Jerome, Arizona 21, or anything by Diane Arbus or David Lachapelle to see how untrue that is. Subject matter, composition, and use of the tool is every bit as important as the tool itself.
Philosophers, critics, and art historians will surely have tedious conversations about AI art in the future. They will echo the tired discussions that surround modern art: why your kid could not have made what these artists did. This is because art often alludes to cultural values, engages with historical works, or pushes art into new directions. Yes, the AI art generator builds many elements of the artwork. But what creates the artwork is the artist’s query and the artist’s ultimate decision to select one image over others, just as a photographer composes many images and displays only what meets her approval. Many alarmist criticisms of AI art sound suspiciously similar to bygone alarmist criticisms of now-accepted media like photography and film.
AI art is art as much as readymades, minimalist art, or photography. A majority of AI art may be bad or in bad taste, but that is not any different from most things scribbled or sculpted in schools or studios. It is unwise to dismiss AI-generated art based on the abundance of bad works out there. Sturgeon’s Law applies here: 90% of everything is crud, including AI art. But the good pieces are every bit as good as traditional art pieces. And even though AI art generators are easy to use, they are hard to master.
What’s more, AI art fits many definitions of art. Philosophers Jerrold Levinson and Noël Carroll, for example, argue that art is made when people intend to create pieces that can be understood as art and as part of art history. AI art generators create images by analyzing and synthesizing pieces from art history and image datasets more comprehensive than any human artist has used. And the users must intentionally select prompts and images they think fit their goals. By sharing these images with others, entering them into contests, and starting critical conversations, they are doing what traditional artists do, and their art is being discussed and examined in ways broadly continuous with discussions of traditional art.
Or take philosopher and art critic Arthur Danto, who famously argued that something is art when it is a part of the artworld. Artworks are made by artists, acknowledged by critics, studied by curators, and sold in auction houses. AI-generated images do these things. They are displayed in exclusive Chelsea galleries and they are auctioned by Christie’s. Moreover, the controversy over Allen’s work sparked debates about this medium in art magazines and periodicals. While there are many more ways of defining art, the point is this: most contemporary definitions of art will be unable to exclude the works produced by artists using AI art generators.
A more interesting future awaits us once we accept AI-generated art as art. Then we can get to the provocative questions. Is the use of AI-generated art a more extreme version of Barthes’ death of the author? How does it complicate questions Foucault raised about the function of an author? Will AI-generated art democratize art and create new problems and political uses, as in the case of the reproduction of artworks? How we answer these questions has implications for important legal matters too. A recent ruling argued that AI-generated art cannot be protected with copyright because it lacks human authorship. But in September, Artist Kris Kashtanova received a copyright for their graphic novel, which uses AI generated images.
There are also ethical questions. Should we be worried that AI art generators will build in bias or reinforce harmful stereotypes? If terms such as “scholar” are correlated only with pale-skinned, masculine-presenting people, the artworks generated from that training data will inevitably be biased in their representation. Relatedly, what happens when AI meets cultural appropriation, as when FN Meka, a white artist, created a Black CGI avatar and hired a Black voice actor to rap questionable content?
AI art is art, and it is no less revolutionary than readymades, conceptual art, and photography. The sooner we get past these kneejerk reactions against it, the sooner we can get to the fascinating, novel issues it raises.
G. M. (“Boomer”) Trujillo, Jr. is Visiting Assistant Professor in the Department of Philosophy at the University of Texas at El Paso. When not teaching, he naps with his dog or daydreams about the conversations Aristotle, Diogenes, Hegel, and Camus would have while watching Star Trek or moshing at a hardcore show.