It’s the Real Thing!

Earlier this year, Donald Trump was driving through the American wasteland in his armored motorcade, furiously scrolling Truth Social, when a group of young Black men waved him down. Touched by their enthusiasm, the twice-impeached white nationalist purportedly jumped out of the car and, all smiles, agreed to take a photo. It soon began making the rounds online. “What do you think?” asked one X user, @MAGAShaggy1958, of this indisputable proof that the former and possibly future president loves Black people.

Many thought it looked fake—and it was. Cue the predictable flurry of “discourse,” the existential dread. This was not simply the Photoshopped manipulation of election cycles past but the spawn of a technology so powerful that it would soon, and perhaps irreversibly, collapse the frail wall between fantasy and reality. Democracy was doomed, and it was artificial intelligence—not billionaire oligarchs or protofascist right-wingers—that would kill it. This has become the standard pose adopted by the panic-prone media, inclined as it is to fall in awe before the power of new technology.

This isn’t altogether surprising. Companies like OpenAI and Midjourney have been happy to position their products as devastatingly effective reality generators. When it announced Sora, a text-to-video generation program, in February, OpenAI highlighted its ability to create “realistic” scenes, to understand how “things exist in the physical world,” and ultimately to help “solve problems that require real-world interaction.” From the start, AI-generated imagery has been characterized by its ability to convincingly learn from and mimic reality. If anything, it’s better than reality because it interprets reality through the analytic logic of the machine, distilling it into its purest form: rarefied data. This narrative helped these companies reach unicorn status and made it so that now even mentioning AI immediately boosts a corporation’s stock price. It also led to the creation of two irreconcilable camps that see this technology as a harbinger of either outright annihilation or salvation.

Yet on the ground, among the humble netizens, a more pragmatic approach is taking hold. In the comment sections under AI images, it’s not uncommon to see people rigorously analyzing them to determine their true origins. They’ll point out subtle features like distorted hands or warped background details as if they were a collector discerning a real Rembrandt from a fake. These DIY art critics don’t fear these images as unquestionable productions of the real but as a style that must be understood like any other.

Insofar as AI companies brag about the “realistic” capabilities of their technologies, and image generators default to a photorealistic style unless otherwise prompted, we can think of AI art as an extension of the realist tradition. As an artistic movement, realism is attributed to nineteenth-century artists like Gustave Courbet and Jean-François Millet, who rejected the idealized subjects of the academic mode—which was historically focused on illustrating dramatized, theatrical scenarios like the Death of Socrates or mythic fables like the cherub-filled Birth of Venus—to depict everyday scenes in an almost reportorial fashion.

As a broader visual sensibility, however, realism refers to a style that attempts to render things “as they are” in a self-consciously neutral manner. We can trace this sense of realism back to the Renaissance period, when scientific and mathematical developments intersected with the arts to produce a naturalist aesthetic that tried to capture the appearance of things with a higher level of pictorial accuracy. It was during this time that tools like the perspective grid were developed to translate three-dimensional space into a flat surface, while increased attention to sciences like anatomy enabled artists to render the human form with startling levels of detail. Some art historians even hypothesize that painters of this era utilized optical instruments like the camera obscura to project images onto canvases, leaning on technological aids to partially automate the visual reconstruction of reality.

Placing AI imagery within this lineage makes clear how its architects leverage contemporary scientific understanding to give image generators credibility and gravitas. Just as the realist style of the Renaissance rooted itself in an evolving technoculture that sought to understand the world through a mathematical breakdown of space and an anatomical understanding of the body, the emergent aesthetic of AI appeals to a belief, however tenuous, in big data’s ability to deliver an accurate representation of the world. Its fluency in the lingua franca of global corporations and surveillance capitalism has given it oracular status in a milieu where big data increasingly has all but the final word on matters ranging from medical diagnoses to prison sentencing.

The full force of AI’s realism emerges from the way its outwardly realistic aesthetic is paired with our analytics-obsessed culture’s sense of what gets to be a credible source of reality. Allying themselves to the reigning “objective” mode allows these companies to reinforce the belief that their tools produce faithful representations of a given prompt—as if they passively translate words into image, the way a camera might capture a landscape. So when a hate-farming MAGA account posts AI-generated images of a “Democrat Protestor” versus a “Republican Protestor,” in which the former appears as a teary-eyed lib and the latter as a clean-cut Chad, the post goes viral, not because they look convincingly real but because they are taken to represent what these archetypes are “actually” like.

Yet as the art historian Linda Nochlin writes in Style and Civilization: Realism, “The commonplace notion that Realism is a ‘styleless’ or transparent style, a mere simulacrum or mirror image of visual reality,” presents a “barrier to its understanding as an historical and stylistic phenomenon.” Though it might position itself as an impartial mechanical process that translates reality to imagery—whether through old-school perspective grids or the occult machinations of algorithms—realism is never neutral. The true-to-life appearances of these images, not to mention the stories AI companies tell about the supposedly objective way they are produced, belie the ways in which ideology seeps in at every level of their creation and circulation. Realism is best understood as the “tensions between reality and its representation,” an “aesthetic argument that constantly shows reality to be a mediated construct,” as the cultural critic Michael Young writes. This style is always in the process of negotiating and establishing our sense of the real through its representation of it.

Soviet Realism, the state-sponsored form realism took under the Stalin regime, gives the lie to the realist mode’s claim to neutrality. Take Arkady Plastov’s Collective Farm Celebration, which depicts farmers celebrating a harvest with tables overflowing with food. The scene is lively, rendered with such high fidelity that you can almost hear the clattering plates and chatter. But this veil of utopian imagery broadcast as fact obscures the millions of deaths that occurred just a few years prior due to Stalin’s rapid collectivization efforts. The lifelike aesthetic of these images presented carefully designed, party-approved propaganda as if it were documentary.

This exemplifies realism’s power to take an “apparently neutral or totalizing perspective while deferring questions about how all representations of ‘reality’ are necessarily distortions,” as critic Rob Horning writes in another context. The best propaganda covers up its own tracks—tries to pretend like it’s not propaganda at all by obscuring the extent to which ideology shapes its production and deployment. That is why the philosopher Boris Groys saw Soviet Realism as a “total art,” realizing the impulse to “cease representing life and begin transforming it by means of a total aesthetico-political project.” In this paradigm, art no longer exists to represent a scene, thought, or idea but to be instrumentalized by a larger political apparatus to shape the real world around it.

The German media scholar Roland Meyer argues that AI art operates from the same premise. It is world-crafting and terraforming while cloyingly insisting that it is operating outside of any ideological paradigm. Rather than reshape the world according to the ideology of the nomenklatura, this art follows the commandeering logic laid out by the for-profit platforms that designed and own it. It’s for this reason that Meyer calls this emergent style “Platform Realism.” This comparison urges us to direct our attention to the particular vision of reality that Platform Realism aims to manifest, the broader project in which it is being employed. The sheer diversity of images being churned out by these generators means that we can’t rely on the content of the pictures to tell us much. Instead, like astronomers studying a black hole, we have to look for signs of mass by observing the distortions AI art leaves in its wake. It’s precisely through close attention to these deformations that we might outline the shape of the ideology operating underneath.

In the chat box used to prompt the popular image generation tool Midjourney, I type in “/imagine: Portrait photo of Asian American 28 year old male with glasses and shoulder length hair in Brooklyn,” which is, roughly, what I imagine an input for yours truly might consist of. A minute or so later, I get four images of bespectacled men with wavy hair and slightly puckered lips staring back at me. One is an absolutely flawless specimen: gorgeous in every way. The other three on first glance appear more normal—like someone you might pass on the street—but after I take a photo of myself and do a side-by-side comparison, I realize that they, too, have an airbrushed sheen of unreality to them. Their skin is preternaturally smooth; their faces are eerily symmetrical and their hair immaculately disheveled. By contrast, my portrait is dumpy. I’m not unattractive by most standards, but my face is asymmetrical. Not to mention the pores and bumps, the imperfections. I like to think of these as endearing idiosyncrasies, but in the AI-generated images, they’re totally absent.

As Caroline Mimbs Nyce notes in The Atlantic, AI can’t seem to stop producing hot people. Sterile, symmetrical, pore-less, hot people. Part of this is likely because these algorithms are trained on so many airbrushed images of celebrities and influencers; but more deeply, it gestures towards the simple yet oft-forgotten fact that the algorithms behind these image generators don’t “see” like we do. Even though computer vision and human vision are both characterized as “vision,” they have vanishingly little in common; the pathways through which visual information is broken down and reconstructed by these technologies differ vastly from our own.

The sheer alienness of AI vision comes out vividly in the case of the distortion most commonly associated with these images: flawed hands. Neuroscientists have noted that humans have specific neural mechanisms dedicated to identifying hands—unsurprising if we believe the evolutionary anthropologists who tell us our hands were critical to the development of our sociality, communication, and technology. For these machines, however, the hand has no special privilege. It might as well be any other appendage, tentacle, growth, or branch. So when I try to generate an image of interlocked hands, fingers sprout from nowhere, begin and end at odd points, and multiply in ways that defy common sense.

Taken together, these distortions betray a much deeper techno-capitalist desire to divorce us from the messy particulars of human life—to “get over” the imperfect human through mechanical technologies that translate our unpredictable, chaotic, material world into something more easily sorted, organized, and commodified. It’s realism optimized. The inhuman vision that underlies AI generators isn’t a bug; it’s the point. It is the incarnation of an ethos that aims to reshape people’s desires and relationships so that it might establish a world of pure transaction. If the needs of actual individuals are too varied and nuanced, we can simply sell to their statistically aggregated algorithmic identities. If people must be given breaks to go to the bathroom or sleep, then we can replace them with artificially generated replicants that can work around the clock. If we’re too flawed to successfully monetize our appearances, then we can just airbrush and filter ourselves to dysmorphic perfection.

AI imagery continues this march toward optimization by flooding our media ecosystem with images that take us further and further from the reality we inhabit. After all, in spite of its realist mode, these pictures are fantasy scenes meant to stoke consumer desire, generate user engagement, and ultimately make money for the people behind them. This sort of dissociative world-building is, as the cultural critic Terry Nguyen notes, the dominant logic of corporate advertising these days, given how effective it is at creating an immersed audience that is always “eager for more.”

This is Platform Realism as a total art—a project less interested in representing the world than in crowding it out with a deluge of sleek, easily consumable content that can be used to continue harvesting our attention. AI images remove viewers from the complex conditions of life in favor of commercially or politically expedient fantasy, all while packaging this fantasy in a realist mode that makes it less ideologically suspicious, easier to take at face value. The future these companies envision is one in which mechanically produced commercials and propaganda are presented as the ultimate reality, while the messy truths of human life are ejected as unpleasant noise or “friction.”

We need look no further than our current poverty of imagination to see how far this project has already progressed: images that all look the same because they’re built off the same datasets, the nth movie remake we’re told we want because the analytics team has divined a profit. Studies even show that we now think AI generated faces are more likely to be real than photos of actual people. When you’ve become accustomed to a sleek and pore-less world, the material world, with all its bumps and wrinkles, starts to seem wrong. Idiosyncrasies and imperfections no longer appear as normal traits to be embraced but flaws to be smoothed over. The primary point of reference shifts away from the world itself towards its false, abstracted, tidy representation.

Ultimately this diversion constitutes Platform Realism’s crowning achievement. It’s been so successful that even criticism of AI falls prey to it—discussing the technology as if it were some magical force that lives up in the sky. But it’s all a sham. The material reality underneath hasn’t gone anywhere; except this time, the false imagery doesn’t hide gulags or forced collectivization but modern-day slavery in cobalt mines and growing carbon emissions. Within a decade, AI might use more energy than most small countries. While it’s right to worry that these falsified images will influence our politics, so will the millions of dollars that tech companies spend lobbying the government to protect their monopoly power. Focus on one cannot overshadow the other.

It is this reality that art about AI, rather than AI-generated images, can reveal. Take Kate Crawford and Vladan Joler’s exceptionally researched Anatomy of an AI System for example, a sprawling diagram that traces the production of an Amazon Echo from its material beginnings in subterranean geological processes, through hardware production, software training, sale, delivery, all the way to its final fate in a landfill. At each point in the process, they detail the laborers, material resources, and profit models that are essential to bringing this device to life. Instead of giving us a dog and pony show that deifies these machines, Crawford and Joler soberly depict the material and economic relationships between the heavy metal mines, smelters, refiners, warehouses, and laborers that actually make AI possible. The technical and bureaucratic schematics that compose the piece might not be photorealistic or sexy, but they are “realer” than any image regurgitated by AI.

To correct the state of AI discourse today, we need to channel this same spirit and train our attention on the messy facts of the human, material world in which these technologies play. Only when we’ve stopped fetishizing “realism” will we be able to turn our attention to very real air, water, ecosystems, and people these technologies consume in the process of generating picture-perfect photos (not to mention the flows of capital and power that make this politically possible). Until then, criticism and discourse will remain stuck—unable to see past the falsely neutral veneer of realism to the complex demands of reality itself.