AI Realism
Why We Shouldn't Compare Paintings and JPEG's
Essay first published in the Botto “Neurealism” book now available.
Last month, Scott Alexander asked his followers a simple question: can you tell which of these two images is a painting by a well-known artist and which is AI-generated? He called it the “AI Art Turing Test”. The test was straightforward: participants were tasked with classifying 50 images as either human-made or AI-generated. The images spanned various styles, including Renaissance, 19th Century, Abstract/Modern, and Digital Art. To ensure a balanced and fair comparison, the selection included renowned human artworks alongside high-quality AI-generated pieces, deliberately excluding images with obvious AI artifacts such as distorted text or anatomical inaccuracies. There were four main findings.
First, the 11,000 survey responders had a hard time differentiating between human and AI art. The median accuracy among participants was 60%, only slightly above random chance (50%).
Second, images with known artistic styles were harder to classify. For instance, Impressionist works were overwhelmingly identified as human-made, even when they were AI-generated. Conversely, digital artworks were more frequently misclassified as AI-generated, regardless of their actual origin. The implication here is interesting. “Style” influences our perception of humanness in the process of creation. A digital style is perceived, therefore, as more likely to be AI-generated.
Third, 60% of the top ten rated images were AI-generated. This suggests that AI can produce art that not only mimics human styles, but also meets or exceeds human standards of aesthetic appeal—at least those of the sample who responded to the survey.
Lastly, responders who initially said they disliked AI art ended up, unbeknownst to them, preferring AI images. This can imply that what they meant by “AI art” is a specific type of grotesque image which the test intentionally avoided and that when presented with a more genuine attempt at artfulness, their preference is not reflected statistically.
However, this is not a test, but a trick. A false equivalency cloaked into the deceptive cape of statistical science. The AI Turing test attempts to flatten art to a mere set of pixels: an enframing that transmutes art’s transfigurative capabilities into a colorful object, a commodity, a standing-reserve, ready to be compared next to another. What the test actually performs is not “statistics” but an abolition of art’s wholeness in favor of a digital flattening, a violent ideological pixelification.
The test foregoes the material reality of art, its weight, its texture. It removes the place where it is encountered, its cultural and historical context. It abolishes any concept of scale. The test actually favors AI-generated images, and because they are playing on their own turf, (the screen), they have an unfair home-court advantage.
Naturally, Walter Benjamin’s 1935 essay “The Work of Art in the Age of Mechanical Reproduction” is a great resource for understanding the implications of removing physicality in art through its reproduction. Briefly, Benjamin argued that the aura of an artwork—its unique presence in time and space, tied to its history, materiality, and physical location—is abated through mechanical reproduction. Consequently, by digitizing paintings for the test, their aura is stripped away. A digital image lacks the tactile reality, temporal uniqueness, and spatial presence of the original, the context of it all. Benjamin would see this as flattening the experiential depth of art into mere visual data, which is inherently alienated from the original creative act. This turns it into a commodity, something ready for rapid consumption and comparison, rather than something to be deeply contemplated. Here we see the slippage that Heidegger mentions in “The Question Concerning Technology,” the proclivity of technology and, by extension, rational thinking, to transform all things into “stock.”
Benjamin does highlight a positive aspect of this disappearance of aura in his essay. Its loss possessed a democratizing potential, and a countering force to fascism, which often relied on tradition to exert its power. The severance leads to a multitude of encounters with the artwork outside of its controlled environment. It denies the ability to yield art as an untouchable power-symbol; as a weapon of propaganda, the loss of aura is the dulling of the sword. Though an optimistic prediction, the current state of the work of art is a bit more contentious. Not quite a tool for fascism, but also not quite a tool for democratic subversion and empowerment, the status of the work of art is contested. The AI Art Turing test is a great example of the powers vying for art’s influence. The very premise of the test implies that the digital image is equivalent to the physical artifact. This is not a mistake, it is an ideological position. One that claims that the entirety of reality can exist within the dimensions of digital life. The same mechanisms that dissolve the aura also recontextualize art in ways that align with technological systems of control, manipulation, and mass consumption, with the means of mechanical and digital reproduction.
When art is severed from its aura, it loses its anchor in time and space, engendering a fragmented and disoriented cultural landscape ripe for conquest. The techno-optimist systems exploit this cultural disorientation to impose simplified narratives that favor the ideology of technological supremacy. When an Impressionist masterpiece is merely a JPEG, it becomes acceptable to compare it on a screen to “just another image.” Everything becomes just another square or rectangular digital container to scroll through. In this new flattened world of pixels, where all has lost value, reality is subsumed by the digital and its endless capacity for generation. Can we tell which images were made by AI from those which were made by humans? At this point, on a screen, probably not. We long for the aura, and yearn for a set of divine aesthetic commandments, but all we have are these tablets.





beautifully concise refutation of this type of framework and 'study.' I think this kind of relationship to art, where only the digitized, flattened end-result version of something matters reveals how little those who champion AI art understand, respect, or care about what the artistic process itself means, both for individuals and culture at large. I personally am on board with the murmurs I hear about a return to physical objects/mediums that, rather than being flattened by their own methods of creation and distribution, are imbued with a presence and lifeness that encourages thoughtful engagement and reflection