On April 1, 2026, models walking the Maison Margiela runway in Shanghai looked like uncanny porcelain dolls. In photographs, it was hard to tell whether they were real or AI glitches. Glenn Martens crafted garments from ceramic fragments, gravity-defying drapery, and handcrafted masks that turned faces into glossy, inhuman surfaces. The porcelain effect—achieved through airbrushing, scanning, and textile printing—resembled a generative error, yet every piece was entirely handmade. A few weeks earlier, a video spread across global feeds. Two masked figures performed microtonal math rock at KEXP studios in Seattle, using double-neck instruments connected to loop stations. The Canadian duo Angine de Poitrine wore Dada-inspired masks, moved like automatons, and produced music that felt almost alien—yet still rooted in contemporary musical traditions.
To those unfamiliar with AI-generated art, these works might seem like reactions to the kitsch monotony of generative models. But for those who follow artists working with AI, they reveal something else: an aesthetic shaped by those who know how to sabotage the algorithm.
It is not artificial intelligence that is the problem, but the logic of optimization that has long governed cultural production.
This is not a contradiction. Algorithmic art did not begin with artificial intelligence. The logic of optimization has shaped cultural production long before neural networks. The pull toward what is predictable and instantly recognizable is rooted in markets, perceptual habits, and the structure of the entertainment industry itself. A Spotify playlist designed to maximize listening time, a film franchise calibrated through audience testing, or an Instagram feed optimized for engagement—these are all systems of optimization.
The so-called “slop”—the flood of mediocre AI-generated content across platforms—has become a common term in aesthetic debates. But the phenomenon predates generative models. Most human-made art is just as predictable as algorithmic output, because it gravitates toward the center of taste, where audience expectations, formal conventions, and commercial demands converge. AI has made this convergence more visible, but the trajectory was already in place. What artists like Angine de Poitrine and Martens are responding to is algorithmic creativity in a broader sense.
Art is not reacting to AI, but to optimization
Looking more closely at these cases is revealing. Angine de Poitrine construct every aspect of their performance around the impossibility of being optimized. Microtonal math rock is already niche; adding microtonality—intervals outside the Western tempered scale—pushes their sound into a territory where music-generating AI models, trained largely on standard tuning, have little to draw from. Their Dadaist masks, domestic staging, and deliberately imperfect aesthetic complete the picture.
Margiela FW26 follows a different but parallel strategy. Martens appears to take the visual language of AI and recreate it physically. The ceramic fragments composing the garments reference synthetic imagery while simultaneously rejecting it: each piece carries weight, fragility, and irregularity. The porcelain effect in ready-to-wear—achieved through airbrush and scanning techniques—produces surfaces that look like Midjourney outputs, yet possess a tactile quality no rendering can replicate. It is a hall of mirrors: the show looks like AI because Martens understands its aesthetic, but the result is radically handcrafted.
They are not alone. In music, Oneohtrix Point Never has spent over a decade building a sonic universe of corrupted pop, degraded advertising jingles, and fragments of muzak reshaped into something uncanny. Arca pushes the voice into hybrid states between the biological and the synthetic, anticipating many of the textures now associated with generative systems. Neither relies on AI (as far as we know), yet both operate in peripheral zones of taste where optimization loses its grip.
Aesthetic value, in an age of pervasive optimization, might coincide with distance from the predictable.
The phenomenon extends beyond music and fashion. The Coinbase commercial Your Way Out, aired during the 2026 Oscars and directed by Oscar Hudson, reconstructs the aesthetic of low-resolution video games using entirely physical means. Sets are printed and painted to simulate polygonal textures, costumes feature pixelated graphics, and actors move like NPCs with mechanical gestures. The result is a collapse between analog and digital, where each can convincingly imitate the other.
Artists who sabotage AI—by using it
The most compelling evidence emerges when we look at artists working directly with AI who produce a similar anti-optimized aesthetic. Mario Klingemann, a pioneer of AI art, offers one of the clearest examples. Since his early work with GANs, he has focused on moments where models fail: faces collapsing between identities, bodies deforming into impossible configurations, surfaces losing coherence. Rather than correcting these errors, he presents them as the most interesting output of the machine.
Among newer voices, Scottish artist Katie Morris embraces absurdity through frenetic, dreamlike scenes where human figures are trapped in impossible situations. She generates around 150,000 images a year and selects only a fraction, turning curation itself into a form of resistance against the idea that AI makes image-making effortless. Her strongest works emerge from embracing unpredictability and strangeness.
The project BAD JPEGS by Mind Wank, presented by fellowship.xyz, represents a radical form of anti-optimization. It asks a simple question: what if AI were trained on the early internet? The result is low-fidelity, pixelated, poorly assembled imagery—precisely what makes it less predictable. It is “slop” turned inside out. Anna Condo takes a different path. With a background in film and photography, she approaches AI through composition and color. Her work pushes saturation and decorative excess toward a baroque, camp sensibility, moving away from the average aesthetic favored by generative models.
Holly Herndon and Mat Dryhurst complete the picture. Their project PROTO merges human and AI-generated voices into a choral form that produces a similar sense of uncanny hybridity. Their earlier work with DALL·E 1, Infinite Images, involved constructing visuals piece by piece, exploring the limits of the model through an ongoing dialogue with the machine.
The most interesting artists today are those who can derail the algorithm, even when they use it.
In Italy, similar approaches are emerging. Andrea Meregalli uses AI tools improperly on purpose, forcing them off track. His project Effetti AI_ndesiderati interprets pharmaceutical leaflets for antidepressants, generating unsettling imagery somewhere between clinical hallucination and perceptual distortion. Photographer Eugenio Marongiu moves away from photographic realism to build dreamlike worlds that critique the pathologies of contemporary fashion more effectively than traditional imagery. Alessandra Condello, coming from parametric design, works with algorithmic logic rather than against it, selecting from machine-generated possibilities those closest to her vision.
Andrea Gatopoulos brings AI into auteur cinema. In The Eggregores’ Theory, he uses older AI systems to generate black-and-white imagery, editing each frame individually, as if assembling an archive already distorted by a machine.
Deviation has always been absorbed—and will be again
It is tempting to frame Impressionism as a reaction to photography, just as contemporary art is often framed as a reaction to AI. But this reading is superficial. The real target is a logic that cuts across all media. The medium itself often produces the “slop,” yet the outcome can differ dramatically. The decisive factor is not the technology, but the relationship to optimization. It is worth acknowledging that the celebration of deviation has a long history. From avant-garde movements to glitch aesthetics, every era has produced its own dissidents of predictability—and in most cases, the margins are reabsorbed within one or two generations.
What has changed today is speed and scale. When the center of cultural production is dominated by systems capable of generating content at high velocity, absorption becomes faster and more systematic. Any successful deviation will be captured and replicated. Yet this is not only neutralization—it also expands our perceptual capacity.
If this diagnosis is correct, asking whether a work was made with or without AI becomes secondary. The more relevant question is this: How far does it deviate? In an age of pervasive optimization, aesthetic value may ultimately coincide with distance from the predictable.
Opening image: Angine de Poitrine in concert at Cartonnerie, Reims, France. Via Wikimedia Commons