The Metabolic Image: On the Cost of Slop

As a writer, artist, and leading scholar of AI, Kate Crawford has long traced the social and political implications of large-scale data learning. In her lecture delivered at Prada Frames: In Sight, Crawford dissected the impact of AI slop on everything from our planet to our cognitive ability—offering a theory of image-making in our new era and the metabolic rift that it heralds.
On the tropical island of Muna in Indonesia, archaeologists have found what has just been recognized as the oldest image in the world: two hand stencils pressed against rock and outlined in ochre, made over 67,000 years ago. One of the hands was altered—the fingers elongated, sharpened, and made to resemble an animal claw pointing ahead. The archaeologist who found it said that the image-maker had “imaginatively transformed the human hand” into something else entirely. The precise meaning is a mystery, but the intent is clear.That image has outlasted its maker by millennia. The act of making pictures was never mere documentation, but an instrument of transformation. We transform the world into images, and those images, in turn, transform us. That process is now being performed by machines, at scale. We are only just beginning to understand the bargain being made in our name and what it will cost.

Photo: Adhi Agus Oktaviana.
Let’s turn from the sacred to the profane: Fruit Love Island. This is an AI-generated micro-series in which anthropomorphic fruit compete for love in a tropical villa. Fruit Love Island became one of the fastest-growing accounts on TikTok in history, reaching 300 million views in just over a week. It’s classic AI slop: mass-produced synthetic imagery optimized for algorithmic velocity. Each episode is generated rapidly for minimal cost and maximal likes. While all of this might seem mindless and ephemeral, it heralds a serious cultural and economic shift: a new era in image-making.Slop makes up 20 percent of what YouTube is recommending, and right now nearly three-quarters of new web content is primarily AI-generated. It is becoming the dominant visual culture of the networked world. AI slop is also a weapon of war. In the past year alone, President Trump has released AI images of himself as a pope, a rock star, a golden statue, and a religious icon descending from the heavens. These images were condemned as blasphemy and viewed by billions. Iran has responded with its own satirical slop: LEGO-themed videos in which Trump appears as a childlike demagogue. This is called slopaganda, mass psychological operations designed to shape your interpretation of the world before the facts arrive. As Jean Baudrillard once argued, the simulation comes first and then it replaces the real. This no longer seems controversial. Slop has flooded the zone.AI slop is more than a cultural curiosity or a technical artifact, however. It is the consumer-facing product of computational capitalism. It is created by rendering all human culture into a standing reserve of data—the pointed claw directing us to the future. But to see where it is taking us, we need to understand where it came from.Technologies have always been central to the history of images, from the use of artificial perspective in fifteenth-century Florence to the daguerreotype and the digital camera. Artists and theorists have long understood that these tools encode the politics of seeing. In 1983, the philosopher Vilém Flusser coined the term technical image, arguing that photography was not an expression of the human hand or mind, but rather an apparatus that carries a worldview. By the 2000s, the artist Harun Farocki had begun to work with operational images—produced by military targeting systems, facial recognition, and drones. Such images are made by machines and for machines. They require no human eye. They are not pictures of a place, but rather instructions about what to do to it—a command.

© Harun Farocki, 2003 / Harun Farocki GbR.
Generative AI extends this lineage into something new: the metabolic image. Metabolic because of the enormous ingestion and digestion of data and energy now happening on a planetary scale. Metabolic images are created by consuming billions of images and texts, all the stuff of human civilization—or at least everything we have uploaded to the cloud. Our existing visual material is broken down and digested by neural networks before being excreted as synthetic outputs. That is the slop that is overwhelming our information ecology. Those outputs are then ingested again, and the cycle continues.This entire process requires an enormous amount of energy, minerals, and water. By 2030, these systems will consume more electricity per year than Japan. Generating a five-second AI video requires the same energy as running your microwave for an hour. In 2025 alone, AI systems used over 400 billion liters of drinking water, more than all bottled water consumed worldwide annually. All of this releases clouds of carbon dioxide into the atmosphere, and the tech industry is on track to overtake the airline industry in terms of its emissions by the end of the decade. As ephemeral as those episodes of Fruit Love Island seem, they leave a dangerous and enduring environmental legacy.In the 1850s, Karl Marx described how capitalism severed the ancient loop that returned human and animal waste to the soil: as people massed in cities, the fields they left behind were depleted, and farming came to depend on industrial fertilizer. This became known as the metabolic rift. Generative AI has created a new metabolic rift, one that is separating humans from basic resources like land, energy, and water, while also creating a rupture in our own practices of cognition and creativity.In a recent study, MIT researchers divided students into three groups. One wrote essays with ChatGPT, another with search engines, and one without any technology. Those who were using ChatGPT produced clear, well-structured writing, but 83 percent could not recall what they had just written. Researchers call this cognitive debt. In other words, a metabolic rift in consciousness itself, a severance between thought and action.This rift accumulates as a kind of forgetting, not sudden or dramatic, but rather a slow retreat from the difficult pleasures of artistic creation: the poem written through grief, the image that surprised its maker, the song that required a person, alone with a piano, to follow it. To outsource all of this is not simply to be efficient. It is to become mechanical. In doing so, we are alienated from our labor. We forget our own voice.

AI will, of course, find important uses. But it is the capacities that we once thought were most human, such as art, writing, and thinking, that are currently being automated by the tech sector. When did we decide that these were the tasks we wanted to surrender? AI slop is the indicator of our current direction of travel: toward an image architecture driven by planetary extraction, concentrated corporate power directing human thought, and the replacement of creative labor with mass industrial mimicry. Industrial AI, like mass fertilizer before it, makes us more productive in the short term but depletes the very material that feeds our imagination.In the Muna cave paintings, a human hand transforms into an animal claw. Like any good image, it has many meanings, but at its most fundamental it reminds us of the difficult work of taking what is and rendering what could be. The gap between the hand and the claw is where imagination lives. What we are building now will certainly outlast us. The question is what kind of mark we want to leave.











