Why ‘AI Slop’ Is an Easy Scapegoat
· Vice
This essay is from the spring 2026 issue of VICE magazine, THE NOT THE PHOTO ISSUE. Buy it now—or get 4 issues each year sent straight to your door, by subscribing.
Mat Dryhurst is an artist, musician, and tech researcher who’s spent more than a decade trying to work out how art and the internet can coexist in a way that doesn’t turn creative people into impoverished, mentally ill slaves. He can be relied upon for ideas from ahead of the curve, even if they sound stupid at first. We asked him why everybody is getting AI slop wrong.
Visit saltysenoritaaz.org for more information.
“Slop.”
To begin with, the term was used to describe low-effort YouTube content made for clicks, then migrated into the machine-learning world as a name given to the most eye-catching, lowest-common-denominator media produced with AI models. Slop is what attention networks reward, engineered to steal time.
Most complaints about AI are really complaints about the platform economy, which, with any luck, AI may help us move beyond. It was the platform economy that made your expensive education economically irrelevant, replaced your creative role with a cheap freelancer, and declared the most worthy art to be whatever grabs attention. It created the incentives for slop. Ironically, it’s the new ability to cheaply produce infinite media to satisfy those demands that may finally break it.
Slop gives us a cathartic focal point for grievances about what has been lost to platform populism and the decline of 20th-century creative industries. For many, it is easier (and more engaging) to imagine the end of the world than the end of social media.
It reassures people that things are changing, but not too much. We’re still scrolling Instagram. This future looks like more of the present: noisier, cheaper, uglier perhaps, but fundamentally familiar.
There’s a meme common in the machine-learning community that captures this. The vast, unknowable latent space of a model is represented by a terrifying mythical shoggoth wearing a friendly smiley-face mask. A reassuring facade over an unfathomable uncharted territory. And yet most of the conversation continues to be about the mask rather than what lies beneath.
What is harder to sit with is the possibility that something genuinely unfamiliar is unfolding, something we don’t yet know how to describe. It’s easier to point at a bad AI image than to bring imagination to a problem with no easy answers.
That’s why so many AI debates feel oddly out of time. People talk about writers losing stable jobs or musicians being displaced as if those conditions still existed. For most people working in culture, those jobs disappeared years ago.
“People speak as if the ability to generate infinite images or songs signals the end of art”
I reject slop determinism and see AI as a remarkable coordination tool. Where social media broadcasts at you, agents and models can socially mediate, prompting you to discover the world and share what you find. The other story of the social internet is how many people already use it to coordinate their lives: to work out, to travel, to learn, with a trusted guide alongside them. Models will amplify that. Think of Pokémon Go: millions discovering things together in real space, coordinated by a virtual agent. That is a better template for the future of culture than the kitsch self-fulfilling prophecy of infinite random media.
There’s also a persistent confusion between media and cultural practice. People speak as if the ability to generate infinite images or songs signals the end of art. Culture is more than this. It is ritual, presence, shared meaning, social context. Media plays a role, but it is self-defeating and reductive to confuse media for art.
DJs are a clear example of something culturally valued that could hypothetically have been automated years ago. Yet DJs are economic juggernauts, well suited to the demands of the platform economy. People care about them for all manner of reasons. Slop catastrophism flattens that. It reduces culture to media and ignores why people value experiences in the first place. AI may dismantle 20th-century industries, but the desire for art and shared experience is a whole lot older than that.
The same AI tools people use to generate lazy media are also the most powerful software development tools I have ever seen. They dramatically lower the barrier to building things. I can now build software that would have taken weeks, or a significant budget, in hours. Software is the language of our time, and once more people speak it, we may not need to rely on a few companies to determine how culture is made and distributed. A working artist can prototype a new environment to appreciate artwork without hiring an engineering team. We do not yet know what that world looks like.
Working with models is new, and something one can master. The eisegesis theory of models suggests they match our intelligence and curiosity. Output quality correlates with input quality. Garbage in, garbage out. The more you customize and contribute to models, the better the result. This medium rewards literacy.
If half the energy currently spent moaning about slop were spent imagining how else we might appreciate art, we would be in a better place. We have a new information substrate and a powerful set of tools. An invitation to propose something better, not just in the arts but in the broader economy that is equally exposed to what is happening.
“AI may dismantle 20th-century industries, but the desire for art and shared experience is a whole lot older than that”
When you build the largest pattern-recognition system humans have made, it exposes us. Big data and personalized feeds show revealed preferences, what we act like we want. When asked, people often say they value complex, challenging media. The data suggests otherwise. That gap between stated and revealed preference is one of the most significant things recommendation systems have surfaced, and further reason not to get lost in the distraction of slop. If we want something else, we have to consume and pay for it.
I don’t think the future looks like endless personalized junk, though the coming decade will be turbulent. Whatever makes someone good at being a writer, musician, or doctor will still matter. How, exactly, remains open.
We are supposedly the imaginative ones. Surely we can do better than fixating on AI-generated shrimp while the tools to build something genuinely new are in our hands.
Using models well is the most literate I’ve ever felt. Understand them.
As told to Kevin Lee Kharas and Claude.
Follow Mat Dryhurst on X @matdryhurst
This essay is from the spring 2026 issue of VICE magazine, THE NOT THE PHOTO ISSUE. Buy it now—or get 4 issues each year sent straight to your door, by subscribing.
The post Why ‘AI Slop’ Is an Easy Scapegoat appeared first on VICE.