Wer prägt die Kultur? Mat Dryhurst über KI, Protokolle und die Zukunft der Kunst

Few debates in contemporary art are currently as polarised as those surrounding artificial intelligence. Discussions often revolve around familiar questions: Will AI replace artists? Is AI stealing copyrighted work? Does AI produce nothing but slop? Is AI inherently authoritarian? For creator, researcher, and writer Mat Dryhurst, these questions are not necessarily wrong, but they are rarely the most interesting ones. He argues that much of today’s public discourse is reactive, shaped by headlines and social media rather than by an understanding of the technologies themselves.

For more than a decade, Dryhurst has worked across art, machine learning, and digital culture. Together with Holly Herndon, he has developed projects including Holly+, Spawning, The Call for the Serpentine in London, Starmirror for KW Berlin and K21 Düsseldorf, and Strange Rules in Venice, exploring AI not simply as a tool for image generation but as a medium for rethinking authorship, public participation, and cultural infrastructures. Rather than asking what AI creates, Dryhurst asks who designs the protocols, systems, and institutions through which culture is produced.

In this conversation with Anika Meier for SLEEK, Mat Dryhurst discusses why he believes the attention economy has shaped culture more profoundly than artificial intelligence, why artists should think like protocol designers rather than image makers, why the future of art may lie less in producing objects than in shaping the systems that allow culture to emerge, and what it means to be an artist—or, as he would rather put it, a creator—today.

Anika Meier You’ve been observing the discourse around AI for the past decade. From your perspective, what is the art world actually interested in when it comes to AI?

Mat Dryhurst Generally speaking, it’s a very reactive discourse. The level of discourse in the popular media—even in niche media—is bad. I’m not saying that to be provocative. I genuinely find it upsetting. If your closest touchpoint to the subject is reading an opinion piece in The New York Times, you’re too far behind. That’s not because you’re not smart or don’t care. It’s because your supply chain of information is unfit. It’s a statement of concern.

We’ve been in this space for ten years, and over time you realise it’s almost impossible to discuss the things that actually matter when the trusted sources of information—and ideally, that should be the media—are so far behind. For example, I’ll give a talk now, and people will immediately bring up water statistics. Those claims have been debunked, but we’ll probably spend the next five years in art and cultural spaces with people saying, “That’s all well and good, but what about the fact that AI is using all the water?”

AM There are still exhibitions centred on that topic.

MD Oh, there are probably grants. (laughs) At a certain point, you begin to wonder whether these easy, ready-made talking points are actually an attempt to discuss the issue, or whether they’re a way of shielding yourself from criticism for not really engaging with the subject. You talk to someone, and they say, “I’m a bit sceptical because I’m worried about the world burning.” And you think, well, you’ve just repeated the thing you read in The Guardian. You read it once, and now it’s become the framework for the entire conversation.

The practitioners and people working in the industry often don’t follow this discourse at all. They roll their eyes at it. It’s so misrepresentative of the reality of the situation. But the unfortunate consequence is that they can operate with relative unaccountability, because the people who are supposedly critiquing them aren’t offering critiques that would ever land. They’re the easiest critiques to dismiss.

That’s the point where you start feeling a bit nihilistic. You begin to wonder whether it’s even possible to use the cultural sphere as a place to have productive discussions about these technologies. Maybe it’s just not capable of sustaining good conversations about them.

AM So what sources of information on AI would you recommend to people?

MD The closest you can get is probably a combination of Twitter, arXiv, and research papers. My own information diet is basically Twitter, papers, and then using GPT or Claude to help translate them, because nobody is going to read research papers every day.

Most of the papers I read emerge through Twitter, because I have a very curated feed. Then I’ll pull those papers into GPT and have a conversation with them: Did they get this part right? Did they get that part right? That takes a lot of work, and I don’t expect people to do it, because it’s much harder than reading a headline in The Guardian.

The unfortunate thing is that there’s no real reward mechanism for doing that work. The truth about AI is, first, nobody really knows what’s going to happen. And second, everything is much more complicated than the dominant narratives suggest. There’s very little appetite for technical complexity. So when you try to distill these arguments into something digestible for smart people who are simply busy, it’s incredibly difficult to preserve all the nuance. It’s much easier to reduce everything to a slogan or to feelings. 

AM It’s the same with NFTs and art. If you want to understand what’s happening in the online art world, you need to spend a significant amount of time on Twitter.

MD To much of the world outside of Twitter, NFTs don’t exist anymore. I’ll talk to people who are active cultural participants, and they’ll say, “Oh yeah, I remember that crazy NFT stuff.” And I’m like, Guys, I told you the whole time this wasn’t going away. In fact, the only art studios I know that are actually growing are those involved in those scenes. 

So on the one hand, you have respected cultural commentators rolling their eyes at all of this. On the other, I’m a gainfully employed artist, and I’m telling you it’s the opposite of going away. It’s one of the few areas still standing, albeit having matured a bit.

AM So what do you say to people—and we’ve also been at Hito Steyerl’s conference Art in the Age of Average. The New AI-thoritarians in Munich—who are largely concerned about AI basically being fascist? This is a view shared by many academics, who often speak as if a technology itself could be fascist. What’s your response to that?

MD If I’m absolutely honest, I don’t know what that word means in this context. If the argument is that AI is a tool that could be used for authoritarian ends, then of course that’s true. If the argument is that the people currently building these systems are authoritarians, I don’t agree.

The most legitimate criticism is that a concentration of power always creates the potential for abuse. Even if the people at Anthropic, OpenAI, or Google—none of whom I see as authoritarians—have good intentions, concentrating that much power always carries risks. They hire clever people to think about it! My question is: why do our elected governments understand less about the most important technology of our time than random people on Twitter? I’ve spoken to elected officials. How is it that politicians in Germany or the UK don’t know what a transformer is?

The people building AI models are going to keep building AI models. That’s the reality. The question is whether we have institutions capable of holding that power accountable without also making really stupid decisions. That’s ideally the role of democratic governments here. Enormous amounts of energy are spent speculating about the allegedly authoritarian nature of AI companies, while we elect people who don’t understand the technology at all. It’s very important we understand this technology.

AM In other words, the problem is political, not technological?

MD If you genuinely hold an anti-authoritarian position, then you need state institutions that know what they’re talking about. You also need to support open alternatives and decentralised models. Holly Herndon and I have argued for that for ten years. We’ve turned down plenty of opportunities to attach our names to projects we didn’t believe in, while consistently supporting open-source and open-weight models. Ironically, many of the people who express the strongest concerns about authoritarianism are now ideologically committed to copyright regimes that would leave only a handful of companies able to train or deploy AI models. To me, that’s a contradiction.

So yes, every new technology can be captured by authoritarian power. That’s obvious. If we invented a flying car, an authoritarian government could use it too. That doesn’t make the car inherently authoritarian. What matters are the safeguards, the institutions, and the distribution of power.

That’s why I often find these conversations confusing. I don’t want to be disrespectful, but I’ve heard this argument so many times. If people dislike Trump’s politics, that’s completely fair—I don’t like them either. But if Trump uses AI to generate memes, does that make the technology fascist? Is Photoshop fascist?

AM AI just made it more efficient to create images.

MD In a previous era, producing the kind of propaganda videos Trump has been making would have required finding someone willing to animate them. Not impossible, but certainly more difficult. So in that sense, I can see how the tool enables that. But it enables everything.

Then the next question becomes: what kind of power do you want to create to stop people from making offensive videos? The closer you get to that scenario, the closer you get to actual authoritarianism. So what’s the proposal?

AM There’s a conflict of interests.

You’ve just mentioned open-source models and the work that you and Holly Herndon have been doing for years, from Spawning onwards. What’s your approach, both as artists and through Holly+, to developing alternatives and proposing concrete solutions?

MD Or simply by experimenting. Our main argument is that, once you have what are essentially infinity machines, the craft or skill of producing any single image or song loses much of its value. We will still make images, of course, but the affordances of these tools are different. What’s more interesting is to work one level upstream.

We think about protocols or models as artistic media. Now you can train models. How would you build one? How would you design a data permission system or a protocol that shapes the media or interactions produced downstream? That’s the opportunity. It would have been very difficult to work that way before, whereas now it’s actually possible. Particularly now that everyone will be able to express themselves in software (with coding agents), which was previously a rare skill. 

We would situate that within the conceptual and contemporary art tradition, where you always work one layer of abstraction higher. Think about artists like Trevor Paglen. His individual photographs are beautiful, but if you’re only looking at the photograph, you’re missing the point. The work is really about the broader context, the research, the proposal, and the systems behind it. That’s very similar to the way we approach protocols. Over time, I imagine more artists will start working like this simply because you can.

In a sense, you’re less concerned with authorship of the final media and more concerned with authorship of the coordination: the tools, the protocols, and the systems. In AI, people call that orchestration. That’s much more interesting to us than worrying about what any single picture looks like.

AM Your current exhibition Starmirror in Düsseldorf builds on The Call, first presented at Serpentine in London, and its subsequent iteration at KW Institute for Contemporary Art in Berlin. Across these projects, Holly Herndon and you invited visitors and local choirs to contribute recordings that become the training data for an AI choir, turning what is usually an invisible process into a collective, public one. Can you tell me more about the project and why it was important for you to make the creation of training data part of th e artwork itself?

MD There are new elements because it’s a very different space, but the principle remains the same. People come in; they’re prompted by the songbook we created and by the agent in the room to sing together. We like to call these read-write spaces. The space retains the memory of everything that has happened within it. Models are trained on those interactions, and then the space sings back to you. It’s a closed loop. 

We’re also working on more projects like this that we can’t talk about yet. But we’re increasingly interested in taking seriously the question of what public space becomes. I think over the next year we’ll start seeing the release of always-on devices, and the idea of ambient intelligence will stop being theoretical. It’s interesting to think about what a space becomes when you know that everything in it is agentic, listening, or interactive in some way.

For us, that’s the sweet spot. We use a line in the exhibition: “Angels appear as what you fear; devils appear as what you desire.” Our practice is about running close enough to the fire that all the hair on our faces is gone, but not fully jumping into it. That’s where the interesting ideas and compelling questions emerge.

AM You’ve already presented the exhibition Starmirror in Berlin. We’ve just been talking about how people who don’t spend all day on Twitter or deeply immersed in AI discourse often have a very different perspective. What kind of feedback did you receive from visitors?

MD Once you’re outside the literati class, the vast majority of people are practical and have really interesting thoughts. In Berlin, we maybe had two or three concerns raised, and thousands upon thousands of people came through the exhibition.

Generally speaking, if you treat people as intelligent, give them the relevant information, and say, “Come in and sing,” they get it. The beauty of the singing ritual is that all the voices are merged together, so nothing too personal is really being exchanged. It’s about participating in something bigger than yourself. People understood that immediately. Thousands of people chose to take part.

Then you go on Twitter and see headlines claiming that people hate AI. Meanwhile, we’re inviting people into a space to help train a model, and they’re curious. They ask thoughtful questions. Most of them are already using GPT when their dog gets sick or for all kinds of everyday tasks. People aren’t stupid. They’re busy. They recognise value when they see it, and they also recognise very quickly when someone is trying to manipulate them.

So far, it’s been incredibly encouraging. I have a great deal of faith in people because I think they’re smart. The problem isn’t people. The problem is the blockage in public discourse, which too often fails to treat people as capable of holding two thoughts in their heads at the same time.

AM Speaking of manipulation, I think you’ve come up with a more accurate definition of slop than Hito Steyerl’s “2 + 2 = 6, 7.” You define it as “the media that attention networks prompt humans to make.” Isn’t that, in itself, a form of manipulation?

MD One of my arguments for optimism is that most people’s fears about AI are actually concerns about the attention economy. The attention economy feels so all-consuming that it’s easy to confuse it with technology itself, or with what technology wants. I don’t think that’s true.

Slop, as I define it, is the media that attention networks prompt humans to make. But you don’t need AI to do that. You can do it with your fingers. You can make exactly the kind of content you think Twitter wants. The same has been true of Instagram for years.

AM Algorithms prompt people.

MD Exactly. I don’t think that’s unique to generative AI models. My argument is that there’s actually a strong case to be made that AI could help us move beyond the attention economy as we understand it. That’s a much longer argument, and it’s something I’m writing a paper about at the moment.

Most people assume that the future will look like Wall-E oder The Matrix, where everyone sits around consuming pornography and AI-generated stimulation twenty-four hours a day. I think that prediction will look pretty dated in twenty years. If anything, I hope we’ll look back at the early business model of the internet as something deeply shameful. We’ll think, Wow, we really debased ourselves. Our monkey brains weren’t prepared for this, and we just let it happen. It needs to be reined in.

By my definition, slop is absolutely a form of manipulation. I just don’t think those feedback cycles are inevitable moving forward, or at least the longer we act like they are, the more likely that is to become the case.

AM What makes you believe that AI can move beyond the attention economy?

MD Have you ever read Anthropic’s constitution? It’s really worth reading. Again, this isn’t me stanning Anthropic. There are plenty of legitimate criticisms you can make. But the constitution is genuinely clarifying.

Read it, or throw it into Claude or GPT and ask it to explain it. It’s basically a set of principles: don’t keep people on here longer than necessary; if someone appears to be in mental distress, encourage them to seek help from a human professional; don’t waste their time; don’t advertise. It’s almost like the Ten Commandments. It’s written by a virtue ethicist.

Sadly, I don’t think many people read this stuff, and it is a big deal. Compare it to the lawsuits against Facebook, where former employees have described how the company systematically tried to gamify every interaction because its business model depended on advertising. That’s worth understanding. There are legitimate reasons to criticise Anthropic. But maybe that isn’t one of them.

Trevor Paglen was talking about sycophancy in AI models, and that’s a fair criticism. Models go through these sycophantic phases. But what’s remarkable is that employees at OpenAI and Anthropic openly criticise their own products. I follow a lot of OpenAI researchers, and they’ll openly say, “Yeah, the sycophancy is ridiculous. Anthropic is beating us on that. We need to improve.” These are engineering problems that will be ironed out over time.

What I like is that these systems aren’t trying to keep you hooked or manipulate you into consuming more. They’re not saying, “You sound like you need a Coke.” Of course not. But people often talk about them as if that’s already what they’re doing, because that’s how these technologies behave in movies.

AM I remember you saying that the people behind the Instagram algorithm have arguably had a greater impact as artists than anyone uploading to the platform. What did you mean by that?

MD Whether they’re artists or not is still up for debate. I made this argument in another interview using Dom Hofmann, the creator of Vine, as an example. I’d argue that he is an artist. He also created Loot, although most people know him only for Vine. The person tweaking the Instagram algorithm is different. I don’t know what their intentions are, so I’m not going to call them an artist. What I will say is that the people working at that level of abstraction are the ones holding the keys to culture.

You’ll remember a time before Instagram, when people expressed themselves in all kinds of idiosyncratic ways. The culture industry filtered and amplified those expressions afterwards. Now people express themselves in largely the same ways because, if they don’t, they simply aren’t seen.

Whether you want to call the people building these systems artists depends on how broadly you define art. I have a fairly liberal definition. In fact, I think there’s a strong argument that a certain conception of art has been over for at least twenty years. I’m not saying that’s a good thing, but I do think it’s an idea worth taking seriously.

Even if you reject that more theoretical position, it’s hard to deny that the people shaping culture today are the people building these systems. Acknowledging that can be an ego hit. It certainly isn’t something I celebrate. But once you accept it, you also realise that if your ambition is to influence culture—ethically, politically, or aesthetically—you probably need to start working at that level of abstraction.

People often say, “My meme changed culture.” Did it? Or did memes as a format change culture? Or did platforms simply create conditions where memes emerged? And what material conditions did your memes actually change? Meanwhile, the engineers building these systems are changing the conditions of culture every single week.

AM And we notice these shifts immediately because we constantly have to adapt. As creators, artists, and writers, we’re always responding to changes in the platforms themselves. When Instagram prioritises Reels or changes its formats, that’s suddenly what everyone has to produce if they want to remain visible.

As a writer, I’ve experienced that firsthand. We moved from blogs to Twitter, then to Instagram’s visual culture. During the NFT boom, everyone went back to Twitter. Then video became essential, and people who didn’t adapt often disappeared. Now many people seem to be converging on Substack, and suddenly long-form writing is back again, while others have shifted to podcasts.

It reminds me of Brad Troemel’s essay Athletic Aesthetics from 2013. It feels as though creators are constantly training themselves to keep up with the changing demands of platforms rather than simply making work.

MD One of the great shames of post-internet discourse is that people were already thinking through many of these questions. A lot of what they anticipated actually happened, but much of it simply wasn’t taken seriously. 

If your mental model doesn’t account for the fact that platforms shape culture, then I don’t think you’re reading culture accurately. Once you have that mental model, you can arrive at all sorts of interesting conclusions. But the first step is almost like learning the alphabet. Obviously you’re not in charge. Any pretence otherwise is overblown.

That isn’t a criticism of people posting on Instagram. It’s a criticism of why I’m constantly mocking Instagram itself. To me, it feels like an abandoned amusement park, or a haunted Chuck E. Cheese. Every time I open it, it’s the same performance: another carousel of identical videos, another conference photo, the same formats repeated over and over again. It feels stagnant.

The only places where I still see life are where people are working one layer of abstraction above that, where they’re competing in the realm of ideas with the people who actually shape culture. That’s the interesting space.

AM If I remember correctly, your Twitter bio has said “creator” for quite some time. Why do you prefer that term to “artist”? How do you define those terms?

MD I started using “creator” after someone wrote an article arguing that Holly Herndon and I weren’t artists. My reaction was: Exactly. By their definition of art, they were probably right. I think that particular conception of art is a three-hundred-year invention that has largely run its course. We still participate in what people call the “art world,” but I find the term “creator” more interesting.

Partly because it annoys people. In their minds, a creator is just some lowly TikToker. But I think it’s worth taking the creator economy at its word. Whether they intended it or not, tech companies revived a much older understanding of creativity, one that predates the Enlightenment, where making a painting and making a boat are fundamentally the same kind of human activity.

I disagree with many aspects of how the creator economy has developed, but I think that classification deserves more attention. 

AM But you don’t call yourself a content creator.

MD Just creator. The ideological move made by the tech companies was to place the engineer building neural networks, the person in Indonesia making traditional huts on YouTube, and the sculptor in a studio into the same category. They’re all creators. Whether anyone pays attention depends on what people find valuable (or compulsive). That shift fundamentally changed the position of art, for better or worse. It perhaps dissolved the special status that artists once occupied.

That’s also why we’ve been working on protocol art. In fact, my biggest conceptual tension is whether the word “art” even belongs there. Maybe they’re simply protocols. We add art because we’re speaking to an art audience. If I were completely consistent, I’d probably just call myself a creator of protocols.

AM But online those distinctions have become much harder to make. You don’t necessarily know who identifies as an artist or who is recognised as one by the art world.

That also struck me about the exhibition Strange Rules that you co-curated on the occasion of the Venice Biennale. You included people like Michael Levin, a developmental biologist, alongside artists. It reminded me of the early history of computer art, when many of the pioneers came out of universities rather than art schools. Even today, those practitioners often struggle to be accepted as artists because they’re seen primarily as researchers, despite producing work that clearly has aesthetic and conceptual dimensions.

It seems like you’re deliberately returning to that tradition and expanding the category of who gets to participate in an exhibition.

MD My understanding is that many early digital artists were accepted because their output resembled something already familiar to the art world. Take Vuk Ćosić, for example. You could hang one of his works on a wall. It might be a reclining nude rendered in ASCII, but it still fit an established format. People were comfortable saying, “That’s art.” That’s not a criticism of Vuk. I think a lot of great work came out of that period.

My broader criticism is that the criteria for entering the art canon are much harsher for people who don’t come through art schools. If someone graduates from the Städelschule and says, “I make environments,” everyone accepts they’re an artist. But if someone else builds the sensors that make those environments possible, why assume they haven’t invested the same degree of research, imagination, and intention? It’s an arbitrary distinction created by an industry. I’m not saying that distinction is meaningless. Art institutions have an important role, and I care deeply about them. But if I’m trying to think honestly, I have to ask what really distinguishes someone like Ken Stanley, whom we included in Strange Rules, from someone who graduated from an art academy.

AM So for you, the interesting question isn’t whether someone is an artist but whether they’re producing genuinely original ways of thinking?

MD Ken Stanley is a machine learning researcher, but by almost any meaningful criterion, he’s also an artist. He wrote Why Greatness Cannot Be Planned, a book about creativity and inspiration that emerged from his attempts to build machines capable of exhibiting qualities we associate with human creativity.

When you really boil it down, people who go to art school don’t have a monopoly on creativity, ideas, ingenuity, or subjectivity. What they have is access to a historical pipeline into the art world.

Many of my favourite artists aren’t even called artists. In fact, I’m increasingly uncomfortable with the term myself. What interests me is meeting someone whose mind works differently and whose work produces something genuinely meaningful and distinct. Sometimes those people are called artists. Often they’re not. Maybe “creator” is the better word.

When I talk to Michael Levin, hear him speak, or encounter his work, I get that same feeling I had when I first discovered contemporary art as a teenager. I think, Wow. Your brain is different and I’m uncomfortable and I’m thinking and I want to make things. Whatever that feeling is, that’s what I want from whatever we’re calling art. And I actually think that’s true for a lot of people. In some ways, I think the art world has constructed an alibi around who gets to produce that experience. Letting go of that pretence doesn’t diminish art. If anything, it expands the field.

AM When you were working on the exhibition Strange Rules, I imagine you also had to think about how to explain protocol art to people outside this particular circle. How would you describe it to someone who knows nothing about it and spends their days looking at paintings?

MD With great difficulty. I think the exhibition worked, but you never really know how much of it sticks with people. I usually go back to the Instagram example because people intuitively understand it. Once they grasp that platforms shape culture through rules, the next step is to imagine working at that same level of abstraction: creating rules or protocols that allow new forms of culture to emerge. That’s also the difference between protocols and algorithms. Algorithms are often highly deterministic. You write an algorithm knowing roughly what the outcome will be. Protocols are different. A protocol can be technical, but it can also be social. It can be etiquette, a belief system, or a moral framework.

The difficult question is how to design protocols that are open enough to allow for diverse, unexpected outcomes while still preventing the things you don’t want to happen. In a way, that’s also the challenge of AI. Earlier symbolic AI systems were more explicit and fixed. Generative AI works very differently. The whole point is to create systems capable of producing an enormous range of possible outcomes. It’s remarkable that they work as well as they do.

That’s also the distinction between earlier cybernetic, feedback-based works, or fluxus scores, and something like protocol art. If I tell everyone in a room to behave like ducks, people will behave like ducks. But the possibilities for emergence are more tightly constrained than in deploying networks.

AM But isn’t that true of every new art movement? It only seems obvious in hindsight.

MD It’s difficult, but then again, every new art movement is difficult to explain. People tell me protocol art is too esoteric. I think it’s only esoteric because it’s new. Take Duchamp. Today everyone thinks the urinal is obvious. But I doubt it felt obvious when people first encountered it. Most people probably walked straight past it. It took years of criticism, writing, and discussion before that gesture became part of our shared understanding of art.

I think protocol art is no different. These ideas feel difficult now because they haven’t yet become familiar. But the intuition that something indelibly changed feels impossible to dismiss.