IN CONVERSATION WITH is a series from Archetype where we interview artists in/at the edges of crypto across music, visual art, design, curation, and more.

Mat Dryhurst is a British artist, musician, and technological researcher based in Berlin, Germany. Dryhurst often works on interdisciplinary audiovisual projects in collaboration with his partner Holly Herndon. The pair is known for their pioneering work in music, machine learning, and protocol art and development. Dyhurst and Herndon also founded Spawning, a company building open tools for public AI.

Dryhurst’s and Herndon’s critically acclaimed works include Platform (2015), PROTO (2019), Infinite Images (2021-2022), Classified (2021), xHairyMutantx (2024), and The Call (2024). In 2023, Herndon and Dryhurst were named in the “100 most influential voices in AI” by TIME.

Dryhurst has also previously taught at NYU’s Clive Davis Institute of Recorded Music, Strelka Institute, and European Graduate School. He also previously served as Director of Programming at Gray Area in San Francisco.

The following interview has been edited and condensed for length and clarity.

Katie Chiou: For those who may not be familiar with your work, can you share more about your background/journey as an artist? 

Mat Dryhurst: The short version is that I’ve been making work for a very long time, often with my partner, Holly Herndon. In our art practice, we develop our own instruments, our own tools, and our own software. We get interested in a topic or a technique, try to learn as much about it as possible, and then often make artworks in various different media, exploring those ideas. The convenient alibi for all of this is that we consider the production of software, tools, protocols and different ways of thinking as artworks in and of themselves. 

KC: You and Holly had a project called Platform in 2015 and then another project called Proto in 2019. Together, they sort of demonstrate a narrative arc of how you think about the evolution of the internet from platforms to protocols. Can you share more about that thesis/evolution?

MD: Holly and I lived in the Bay Area during the web2 tech boom. The observation that became really apparent going back to around 2008 was that the people who were tweaking rules upstream of the production of media––so the people who were designing platforms––were making very consequential decisions that had downstream effects on aesthetics, art, and culture. I remember, for example, having a long argument outside a music venue with my friend Aza Raskin about his startup Songza. Songza was the first to really explore playlisting and this idea of testing as being a really good way to get people engaged with music streaming in that new format. I argued with him about how that format would break up the music album format in a highly consequential way. Holly and I had many, many different experiences similar to that in the Bay that then flowed into our work. We were also touring musicians at the time. We were witnessing the nature of the record changing, and we had to ask ourselves, “What are the possibilities here?” and “What could you use the opportunity of a record and live tours for in this new context?” We started putting together Platform from 2012 and it came out in 2015. A lot of that time was spent looking at the roles of platforms in art making and in our lives, which might seem kind of obvious in retrospect, but at the time, it wasn’t really obvious. It was actually something we had to fight for. I developed a piece of software called Saga that was an early attempt to produce, what I later learned was, a smart contract system that gave people more agency over the media and art they created online.

At that same time, we had friends in the Bay who were developing federated, distributed systems. There was a lot of discussion around Bitcoin, and I became very interested in Ethereum. It made sense to think about different kinds of platforms, different kinds of ownership models. Ethereum itself felt like a very beautiful idea there. While becoming very interested in crypto as an art medium, we also became very interested in machine learning. Between 2015 and 2019, before we released Proto, those seemed to be the two areas where there was the most potential. In that period of time, we developed a fluency in machine learning, and arguably preempted a lot of the discussions that are happening today—around AI and art, data ownership, and the different affordances and conditions of this new information substrate. I see the two as being very complementary. 

KC: Do you feel like your interests in crypto and AI were/are independent of each other? Or did/do they feel inextricably intersectional?

‍MD: Yeah, it's funny—semi-independently. We've definitely been involved in both fields for quite a long time, but really for our own purposes. Now, it's easy to say that there's a very fully formed crypto industry and a fully formed machine learning industry. When you go back to the early-to-mid 2010s, that wasn't really the case. We were just talking about research questions and different things that we could do with software. So, in a way, we were fans of those topics more so because we're always interested in new opportunities to explore ideas.

For crypto, the affordances of decentralized networks, and particularly the money legos approach—the ability to quickly prototype something and have it live on this network—was just really useful. I was also obsessed with the DAO idea of making an institution with a shared bank account. I'm not as idealistic about those ideas today, but I’m still very interested in them.

On the machine learning side, it was really clear that it was going to be a massive deal. I guess those two areas were both enabled by the GPU. That's the central unifying factor—you could now do stuff at home with a consumer GPU. With that premise, you could see that there was a massive opportunity to bring some imagination and context to the field. Both Holly and I come from an experimental music background. There's a very long tradition of artists and musicians getting involved early in the development of the recording industry—the electrification and digitization of music, the internet, and so on. For us, it was quite incredible to see that opportunity present itself for our generation.

In my mind, they were never fully separate. They're just two new ways of looking at things that happen to intersect. If you go back even to a lot of the early writing with Ethereum, the concept of the DAO, for example, was quite transparently inspired by Daniel Suarez’s Daemon, the science fiction about runaway AI systems. So a lot of the early conceptual work going into the Ethereum network and others was thinking about: “What if machines had money? If the economy were to continue to virtualize and machine learning were to develop at quite a rate, what kind of infrastructure might we need to deal with that throughput?” So in a way, they were always wedded.

KC: You and Holly recently published a book called All Media is Training Data. I’ve seen many different—both pessimistic and optimistic—interpretations of the title. Can you talk more about the thesis there and maybe clear up any misconceptions?

MD: On principle, we believe that once you put something out in the world, all interpretations are valid. There were a couple of different names proposed for the book, and the ambiguity was understood from the beginning. 

I do feel there's a tendency when you're talking about AI, for people to use it as a bit of a Rorschach test to project their biggest fears. It felt appropriate to use an ambiguous title in that same vein. Generally, you can always guess what our interpretation would be because we are insipid optimists about things. We're earnest, to a fault, I'd say.

Our idea was simply the observation of saying, “Everything that has been recorded and is available in public is to be understood now as training data for mind children of the future.” It's not a new observation. The idea of mind children comes from a Hans Moravec text, I believe from the 1980s. This is well-trodden territory. But ultimately, our position is that there's great beauty to that—that the long arc of culture is a matter of people contributing to a shared story, contributing to this thread that is greater than oneself. That actually is what culture is, and that should be embraced.

Of course, the other reading of “All media is training data” is a dismissal of the sanctity or the specialness of human art. That is a problem, and that is a dynamic. I definitely know of some people who think of it that way, “Okay, thanks for all the data, now we can move on.” I don't take that position.

Either way you approach the statement, however, for us, it was really important that it conveyed that this is the reality, not some distant techno-futurist idea. Once we acknowledge that, we can think about what the opportunities are. Now that we are training machines in the future, what kind of new challenges does that bring up? What kind of new questions? How do you want to train them? What kind of values do you want to encode in intelligence in the future?

There's this nice opportunity for reflection––not quite Roko’s Basilisk kind of fearful reflection—but more a question and an opportunity to tweak some things in a direction we prefer.

KC: On the note of training data, I think another lens people use is monetization-focused––“You can use my data but pay me for it.” I think there’s massive opportunity there, but it quickly gets very thorny in terms of equitable information rights, information access, distribution, and these sorts of things. How do you think about this topic through an economic lens?

MD: As a case in point, I always think it's funny looking at Twitter/X. For years, some of the driving motivation behind the decentralized experiments in the early 2010s was, “We're posting all this stuff, people are getting value from it, and we don't get any value back. What if there was a way to have a social network that paid you?”

It's funny now, in retrospect. One of the biggest issues with Twitter today is the payment system—it creates strange incentives. You now have a feed inundated with slop that people make in order to grab your attention in order to get paid for it. Oddly enough, some of the ideas that would have been considered progressive ten years ago, or the ambitions that were being considered progressive ten years ago, are now to be looked upon with some degree of skepticism. I actually now really miss the days where nobody was getting paid on Twitter because then at least you had to be intrinsically motivated to participate.

Monetization is complicated. We hold two positions. Our fundamental position is that we believe everybody should ultimately have the right to choose what they participate in, which is why I like the idea of decentralized networks. You can set direct pricing, and there's no centralized authority telling you how to run your business. I like that idea in the abstract. 

Holly and I released the project Holly+ a few years ago, pushing forward this idea with our own information and our own works. “What if you as an identity, or your data, would be default in the public domain?” If we're involved in a project, you're allowed to do whatever the hell you want with it. That's our principle. I think there's more value created in that process, and I don't see it as being deleterious to the original in any way.

That being said, I understand that other people would take a very different stance. The work we've done with Spawning, for example, asserts our position that consent and matters of monetization and licensing are complicated enough that it would be nice to build in the ability for people to opt out, if only to create as diverse as possible market opportunities to explore what monetization or value might look like in future.

KC: Switching gears a little bit, your latest exhibition The Call explores AI as a coordination technology. I’d love to hear your latest thoughts on DAOs and decentralized coordination networks broadly—what’s worked or hasn’t worked and how can the space move forward?

‍MD: If you zoom out a great deal, you could make the argument that Bitcoin is a DAO, or that Ethereum as a network is a DAO. In that case, it's been incredibly robust and is working really well. Maybe scale is all you need. Maybe for DAOs to work, they have to be like countries in that sense, a kind of public. There are also other scenarios. If you go back to the original DAO ideology, the promise of new kinds of institutions, I think a lot of that has actually transpired a bit. If you look in the NFT universe, there are collector DAOs that have played huge roles in shaping a narrative amongst different collectors who are particularly passionate about machine learning art or generative art.

There have also been tons that just collapsed. There have been a bunch of investor DAOs that ran into their own problems. Some still persist. Some don't seem to exist anymore. That's also fine. The idea of an institution that emerges for a moment, does good work, and then dissolves when it's no longer needed is also kind of cool. The liquidity of an institution or of a collective is maybe a feature, not a problem.

At the same time, I've also heard from people who have had horrible experiences. In the heyday of NFTs, there was definitely a premium on promising that a project would live forever, and all the collectors were now part of this group. That worked out in some cases. CryptoPunks are a legitimate group of people. There have been ups and downs and some drama in that world, but it's still trucking. There are other NFT communities that came and probably shouldn't have ever made the promises that they did. They became a crazy burden to whoever started them. So, it's really a mixed bag. 

In general, the principles have largely adhered, so I'm still pretty interested in these ideas. The challenge, of course, with these open, decentralized networks is you're going to see a lot of things crash and burn. That's the thrill of it. And the risk of it. It's an uncontrollable economic engine. For me, I find it thrilling—even though I've had my fingers burned, as probably everyone has. It's kind of a rite of passage.

These networks, in my mind, are places where ideas are speedrun. Someone comes up with an idea, a bunch of capital is deployed from people who support it for whatever reason, and then things happen. Nine times out of ten, it will crash and burn. One time out of ten, it becomes a standard that everybody builds on.

KC: To pull that thread specifically toward creative spaces, I had a conversation recently with a friend about artist grants and similar crowdfunding-like structures. A critique I heard is that capital in those structures is often committed semi-blindly, there’s no actual point of view that binds the artists with the funders in a way that feels fundamentally meaningful or connective. That’s not a question, but I’d be curious to hear your thoughts there.

MD: I'm honestly a bit of an institutionalist at heart. Even though I will be as critical as the next person about the art museum or the snobby gallery that excludes or gatekeeps, I think it’s great. I'm of the position that gatekeeping is a necessity. I just think there should be more gates. There should be more gardens.

Institutions—whether you're a brick-and-mortar institution, a collector group, you're starting a gallery, or starting a project—should have a strong thesis. You could change that word with the word protocol in order to encourage others to join and produce more work under that strong thesis.

That was always my hope for the space. There are some cases where that thesis is weaker or looser. In those cases, you get a lot of waste and many experiments that aren't all that interesting. In other cases, far from it. In other cases, you get people who are aligned in terms of their interests, producing works in a more interdependent configuration than existed before. You can do great things in those configurations.

The question there is really about the strength of the thesis and the strength of the conviction, the strength of the ideas. I would just argue, generally—and I make this argument to museums and more brick-and-mortar institutions who are also struggling with dealing with the transition online—to stick to your guns. There's a point of dilution where people don't have a strong enough thesis to be able to set parameters—tight rules  or even strange, weird new rules around things. That's what I don't like. I don't like it when the big museum in New York is trying to catch up with the trend on Instagram. Ultimately, that's kind of the point of institutions—whether they're new or old—to have a position.

As a case in point, Serpentine Gallery in London is a very storied place that I revere. When you're invited to do something at Serpentine, it's an event. It stands for something. It has a history. You could say the same for the Museum of Modern Art or the Whitney. I think that's a good expectation, and that ultimately we've all benefited from these different competing positions being out in the world.

As a case study, I find memecoins conceptually fascinating, but it is very rare to encounter one that meets that criteria in an interesting way. You're coordinating people under a certain word, but the rules and interactions that you set—the protocol you set around it—doesn't seem distinct enough to produce a different outcome. That is not to say that you couldn't use those mechanisms to do something more interesting. It's just very rare to encounter them.

KC: While I largely agree with you, a popular counterpoint is that if you harness enough attention/mindshare and/or capital in dense concentration, maybe some cohesive and aligned thesis could actually come out of that group in a deeply emergent way.

MD: It’s definitely possible. There’s always edge cases, but it’s tricky. I've fallen prey to this a few times, but there's always a “canonical” example that's used to prove a thesis.

You look at Pepe, or Doge—Doge as kind of the original meme coin. It's become this behemoth. It was the first one. But there’s also this entropy to it, where the more times it's replicated, you get this diminishing returns effect. That, to me, weakens the overall thesis. Whereas in other circumstances, you don't get diminishing returns. People establish a protocol that just consistently works. I guess a lot of it comes down ultimately to incentives.

One critique you could easily have of the ICO period is: was it a good idea for people to IPO on a PDF? Was that a good idea? The great example you could point to is Ethereum. Clearly, that was a great idea—look at all this crazy value that's been created. It clearly worked. But 99% of them didn't and were awful. The incentives were completely askew. 

On principle, the idea is actually super interesting. So it's our collective job to retroactively look back and say, “How do you switch the incentives here to make this more interesting?” 

KC: To clarify, when you say incentives, you specifically mean economic incentives?

MD: Yes, but also social/reputational incentives, basically the rules of the system. Let's say you go back to the original ICO era—you publish a PDF, make a lot of grandiose promises, a bunch of people ape into it, you then leave with the money and hire someone in their 20s to keep a Telegram channel going and pretend that things are happening—that kind of token design, ecosystem design, protocol design, is clearly a failure. It's highly dependent on trust and reputation. It's really a matter of being able to read intentions and understand—and also the really tricky part, which is the other dimension in this whole protocol question—reputation. In the arts, for example, you have an art practice. You set the terms of that practice, you try to stay true to it, and ultimately not take the shortest path to make a bunch of money, because that nukes your reputation.

KC: As an example here, what are your thoughts on Botto?

‍MD: Botto is a great example. I know Simon well, and I like Mario very much. The goal was always clear. There's a community, people who are really animated about it. It's a model that continues to grow and continues to evolve. I think it's one of the few enduring examples in that space that stuck to their guns and are a great reference point. I haven't collected yet, but it’s a case in point of an actually good idea that endures and is worth people's time and attention. There's been a few of those.

KC: In an age of abundant information and technology, do you think the methods by which young people, specifically artists, develop their own voice fundamentally changes? Or is that par for the course generally across time. Related to that, what are your thoughts on sloptimism and how people sort through noise?

MD: I’m 40 years old. I think a lot of people my age and some even younger never quite reconciled how much creative practices have changed. A lot of people, particularly on the AI question, talk about creative careers as if it were 1999 or something. That’s not the case.

For example, the expectation that the musician makes a living selling their record or something like this. For the majority of musicians I know, a record became a mechanism, a focal point, to coordinate live tours around. It’s a vanishingly small number of people who actually make a living from recorded music. Many more tour. So when we talk about the media form of recorded music as being the site of value, I don’t think that’s been the case for a long time. I can imagine for a 21 year old, with very few exceptions, that’s a very abstract notion. 

The other side: I do think AI is a kind of epochal shift. There are things that come and go and change with the times, and then there is, well, this is a big deal. Everything changes. When that happens is hard to tell, but it’s a special bird, a special challenge. The idea of bringing imagination to that domain feels urgent. I can’t reliably say that things will look the same for creative people, or even the battle lines, or the distinctions of who is or is not a “creative person,” contextualized by the traditional culture industry. I don’t know whether that exists in 15 years. Which is not to say art doesn’t exist, or people won’t be creative, or make art. But I think this new information substrate is a bigger deal than the internet. 

On the point of sloptimism, I think Dan Keller coined that. I used to call it high slop.

KC: I actually just interviewed Daniel, he referenced you and Holly a lot.

MD: Yeah, Dan is a very old friend. That’s really funny. 

If I go back to 2012, both Holly and I spent years trying to hammer home the notion that contorting a creative practice or identity toward platform incentives was a humiliation. Not to say I judge anybody for doing it, people have done interesting things. 

But our position was: It’s ridiculous for us to go on Instagram as musicians or conceptual artists and be invited to take pictures of ourselves all day to seduce somebody into liking us in the hope they might engage with a book we write. That is a debased understanding of culture. We were very hardline on this. Of course we participate, because you have to, but I hate it. So if you're telling me that the ability to quickly automate media to meet an algorithmic goal will be overwhelmed by automated media and characters cynically creating slop—slop originally being YouTube clickbait. If AI is going to overwhelm our current valuation mechanisms and cultural incentives: bring it on. I love that idea. I think it’s a forcing function for us to reconsider curation of media, to reconsider the valuation of people with interesting contributions.

I don’t want to be an old man yelling at the clouds. A lot of people use Twitter well, use it as a medium. There are incredible YouTubers whose work I enjoy. But for the most part, the logic of web2 was a mistake. People were obsessed with network effects, with the conceit that more attention means better. That’s been disproven. It was right enough for companies to grow, but culture has been debased for it.

So my optimistic take isn’t about Italian brainrot or whatever, but that this phenomenon will overwhelm our current media landscape enough to make us reconsider how we fundamentally value media.

I’ve pulled the trigger on subscribing to magazines I used to read now and then. I now have Harper’s and The New Yorker subscriptions because I can’t take the slop. I subscribe to Substacks because of the signal-to-noise ratio.

So that’s my sloptimistic take. I don’t get enjoyment out of watching slop. But the overwhelming nature of the slop will force us into making interesting decisions about culture that we should have made 10 years ago.

KC: What potential externalities related to AI and cultural production/information dissemination do you think is talked about too much and talked about too little?

MD: I think what's talked about too much is the low-hanging fruit—replacement narratives.

That’s often silly and belies a fundamental misunderstanding of what art is, which is a consensus network between humans. It’s not pictures or whatever. That kind of category error leads to an interminable amount of conversation that doesn’t need to happen. Also, generally, the conceit that the culture industry—as we currently understand it—is under threat, and that means culture is under threat, is absurd. 

What’s not talked about enough is that the culture industry as we know it was a thriving anomaly for less than 100 years. There are all these assumptions that will remain fixed. In crypto, people have a high tolerance for abstraction. They intuitively understand that the way you release something is also part of the thing. That’s slept on. It's so liberating to just understand the obvious truth that that is not the case, that these things are fluid, and that's part of the fun. And then once you realize that, you realize we have much more agency over the future than we may believe that we do. You have to get over the preoccupation with things staying the same.

‍—

Disclaimer:

This post is for general information purposes only. It does not constitute investment advice or a recommendation or solicitation to buy or sell any investment and should not be used in the evaluation of the merits of making any investment decision. It should not be relied upon for accounting, legal or tax advice or investment recommendations. You should consult your own advisers as to legal, business, tax, and other related matters concerning any investment or legal matters. Certain information contained in here has been obtained from third-party sources, including from portfolio companies of funds managed by Archetype. This post reflects the current opinions of the authors and is not made on behalf of Archetype or its affiliates and does not necessarily reflect the opinions of Archetype, its affiliates or individuals associated with Archetype. The opinions reflected herein are subject to change without being updated.