Mat Dryhurst on Becoming Infinite
Mat Dryhurst on Becoming Infinite
After visiting Mat Dryhurst and Holly Herndon’s The Call at Serpentine Gallery in London, Peter Bauman (Monk Antony) discussed with Dryhurst the theoretical underpinnings of the duo’s practice, how The Call aligns with this framework and invigorating the AI-data discourse.
Peter Bauman: You and Holly like to say that “creation is collective.” The phrase serves as one of the key conceptual foundations of your practice. How does blockchain functionally assist in this core endeavor? Could you achieve coordination at the level and scale possible now without blockchains?
Mat Dryhurst: You surely can but blockchains help. I was talking to someone recently about how memecoins have emerged as a kind of multiplayer game—something people play with others on a weekend. You can, of course, coordinate such things through Reddit or Twitter. Composable blockchains just make these kinds of coordination and value games much easier to throw together.
I’ve argued for a while that this is the real benefit of crypto—groups of people speedrunning different potential ideas or economic configurations in more or less real time. Putting those experiments together with other tools would be far more laborious and costly.
Holly+ for example, was an experiment we could not have afforded to run without the ability to use the smart contract, NFT and DAO infrastructure available on Ethereum. We didn’t really make any money from it but it would have been so prohibitively costly and time-intensive had we not used open crypto rails that it likely wouldn’t have happened at all.
Peter Bauman: You two describe another core underlying principle of your practice as “consumption is production.” Can you walk us through that?
Mat Dryhurst: The “consumption is production” point is mostly a comment on how GenAI has the potential to fully dissolve the boundaries between what we consume and what is made. The social platform era already teased this. For example, Spotify thinks it knows what you like—down to a very precise analysis—and as such recommends you new things it predicts you will like. But a human has had to make that media. GenAI is slightly different, as your taste profile can itself become productive.
Using the latest audio generation models, it takes twenty seconds or so to produce a three-minute song on one GPU, which is a new paradigm for media as we are ostensibly generating time.
Media is created faster than we could ever consume it. In this paradigm, it follows that taste profiles, embeddings or models themselves become something valuable upstream of the media they produce. This harmonizes a little with our ideas on protocol art, where it feels increasingly like the most interesting work is happening upstream of the media we usually associate as the central focus of art.
Peter Bauman: Your latest project, The Call at Serpentine, showcases this new kind of infinite media production. The Call features three works of art in one, each demonstrating a core element of the AI training process. The first element you encounter in the site-specific installation is The Hearth, an organ-like instrument powered by GPU fans.
The Hearth physically references the architecture of neural networks and the physicality of an instrument, while the name suggests a place that brings people together. It represents gathering and ritual around an awe-inspiring, quasi-religious object. What is the work suggesting about our relationship with AI?
Mat Dryhurst: We liked the idea of the show having a hearth and had been reading Inventing the Individual by Larry Seidentop. In the book, he writes about the hearth and its role as a quasi-religious site for the worship of ancestors. We had originally wanted to create a library inside the gallery for the show, which for various reasons didn’t work out. We looked a lot at old English chained libraries, where sites of worship also served as this fortified archive. That seems like a more interesting way to think about the AI data question.
If you take our thesis that AI is just us in aggregate—something larger than ourselves—it might also make sense to treat data with some reverence.
The gesture of submitting yourself to an archive or making yourself public domain can be something beautiful and noble—an intergenerational lineage. The object, data and models were intended to be awesome in that spirit.
Peter Bauman: Can you talk more about the brass narrative imagery found throughout The Call, specifically the significance of the lur and child? Lurs are ancient Scandinavian instruments that were used to summon—or call—warriors to battle in the Bronze Age. Is the piece meant to be preparing us for battle today? What are the beliefs and values that the object is meant to communicate?
Mat Dryhurst: We try to keep some of this stuff ambiguous. But when dealing with themes of transcendence, submission and emergence, it makes sense to acknowledge many of the fears associated with the subject too. The Lur and the alien triffids on the objects connote traps and monsters, which is an association many have with this data and AI question that you can’t disregard.
Like in any good fable, however, good things come to those who take a leap of faith. There are so many competing myths around AI. The individual entrepreneur transcending limitations to optimize themselves to explore the galaxy and the Marxist extraction of individual labor seem to be the common ones that frame what we often find to be boring culture wars around it all.
Our AI myths are a little different. Once scrutinized, we find there is some fundamental truth to us not really being easily divisible individuals at any level, that culture is complex, enduring and generative, and that data and models invite us to question those stories—and coordination logics—so we can think of new ones.
Peter Bauman: Drawing from that challenge to individualism, The Call’s participative element is The Oratory, where the audience can interact by singing with the model. This element again has religious undertones—complete with choir, altar and even a miracle-like experience of having an entity respond to your voice. The work explicitly alludes to media becoming infinite. But what does it say about humanity’s potential to become infinite?
Mat Dryhurst: In the show we create a parallel between becoming infinite and becoming public—submitting yourself to the public domain so that others might benefit from your contributions—as being somehow kind of freeing. In The Oratory room, we reference the great Lebanese poet and symbolist painter Khalil Gibran, who described art as a process of becoming infinite. His works recently entered the public domain, making them infinitely composable.
We took this same approach with Holly’s persona in Holly+, that in a sense the greatest honor one could achieve as an artist is to become infinite through adaptations made by others.
To be forked, to have your ideas expanded upon, to have others make something new from what you have contributed, to live in memory—these are all the highest honors from a cultural perspective and are not all that compatible with our fleeting preoccupations with fortifying the individual.
Peter Bauman: You brought up the “AI data question” and have recently commented on the quality of the discourse around AI and data. What does the mainstream discourse most misunderstand about the issue?
Mat Dryhurst: This is such a massive question and probably a bit boring. The AI data question is just very complicated and we and our collaborators have been trying to break ground on it for many years now.
Our position is that ultimately the most likely positive outcome will be for there to be a universal consent protocol that allows for people to reserve rights over their data in order to share models of themselves or omit their data from models entirely if they prefer.
We say this with some qualification that might take a while to fully explain but it appears that is where things may be going with a great deal of effort expended on our part.
The problem is that there exists very little appetite for nuance in media networks that increasingly incentivize blunt arguments. My fear is that the mutually reinforcing anti-tech and pro-tech team sports create a false sense of awareness of issues for regular people who should be getting solid information. There are parallels with crypto, where this kind of new medium of tech entertainment discourse—largely led by non-experts who say evocative things—would have many people believe that crypto died a few years ago (and yet!). Whatever you think about crypto or AI, I’m just not in support of discursive games that feed people bad or misleading information.
Peter Bauman: How can we improve the discourse?
Mat Dryhurst: It feels a bit futile right now, if I’m honest. I guess we improve the discourse by attempting to tell people the truth as best we can and consciously attempting to avoid baiting or misleading people, as are rewarded in these entertainment feeds.
Peter Bauman: What questions does The Call pose about AI and data collection, permission and governance?
Mat Dryhurst: Quite a few. There are two things going on: one, our thesis of becoming public or dreaming of a public AI as being a thing of beauty. But we also acknowledge that this should ideally be a matter that people consent to. As such, the Serpentine Future Art Ecosystems team has put a lot of work into making it so that the choirs who participated in The Call own their own voices. We had to negotiate with them to use their voices for the art, just like anyone else would.
We take this public AI thing quite seriously. Our company, Spawning, has been doing some great work collating and enriching public domain datasets. This year we will release a fully public domain image model to take these ideas further. We think there is beauty to the idea—more beautiful yet if we can actually make it real.
Peter Bauman: Consent is so important. We’ve had our site scraped at Le Random and it didn’t feel right so we fully support what you’re doing.
Another thing that delighted me about The Call was the absence of screens. Can you talk about that decision? Why was it important to emphasize the physicality of the space and its artifacts when, as Holly has said, “The [non-physical AI] model is the artwork”?
Mat Dryhurst: We try to follow through on our ideas with as much fidelity as we can. The AI model is the artwork; the data is the artwork; the protocol coordinating it all is the artwork.
For better or worse, most people now associate “AI art” with looking at digital images on screens so it felt important to omit screens in order to really hammer home our emphasis.
We have made a lot of screen-based work and love a lot of screen-based work but the point of this show was to emphasize the protocol. So far, I think that has been successful, as removing screens has invited people to contend with our ideas in ways that perhaps screens may have obfuscated.
.jpeg)
For previous projects like Holly+ or xhairymutantx [commissioned for the 2024 Whitney Biennial], it's often hard to explain to people that the artwork is the model and the protocol around it. These are models that produce infinite media—lots to hear or look at.
But the art is really situated in how and why the models were trained and how and why they are designed to be used. I think that over time this will seem less unusual.
Peter Bauman: You mention the term “AI art,” likely with some ambivalence. Is there something more precise you prefer? You’ve also spoken often about protocols. Do you see “AI art” in the traditions of generative (autonomous systems) and systems-based art—somewhat like this diagram I shared on Twitter?
Mat Dryhurst: I don't particularly like "AI art" but feel it will stick around in the same way "NFT art" has. “AI art” generally refers to images that look like they were primarily generated from an AI model. We have used the term "protocol art" for many years to get close to what we do, which is related in a way to systems art—but different.
Holly+ for example, is an AI model, is a dataset but is also a set of permissions for the permissive sharing of Holly's voice and identity. Again, the whole protocol is where the art takes place. There are analogies to artists who build smart contracts to determine specific ways to interact with an NFT project. Historically, it is closer to systems art—or process art—in that the output media is really just one part of the art project.
But it's a little more participatory than a lot of cybernetic art, which was really about chance operations—Rube Goldberg-style exhibitions of feedback. Protocol art is a little more opinionated than that; our systems/protocols are set up to enact a certain kind of interaction and outcome—a little more like an AI agent than an arbitrary feedback system. We have a book coming soon that goes into this, with some essays from people also thinking about protocols related to The Call.
My initial interest in Ethereum was the ability to compose protocols of interaction—little economies, opinionated systems. For example, I would consider Art Blocks to be protocol art, as the smart contract, its opinionated expression of scarcity and value, is a protocol that has been forked countless times. We have a small group where one of the original debates was to set up a collector group for protocol and smart contract art.
Peter Bauman: Protocol art reminds of Frieder Nake’s “algorithmic art” or Jean-Pierre Hébert and Roman Verostko’s Algorists. Are those traditions you see yourself reimagining—perhaps with a more participative or opinionated bent? What do your interventions with protocols allow that other expressive forms do not?
Mat Dryhurst: I can see similarities and differences. Similarities are that you are looking at building your own systems to produce bespoke outcomes. We've always been interested in that. In our older works, like Saga or Platform from a decade ago, the gist is that we are most interested in building our own tools in such a way that the output is ours, which shares a spirit with algorithmic artists.
I guess the difference is more biographical. I studied for example, with Casey Reas and have always been somewhat adjacent to artists working on algorithmic work.
But my focus has been more on coordination, social scenes, values and value—which was actually what drew me to crypto so early.
Ten years ago, my recollection was that very few people focused on algorithmic art were all that compelled by crypto, as it was quite abstracted from using software to produce media. They were more interested in producing software to coordinate. Those two fields, to my recollection, only really converged when Art Blocks became so successful.
On the whole, there are more similarities than differences but just biographically, I always saw our work as different from that tradition. I was always more compelled by figures like Ted Nelson, who was interested in different ways the Internet could work.
As for protocols and AI, I wouldn't say they are exclusively interesting. It's just where our interests took us.
You can experiment with interesting protocols with a pen and paper and I might find that more interesting than most AI work.
Ultimately, I'm a modernist and tinkerer in spirit. The emphasis on protocol over any one technological approach is useful as it means I'm pretty medium-agnostic. For example, many of my favorite artists don't necessarily have to be technologically advanced in terms of the tools they use. But they do establish a distinct protocol or practice over time that any work they do has a recognizable character and approach.
Peter Bauman: You just said, “My focus has been more on coordination, social scenes, values and value.” How does this emphasis on the socio-technological realm challenge the critique—by, for example, The New Yorker’s Ted Chiang—that art should simply be reduced to the number of choices an artist makes?
Mat Dryhurst: I acknowledge Chiang's point. I take the position that the burden of proof to claim something is not art should be much higher than claiming something is. Intention is not so easy to account for. In fact, in making music for example, as with other media, oftentimes you try to get yourself into a state of mind where things are not so calculated.
That said, in the context of “AI art,” I do think there is a difference between jamming on a text-to-image model someone else trained and some of the bespoke model training, data collection, etc. we do. I don't think one is more "art" than the other but they are different. I would argue the construction of models is a somewhat novel proposition—just because the considerations are so alien to prior ways of working.
I think what frustrated people about that article was less the claim of choices—which I think is worth critique but a defensible claim—and more so that Chiang didn't seem overly familiar with how deep you can go into creating art with AI tools. He did, however, leave open the possibility of working with models in a more classically authorial way.
Like I mentioned, my interest in protocols is medium-agnostic. It derives from the observation that the upstream development of protocols, platforms, etc. is a site of art that is rarely acknowledged. I use the example of platform incentives and UX ostensibly creating media through others—the feed tells people how to make work for it. This is another form of modeling with humans in the loop and is a little Deleuzian [challenging traditional structures].
In the spirit of Gilles Deleuze’s war machines or Reza Negarestani's Cyclonopedia—which posits oil as an intelligence that instrumentalizes us—you could also argue that the platform and attention economy decision to put data above all else is ostensibly choreographing humans to produce more of it for some unknown goal. The current state of the AI industry would not contradict that theory.
---
Mat Dryhurst is an artist, musician and researcher who works collaboratively with Holly Herndon on medium-agnostic, protocol-based work.
Holly Herndon and Mat Dryhurst’s The Call is on view at Serpentine North, London, until February 2, 2025.
Peter Bauman (Monk Antony) is Le Random's Editor-in-Chief.