AI in the subaltern future
Interview with artist Vishal Kumaraswamy
This is an edited transcript of an interview. You can read a condensed version here.
Q
Are you in Los Angeles?
A
I’m in Seattle actually. I just started a PhD in September, at DXARTS.
I’ve never been to the US before. I think it’s quite an overstimulating experience! The university work is good — it gives me time to focus for a little while, it feels precious.
What’s the focus of your PhD?
I’m continuing some of the media performance works that I was making the last few years. Some of them had research components that I never realised. I never had the resources to just play around with some things. So over the next four years, I’m going to work on each of the components. Especially in terms of scale — being able to scale something, from my one-person operation to using the resources of the university.
I’m not sure about Seattle, but where I come from — Silicon Valley — there are tons of Indian people. Especially from Bengaluru.
Yeah, Seattle is full of people from there. I keep running into people. One of my childhood best friends lives here — I found out like two months before I moved here.
Also, Seattle is very much like Bengaluru — it’s a tech city. I think tech is doing strange things to the city, that you can kind of see. The sad thing the two cities share is the loss of local culture. I mean, Bengaluru — the culture got decimated.
It’s not entirely just tech, but prioritising the building of infrastructure — office buildings and tech parks and Special Economic Zones. The local culture had already started to erode, and these past ten years, with the current regime it’s become extremely difficult. So much so that we don’t even have spaces to gather, as artists or creative practitioners.
And you got to the US with its new president!
Yup! Perfect timing. [laughter]
Back to Bengaluru
You were here in London for a while — you went to Central St Martins, right?
Yes. I think London is one of my favourite cities. I did my MA in Photography there. It’s changed since I was there — it’s gone back to more of a traditional photo programme, but when it was run by Daniel Rubenstein (a student of Johnnie Golding at the RCA [Royal College of Art]), a lot of the thinking and process was very heavy on critical theory and philosophy. So we didn’t do any photography — we started the course on Day 1 with “Photography is dead and we’re moving on”
It was great — very transformational. The trajectory of my practice, which was very photo-based at that point, moved more into tech and video. Also, London’s cultural scene — that’s my metric for what a city should have. I was a broke student, so being able to run around the city with literally no money, and yet to experience culture — that was really special.
Yeah, looking up your work, I came across an exhibition of just photos — I guess maybe when you were just starting out. But actually, I first heard about you through your work at HKW in Berlin — the AI exhibition a few years ago.
Yeah, I was there, moderating a conversation with Tung-Hui Hu — a writer and technologist working a lot with AI. He wrote this book called Digital Lethargy, also A Prehistory of the Cloud — I read that when I was in London, and started following his work. It was an interesting set of propositions around why “the cloud” became this really important tool, and how most technologies can be traced back to military and industrial production. We talked about the role of terminology and neologisms that are required to talk about things in the current moment.
You had a video work in that exhibition, too, right?
I had two actually. They were from this body of research that started in 2018. I was looking at transformer models — the basis for ChatGPT, for example. Just as I was, somebody who worked in an engineering department at a startup — they had been working on translation bots. It was someone I knew from Bengaluru, and he knew I was, like, hopped up on poetry bots on Twitter: “This is great, you can just send it things, and it will write a poem for you!”
He sent me a kind of key to GPT-2, which was the crux for the first well-scaled model, the first proof of concept for something that was reliable and allowed you so much control over the accuracy of what it was going to spit out. It’s what a lot of creative coding classes use. I’m using it now, to build small language models.
So I started having a conversation with this thing, and co-writing about some things I was thinking about. Then I got commissioned by the RCA in late 2019 to do a project called Empathy Loading. I made a three-channel video that looked at some of the concerns about bias happening at that time — which I had come to on my own, with GPT-2. It was one of those moments of accelerated insight into how these models were being built and not really working for a lot of folks.
I looked at it from a simple perspective of how Bengaluru builds the majority of the world’s tech. It supplies a lot of the technologists and engineers and the infrastructure, the backbone. Even for something as cutting edge as AI, the grunt work, the real labor is still going to happen in Bengaluru. Star Wars is still going to get its vfx made there.
It doesn’t matter what the scale of the project is or where it’s coming from, for the rigor and depth of expertise, you have to go to Bengaluru. Nowhere else are you going to get tens or hundreds of thousands of people so highly skilled. It creates internal competition — people want to get better and better, they’re a lot more invested in it. And the ecosystem there really supports it.
There are all these narratives about how AI is being built. But where’s the story of Bengaluru? The city I grew up in, going to the computer marketplace, messing around with peripherals, building computers with people. This was just stuff you did if you were a nerdy, techie kid — you wanted to learn how to do things.
You had to go to the repair marketplace. Even the pop-up stores selling peripherals only started in the mid-2000s. In the late 90s, you’re going straight to the market, and making friends with somebody who would let you play around with something they were playing around with. “Here’s this new Nvidia board — a GPU I got that was cutting-edge, for example. You actually get to see and feel and tinker with things.
So in 2019 I was trying to make work that was a little bit an allegory about that formative experience — not just mine but a lot of people growing up in Bengaluru, why they become software engineers. That’s why it had such a hold on us — it was that tactility, and immediate interaction.
But it was also something that very quickly became a subculture first, and then became the most dominant part of the city. And in that, what was lost was all these folks who had been tinkering and playing. School did some of that — they were like, “It’s Windows95.” And I’m like, “That’s boring for me, because I already did Windows95 on this guy’s computer.”
You start to ask questions about all these narratives coming from the West. How do we have those conversations in our context, and do it bilingually, or trilingually?
So at HKW I showed two works. One was called ಸ್ವಾಯತ್ತತೆ Swaayattate (Autonomy). That was the commission for RCA, and also Furtherfield in London. I worked with a bunch of curatorial practice students,
And from that, I made two specific, tangential works — actually one work but a two-channel video. It looked at other forgotten, unspoken histories of literature — because I had just been playing around with the language parts of GPT-2, and I was interested in thinking about, like, where are the holes in this model? A lot of them were about race or caste.
Instead of trying to make it very didactic, I went down the path of working with two poets I really love. One called Bob Kaufmann — a Bay Area poet, one of the few black Beat poets. He was really in the beat ethos, and he died destitute. I finally found his books for the first time here — I had only read them as PDFs.
He is the person who, all of those Beat poets like Kerouac and Ginsberg, he fit into the character that they’re talking about. He kind of gets left out of that history. But the community — including all these big poets — constantly keep pushing him up there, in their mentions and paying respect to him.
It was similar to a poet I was looking at from Bengaluru — someone from my community: Siddalingaiah. He was writing in Kannada — the language I speak. A couple of his books have now been translated into English.
So I was looking at the two of them and thinking that maybe there was a way to do a kind of comparative lit study. Instead of talking to each other, their works could talk through this GPT-2 model.
The medium is the mess
So now are you engaging with generative image and video?
Less with images, some video stuff. I think visually, for me I find it difficult just to let the machine do a lot. There is something about working in visual and image cultures that is often not seen or represented. Being at the receiving end of that unseen experience, you understand that the majority of image production is happening for a very specific class and section of people — both the technological tools being developed and also the culture that is reproduced.
I’m still keen on thinking through text. I played around a lot with the datasets for This Person Does Not Exist, and that was so long ago, I think it helped at the time a lot of us think through how computer vision can produce a certain uncanniness.
If you’re working in that space, you start to ask questions about things that aren’t directly visible — what you might understand as experiential or phenomenological artefacts. In other words, you understand that something has happened, but the edges of it — it’s the kind of breaking of it, the parts the tech regards as not interesting. That is where there’s something that’s crucial — I think that is where art is generally made, a very human process. The tangential part, something you discard or delete or erase — those are actions with a lot of agency. A full spectrum of autonomy is being expressed.
So for me, the image making part, I find it too sheen. It’s too smoothened out, polished — and not in a good way. It’s like shoe polish — you can keep shining it, but it will never become a mirror. It’s hard to know what’s happening in the black box.
The thing I do like about some of the genAI tools is in the pipeline — image editing pipelines where you can really use things that are frustrating and difficult for one person. Sora has been interesting to look at. I was playing with Runway for five or six years — before it became this strange web-only platform. You used to be able to locally train the models.
Now it becomes difficult for a lot of contemporary, critical art practitioners to locate ourselves within the image. That’s too much of our own endeavour — expressing how something is happening experientially.
I’m still working with language models, and trying to work with smaller ones, more bespoke ones that can be trained on a corpus of data that I have made or generated. I think there’s some potential to continue some of the things I was thinking about.
How much do you get into coding yourself?
A little bit. I know a little bit of Python. I’m doing a lot more now, being in this data-driven arts class at the moment. I’m looking a lot at early internet art — HTML, CSS. I know enough to make some things within a very specific set of constraints.
The tool that I’ve used most is TouchDesigner, which is built on Python. Node-based interfaces work better for me, because I’m able to see it at a glance. I find difficulty in line-by-line coding. We’re using some co-pilots now, like CoLab, and the GitHub one.
Lately I’m thinking about AI not so much as an artist but as an audience for art. In a very real sense, there’s social media. But there are a few artists who are making work explicitly for AI, either as training data or finished work. If AI becomes the main audience for art, do you think it might change the kind of work we make?
I think it will change the work some people make. I’ve also been working as a curator the last three years or so. I think there are interesting conversations about the volume of production — how much you can make. Before, you would need to be some kind of polymath to get to a fever-pitch level of production.
It’s something a lot of media artists go through, because you have to keep learning new tools. I have to learn about audio production, I have to learn coding, and I have to know how to make things and break things.
I think for some artists, there is a possibility of shortcutting some of those things. Within contemporary art, there will always be the commercial aspect, and the more critical, pedagogical aspect, where it’s generating more conversation and discussion. What is changing because of AI is the infrastructure that supports art making — the funding mechanisms, the institutions.
There’s no way you can go on any Meta platform without using AI. So this undercurrent is eroding the reference points to other humans — like me being able to reference a Joan Miro painting. I would have to leave the ecosystem of digital technology. Because even if I look up Joan Miro, it won’t allow me to sit with it, just sit with the possibility of imagination, being able to think about something. It very quickly pushes you to produce something, feed the machine.
So there are people who work at volume, but there are also those who work slowly — I work pretty slowly. Then there are artists like Ian Cheng, who is making the critical things, and because of the scale of production that he’s able to get to now, he’s able to retain some of the criticality. When I saw his work years ago at the Serpentine, it was very different — from Emissaries to now.
So what is the popular reference point for art, and how much does that shift? I actually believe that the conversation about AI will go away, especially in the arts. It’s already got to the point where you start asking the questions like in the 1950s to 80s — when movements in art came about, conversations about aesthetics — a lot of that is being rewritten, or unwritten.
For example, I’m re-reading Marshall McLuhan right now. And I’m like, “How many times is this guy going to keep becoming relevant?” It seems like every five years. He’s never not going to be relevant, because the medium is changing its functions, but what it does to people is not really evolving.
A year ago, I was talking to someone in Bengaluru, who now runs his dad’s computer marketplace. And he was telling me that if you want the latest tech in anything — screens, wires, anything — you have to go back there.
They’re dealing with bespoke requests people are making. He was showing me a flexible LED screen, which is made for one specific client, and purpose. And he said, “When we were growing up, I thought screens would go away!” They’re just not.
So the modality of engaging with the world remains a screen. The size of the screen changes, but all the promise of virtual, augmented, mixed reality — they’re getting sidelined into something that’s novel.
To me, this creates caste systems. Look at Apple with their VR headset — to make only available to certain people, very clearly creating a tiered system. The discourse might be about accessibility, but the technology is about inaccessibility.
The subaltern speak
The themes in your work: caste, class, race and technology. You come from a caste called Dalit, you call it an “untouchable” caste — what does that mean?
The caste system is a social hierarchy, it comes from a very specific Hindu system. In its simplified version, it’s got four castes based on occupations: you’ve got the Brahmins — the “learned people”, the priests and advisors. You’ve got the Kshatriyas — the warriors and fighters, usually also kings. Then you have the trader castes, who do all the metallurgy, blacksmithing, goldsmithing. And then you have the Shudras, who do the bulk of the manual labor.
And then you have the fifth one, Dalits — we’re outside of the codified caste system. The worst of the worst jobs — dealing with human bodies and animal carcasses, disposing of waste, and things like that. There’s a lot of misunderstanding that this was something that happened in ancient times. But it became codified into the social fabric of South Asia. It’s how cities are segregated, how societies form, how our constitutions operate.
Interesting things happened when the British rocked up in South Asia: they find a system that is extremely efficient at oppression, because it’s been honed for millennia. The reason that colonization was so successful there was that all they had to do was plug and play.
This created some unintended consequences that irritated the dominant caste groups. The first dictionaries for our languages, for example, were not intended for a certain caste group, just made widely available, because the British were on a proselytizing mission. They all do it — the Dutch, the Germans, French, Danish. Because their understanding of a base level of accessibility is language — they still believed that is something everybody should have.
But coming to a time period where you have communities like mine, who for centuries have been kept out of access to a lot of things — clean water, anything that could be contaminated, dominant groups look for markers that set them apart. It’s the same conversation with race, or any social system that creates a subaltern — the word comes from Gramsci, who was talking about political prisoners. A sub-altern is a sub-class of people, they’re being dehumanized, being called substandard.
And now, this is the thing that is being replicated in the tech world, because those we would call the oppressor castes are now the CEOs and senior leaders of Microsoft, Google, OpenAI — the kids that Musk and everybody hires. Within that circle, there is no left and right, only holding onto power. Any other positionality is only aesthetics.
So how do you go beyond this power structure that manifests in technology? Both on a consumer level but also at an investigative or critical level. How do you break those tools? Sometimes literally. Other times you try to create counter-narratives. Simply creating a corpus of artworks from communities like mine — that itself is the counter-practice.
So with technology, I see a glossy screen, which is like the sheen of genAI image making. But I don’t care about the surface-level sheen. What I care about is, why is it only generating certain kinds of images? It was interesting when it was spitting out people with seven fingers. Because it forces you to rethink — instead of trying to make it efficient or honest or true to life, if you actually go down that seven-finger path, you generate something that is unlike images we’ve seen before. Which is the task of the artist, the imagination of the designer.
Do you think subaltern futurism can really create a change, to empower people to mobilize and imagine a different future?
I think it’s more of a prompt within the communities. It’s not a kind of umbrella for people to gather under. It’s about thinking, what is the role of a personal practice, a life investment? For me, what feels luxurious is having enough time to make mistakes, and be faced with failure that isn’t like, so total that it will erase your entire practice. But instead a space where you can fail and be more generative in the next iteration.
It’s futurism in a way that we’re not trying to speculate about the most insane, Neuromancer, William Gibson kind of scenario. That requires that we share an understanding, a status quo, in order to get to a position of imagining the absurd.
My position is that our lives are already absurd. They’re already divergent. Some artists I work with, their desire is to kind of intersect into the mainstream — whether it’s art or design or whatever field — so that they can feel included.
But some of us are trying to say that the mainstream, even if you do enter it, you still find a sense of dislocation. The worst possible outcome is getting into that space and then realizing that you’re the only one of your kind. So now you’re faced with educating a lot of people.
Is there a need to go into a room like that? Or do the rooms we’ve created for ourselves have knowledge and value? How do we prioritise our own time? There’s an artist I follow in the UK, Larry Achiampong — he has a book just out, called If it Doesn’t Exist, Build it.
That’s something a lot of communities have done. And for me, that’s where innovation in tech comes from. Some scholars talk about the importance of caste in tech, because our life experiences are very different. You can have 10,000 engineers from the best schools in India who all end up in Bengaluru. But you can have ten who are from castes like mine, and because of the different timelines that they move through, when they actually get to the same place, their point of view is so different, they don’t look at things the same way. And all they need are support structures that allow that to come through.
I think a lot about how we can learn from design. There are things about innovation in design that are about how a diversity of views and experiences are important. You get away from the superstar designer, and more to a team. In design, they talk about teams, but I’m just talking about community. By working together, we increase the corpus of information, of experience we have.
All of the things that academia likes, like peer review — I feel like it can happen, just in a different way. We just have to apply it differently. I especially liked learning from product designers in Bengaluru. There are these brothers, the Foleys, who created the Commonwealth Baton. They had a big sort of public design process, with ads in the paper, getting people to send in designs. I didn’t know what product design even meant, I was in school, and I’m like, “What do you mean I can just sketch something and it will just show up in this baton? How does that even work?” But I saw that there is something there — inviting people into a design process.
Support structures
You’re on this trajectory — moving into the mainstream art world, getting bigger shows, commissions, residencies, exhibitions. I guess it’s a challenge how you bring that spirit of social art practice with you, instead of becoming this big name solo artist.
Yeah, I have friends who’ve worked in Anthony Gormley’s studio in London, and others who used to do the spin paintings for Damien Hurst. Whenever they talked about how it was like working in a factory, in the erasure of their own subjectivities, there was also something about having a basic standard of living — paying the rent. Which means that you can afford to rent a studio, continue making your own work.
It’s sad that paying the rent is the primary concern of a lot of artists. This is something that is not talked about a lot — the aesthetics of labor, whether it’s for yourself or someone else. Everything I do there informs what I can do here
Then there’s a longer conversation about labor and caste. Like I said, that’s why caste continues to exist That is something I am thinking about: how could you set up a mechanism like one of these big-name artists has, but without the superstar persona? Anselm Kiefer — he works out of these bunkers in Germany. All these big marquee artists — is that what you need to create an infrastructure that allows for several people to build?
Do you think a gallery can do that?
I have my own problems with the gallery system. But I do understand the function.
I did want to ask you about your own practice — some of the conversations you’re having. Something you mentioned in your talk in Korea: the process of doing design, and if you think about where those boundaries are. What does it actually mean to do design now? AI’s trying to design for you, but it’s not really doing it. Although I think artists and designers think of that “doing” in different ways.
In one of the other stories, someone told me that the designer becomes more like a curator. If AI is making all these different versions of something, then you’re selecting things rather than making them yourself. You’re directing — that’s the other metaphor: a film director.
I ran a design programme at the RCA with “design” in the name, but it kind of transformed into a fine art programme. Because we just ignored the boundaries between art and design, and just made work. Sometimes it would be more design, sometimes more art.
I’ve been talking with a friend about taking ideas and theory — the most human part — and make that the center of a practice. Now that AI can do all the coding, and maybe the making, to some extent. And the business stuff that none of us want to do. Then could we focus on just ideas and thinking, and still be able to make work. I think people will be interested in that.
This is the optimistic vision of AI — that it’s taking over all these tasks that no one wants to do, making time and space for thinking. I think you’re right, though — those conversations will probably go away, and it’ll be like the internet — it’s just everywhere. AI is quickly getting there. For better or worse.
So I think you’re doing exactly the right thing — thinking about how it’s creating new social hierarchies. You repeated what I say in that article: designers feel like they have to learn and use AI software.
So what I see is that some of the practical aspects of skills are getting easier to do — with coding for example. There’s really something there. AI image creation tools might always be limited in some ways, but it’s the artists who are ahead of everyone — making their own datasets and models. That’s where innovation comes from.
An exhibition just closed at the Serpentine of Holly Herndon and Matt Dryhurst. They make music mostly, and they recorded with a bunch of choirs around the UK, and fed all that into a model. But more importantly, they created a whole governance structure, so that the gallery acts as a data broker. Super interesting. That’s one way to do it.
It’s always the artists trying new things, breaking things. And that’s always been my goal — putting advanced tools and technologies into their hands, and see what happens.
So what are you working on now?
I’m doing two things. In this data-driven arts class, again working with small language models — especially creating a corpus of datasets about caste that doesn’t exist. Then what do we do from that point onwards?
Text generation isn’t the exciting part. The exciting part is the creation of the datasets, because there is some level of human agency to it.
That kind of comes into things I’m thinking about, around the iconographic relationship to language. Language as images, particularly in Kannada, the language that I speak. We know enough image-wise to understand the rules that dictate our own presence in certain circumstances.
Back in 2021, I made one component of it, which was looking at access to public spaces, and the times when the caste rules are suspended — religious processions or funeral processions, for example.
That is the essence of the PhD over the next four years, which is thinking about movement in performance. But right now, there’s that other project I’m thinking about, around the articulation of alphabets. If literacy isn’t important — it’s always been inaccessible — then what is the way we think about and work through language? The shape and form — the imagery — projecting our own experience onto that.
Like poetry?
Yes — the affect, but taking away the readability. Poetry, for me — the language helps me understand how my own experience has been similar, or opposite, what’s being said. Or tangential. Sometimes it sits with you, and it comes back in your body, later.
What if I just hear it, in a language that I don’t understand? Some part is still translated as a subjective experience — my experience, as sound hits my eardrums. What I make of the rhythmic quality, the timbre of the voice, whatever. How that helps me absorb the information, but also to absorb somebody else’s experience.
So I’m thinking about going back to work on the speculative alphabets project, especially in Kannada. Can we still call it an alphabet if readability isn’t a thing anymore? If lexicon and vocabulary is an experience, then how do I think about making that into some kind of corpus?
That’s actually one place I’ve been playing with image generation — from a dataset based on drawings that I make. It can spit out versions and iterate with me as a collaborator, instead of trying to make something polished and sheen. Let’s iterate together, step by step. And see how this informs new lexicons, whether it’s sonic cultures, movement cultures.
All of which still goes back to practices in my own community. How those experiences are divergent, because our relationship with the language is so different, that means it produces a different understanding and a different expression across a range of other forms.
I think some artists are doing very interesting things with film and sound in that regard — especially in my community. So I’m trying to think along with them. I’ve been sitting with this alphabet part for like three years, but now I actually have a server at the university. So I can create something!
