The bank clerk at a loan appointment is playing the character of a bank clerk. A tarot reader plays the character of a tarot reader. An artist at an art show plays the character of an artist.
It also relates to AI and LLMs. LLMs are told to play a character. They’re given system prompts you never see…
…The text that [LLM] Claude generates is written in the first person. You can ask, who is that “I” that is speaking? I use that as a launching-off point to ask a more human question: when I speak, who is the “I” that is speaking?
Marcus Round’s sen(i)ence is hard to describe. Is it a piece of computational art? A performance (complete with a harassed bank clerk / casino dealer living in Brazil-style techo-bureaucratic dystopia)? A shared sentence-building game, with a Tarot inflected twist? A philosophical exploration of the ‘I’ played by LLMs, and what it means to be human?
Through a combination of computational savvy, meticulous world building and exploration of the philosophical underpinnings of self, Round somehow manages to bring all of these elements to sent(i)ence, creating something greater than the sum of its parts.
We welcome Round into the Voidspace to tell us about collaborative algorithms, uncanny CAPTCHAs, and why we need a human to show us the heart of the machine.
Voidspace:
Welcome to the Voidspace. Let’s begin with asking who you are and what your practice is in the interactive arts field.
Marcus:
I would describe myself as a computational artist whose practice revolves around LLMs — large language models — and how they interact with human agency. I’m interested in how language relates to agency, and in LLMs as a site of intersection between human agency and the emerging phenomenon of computer or AI agency. Specifically, I focus on language as the site of that interaction.
I make installation games in which people interact with each other through a system I’ve created that involves an LLM and constrains their use of language.
Voidspace:
We’re mainly here to talk about sent(i)ence, which you showed at Voidspace Live, and is your most recent work. I’d be interested to know whether there were predecessors exploring similar issues.
Marcus:
sent(i)ence was the final piece of my Master’s in Computational Arts at Goldsmiths, so the whole degree fed into it. My original undergraduate degree was filmmaking, so there’s some influence there. I never had experience with theatre‑making, though it always interested me.
The previous piece with the clearest link to sent(i)ence is Conversation(AI)symmetry — both pieces were exhibited at the V&A Digital Design Weekend, in 2025 and 2024 respectively. It was an installation game for two people. It wasn’t narrative‑driven — it was a game people could sit down and play. They role‑played a first date, but they had to pick words given to them by an AI and construct the conversation in a playful way.
I was making it in late 2023, which was riding that first initial wave of the ChatGPT release. It was fun exploring how people increasingly use LLMs to express themselves — getting ChatGPT to draft emails and so on — and turning that into a playful game.
Voidspace:
I think it’s really interesting that you followed that interaction between human conversation and machine conversation. I’d offer a gentle challenge when you say that wasn’t a type of theatre — that it was purely a game. I’d offer the idea that a kind of theatre I’m very interested in is the theatre two participants create between themselves when they engage in a formal structure that holds their interaction.
Marcus:
Yeah. I agree, and I was consciously aware, particularly with that piece, that I’d constructed a system for two people to interact, but they were very much on display to an audience. They were in the middle of a gallery space, and people would come in and see them.
I always think about the moment someone comes around a corner in a gallery and suddenly sees two people at a romantic dinner table — there’s that moment of “wait, what?” It’s the same with sent(i)ence: they come around the corner and see a poker dealer at a table in the middle of a gallery. It’s strange.
Voidspace:
I like that with your film background, you’re thinking about the tableau — the idea of the filmic tableau that others witness.
And I think there are two kinds of theatre here: the theatre from the outside, where passersby encounter this tableau; and the theatre from the inside, which is the theatre the participants create between themselves.
Both are equally important. There’s something about how these systems have complex inner workings you can see from one perspective, but from the user perspective you’re getting the theatre from the outside.
Marcus:
Something significant I learned from Conversation(AI)symmetry, which then influenced sent(i)ence, was that when I was exhibiting Conversation(AI)symmetry, the two participants were performing in a way that everyone else could watch. It was quite funny to observe.
But also, I was there as the artist — not in character, or was I? I wasn’t performing a role, but I was still “Marcus Round, the artist”, greeting people and explaining the game. I’d say, “Sit here, scan this QR code to get it running, swipe the words like this,” and so on. It became a looping ‘performance’ where I repeated the same instructions endlessly.
I think a lot of artists experience this in gallery settings — people come in, you talk them through the work, you explain the themes, and you feel yourself slipping into a groove, like a record needle. I really noticed that. And as I was developing sent(i)ence, which is a more complex experience, I realised I’d be repeating the rules and the setup over and over. So I decided to script that. It becomes part of the piece — me explaining the rules, asking people to sit down — and then I can twist it, introducing more ambiguous conceptual themes.
Voidspace:
For readers who haven’t experienced sent(i)ence, could you outline what a participant actually encounters? What is the piece, for someone who hasn’t seen it?
Marcus:
It’s an experience for two people. I’m there as the artist‑performer, sitting at a table. The table is essentially a poker table. I’m wearing a waistcoat and bow tie — I look like a casino croupier. So when you come around the corner, it looks like a croupier is waiting for you to sit down.
Except it’s also a desk. I’ve got a keyboard and a screen. My demeanour is professional, slightly bureaucratic. So on the other hand, it feels like you’re sitting down at a bank desk, applying for a loan, or maybe at an embassy for a visa application, or even a job interview.
Voidspace:
I love that duality — the scenography of one thing layered with the effect of another. Having both at once gives you that jolt out of reality that lets you engage more deeply. It sparks curiosity and a sense of unreality that opens you up to the experience.
When I did the piece, I noticed that there were also other visual cues around the table that I really noticed. First, I’d love to hear about those, and then what we actually do as participants and how the experience unfolds.
Marcus:
As you sit down, I give you a little introductory spiel. I’m fully in character, and it’s still ambiguous what you’re about to do. I start shuffling cards — you can see I’m going to deal them, as if I’m a croupier.
But around the desk are stacks of books. I deliberately made it feel cramped, like I’m a low‑level bureaucrat working in a broom‑cupboard office with books piled everywhere.
Voidspace:
Love an immersive broom cupboard.
Marcus:
And the books are arranged so you can clearly read the titles. You might start to sense themes — AI, language, consciousness. The piece is called sent(i)ence, so that’s your only clue.
Voidspace:
I remember Borges being there. What else?
Marcus:
Borges, Calvino — a lot of people recognise Calvino. I’ve placed If on a Wjinter’s Night a Traveler quite prominently, with its recursive cover. If you know these texts, you immediately situate yourself.
Voidspace:
If you know them, they cue you into the themes you’re about to encounter. But they also add a layer of abstraction. Long before AI, recursion and self‑reference were explored in printed text.
What extent did those texts influence your thinking?
Marcus:
They absolutely were strong influences. You might think I bought them as props, but at least 90% of those books were already on my bookshelf before I even knew what sent(i)ence would be. Maybe 5% I bought as research, and another 5% were props.
Voidspace:
Which specific texts influenced you the most?
Marcus:
Calvino I encountered fairly recently, but Borges — Borges is an incredible reference point for AI and LLMs. He was writing long before AI, but he’s only become more relevant.
If on a Winter’s Night a Traveler is relevant to LLMs in terms of prompt and response. An LLM is essentially a massive database of text: you give it a prompt, and it completes the rest. People think of LLMs as chatbots, but underneath, it’s just text completion.
If on a Winter’s Night a Traveler is a series of first chapters — you mentally fill in the rest, trying to find threads between disparate texts. At the end, all these hypothetical books feed into one meta‑text.
Voidspace:
Feeding into text and meta‑text.
Marcus:
Exactly. That’s what’s happening in sent(i)ence. You see all these book titles and start generating or imagining something from them.
Another major influence is Douglas Hofstadter, especially I Am a Strange Loop. I read Gödel, Escher, Bach years ago — incredible book — and then I Am a Strange Loop, which focuses on the word “I”: what we mean when we use it, where it comes from, and the relationship between linguistic “I” and subjective self. That’s at the core of sent(i)ence.
Voidspace:
I love that. So that’s what greets you when you first approach the piece. What happens next?
Marcus:
After you sit down, I welcome you and explain that we’re going to write an affirmation. I say that an affirmation is a personal expression of hopes, fears, and desires — which, in the interest of efficiency, I’ve built an AI system to autocomplete for you. People usually chuckle at that point. You can see what’s going on: it’s ostensibly a therapy session, but with an ironic twist.
I deal you the cards. Each card has a short phrase: “I am afraid,” “What I didn’t do,” “My father,” “I’m not ready yet.” Very personal, first‑person fragments. You’re tasked with playing them into five poker slots.
Instead of making a royal flush, you’re making a sentence — something like: “I am afraid of what I didn’t do, but I know my father will forgive me.”
As you look at a card, and decide whether or not to play it, your mind goes somewhere. “What I didn’t do” means something different to you than to the person next to you, and different again from what I thought about when I wrote it. Then it’s fed into the LLM, which generates the continuation — the statistically most likely next part. Arguably the statistical average of what everybody thinks about after reading that phrase.
Voidspace:
Something I find fascinating is the layers of structure upon structure. And I love that the affirmation is iteratively, collaboratively built. Each person plays a card, and you can use poker chips with conjunctions on them to make sense of it.
There’s randomness and intentionality — like poker — applied to a collaborative poetic experience.But the individuality is deeply subjective and privately held. I play “what I didn’t do,” and my subjectivity goes into the machine; the other person responds with theirs.
Marcus:
Yes. You can’t see each other’s cards. You’ve only seen your five; you don’t know the rest of the deck. Every phrase is new. You play “What I didn’t do,” your partner reads it, associations spark, they continue the sentence — maybe “is,” on a poker chip, then their next card: “What I didn’t do is what we don’t talk about.”
It’s written in the first person — “I” — but whose “I” is it? Yours? Theirs? Mine, as the artist who’s seen hundreds of these? The LLM’s statistical “I”? All these interpretations swirl.
And that concept of short phrases evoking things in your mind, ties back to the books on the table: a book title is a tiny phrase pointing to hundreds of pages. The title gestures toward meaning, but it evokes an imagination.
Voidspace:
Exactly. And people will know some texts, not others. No one except you will recognise all of them. There’s always a mix of knowing and not‑knowing, and the power of evocation fills the gaps.
In collaborative poetry, you work toward shared meaning. But here, you’re combining subjective, internal meanings. Each phrase means something different to each person. So if you add multiple “I”s… does the “I” disappear?
Marcus:
As I developed the piece, I struggled with the question of what exactly participants were going to be doing, within the fiction. I knew the mechanics — poker cards with words, semi‑random sentence construction, LLM completion — but what do I call it? At one point I thought, “Are we writing poems?” I could have said, “We’re writing a poem, and for efficiency the AI will autocomplete it.”
But the idea of an affirmation — a personal statement like “I am worried about this, but I know I’m strong” — constructed from two people using a random deck, really hit the core question: is the sentence meaningful? Sometimes I’m genuinely struck by the sentences created before me, as I sit there in character. They feel poetic, relatable. But are they meaningful? Or is it just language being reconfigured into something that seems meaningful? That’s also the question with LLM‑generated text.
Voidspace:
That’s really interesting — and a great bridge to describing what the LLM actually does. What does the participant see the LLM doing, and how does that compare with what it’s actually doing?
Marcus:
One problem is that the piece is hard to describe — you really have to experience it. But: you and your partner play the cards, construct a sentence. I, in my bureaucratic desk‑clerk persona, say, “Very good, I’ll type that into the system now.” I type it into the computer and say, “I’ll send that off to the AI for autocompletion.”
When the AI responds, I turn back and read the full sentence aloud. You wrote the first half; the AI wrote the second. I read it in my bureaucratic tone, as if we’re efficiently completing a form.
Voidspace:
The bureaucratic tone is interesting. People use LLMs for personal matters, and it feels intimate, yet there’s always that awareness of the corporation behind it — a mechanised, bureaucratic process.
Marcus:
Yes. In some sense, what we’re doing is a kind of therapy. In the fiction, people come in to write affirmations using cards; I feed them to the LLM and return the result. There’s a bureaucratic layer — like health‑insurance‑mediated therapy. I imagine an alternate reality where ChatGPT exists, but you have to go to an office to use it. You queue, sit with a low‑level bureaucrat, and they enter your query and read back the answer.
Voidspace:
At Voidspace Live you actually had a queue outside. That added to the effect.
Marcus:
Yes. Logistically I couldn’t manage the queue while performing, but I did think about having a “take a ticket” system — “Number 17, please” — and playing that up. But I had to focus on what was inside the room.
Voidspace:
You’d need a little PA system so you wouldn’t have to leave the table.
Marcus:
Exactly. Then as the LLM is working, I give a technical explanation. This is where the line blurs between me as the artist explaining my piece and me performing a scripted role. It’s a computational art piece, so there’s a technical element. I am doing something interesting with the technology, and I explain it truthfully. But for some people, it comes across as techno‑babble — “feature number 106,425” — which is fine. That’s part of our aesthetic relationship with LLMs: a deep algorithm we don’t understand.
But some participants come in who work in AI or know the tech. They understand my explanation and the novel technique I’m using to steer the LLM, and they find it interesting.
Voidspace:
I don’t come from a tech background, but I could follow enough of your explanation. It didn’t feel like techno‑babble — it added to the dramaturgy. It felt like part of the process. I’d love you to give that explanation here — the one you give in the piece.
Marcus:
To do that, I have to bring in the tarot. When I say I’m about to type your sentence into the AI, I first bring out another deck of cards and say, “I need you to pick whichever one of these you feel personally drawn to.” I never mention the word “tarot” — deliberately. In this alternate reality, tarot doesn’t exist.
You pick a card, I turn it over. At this point, the audience senses it’s tarot — the tall cards, the spread, the ritual.
But when I turn it over, it’s just text. It’s a tall, narrow tarot‑shaped card, but it only has text on it. It does have a title — The Hanged Man, The High Priestess, something like that — and it may be upright or reversed.
I say: “The sentences on that card are real sentences once written by real people, now buried deep in the database of the AI’s mind, and compressed into statistical patterns it has learned to recognise and reproduce. Our engineers have identified these patterns, curated them into a group, and assigned them a name — in this case, The High Priestess, which you’ve drawn in the reversed orientation.”
I then list the various features associated with that card draw, very matter-of-fact: feature 36,401, “concepts and discussions related to morality”; feature 26,984, “references to knowledge and its various aspects or contexts”; and so on.
This is all based on real-world work done by the AI labs around what’s called mechanistic interpretability — trying to understand what’s happening inside the “mind” of an AI. They identify features that activate consistently: a neuron that fires for family‑related text, another for medieval history, another for lists or quotations. Anything you can imagine appearing in text can have a measurable signature inside the model.
You can hone in on these thematic structures the same way MRI scans show which part of a human brain lights up when someone sees a cat versus a hippopotamus. Conceptually, it’s similar.
So I took features like family, loyalty, uncertainty, overcoming difficulties, and grouped them according to the Major Arcana. I did this by feeding descriptions of each card’s divinatory meaning into the model and seeing which features lit up. If a card was about fairness, for example, I could find the cluster of neurons associated with that.
Then I brought out the real sentences — the ones most closely associated with fairness, loyalty, new beginnings — and put them on the cards as part of the design.
Back in the performance, I say: “Great, you’ve drawn The High Priestess, reversed.” I enter that into the system in a bureaucratic way and click go. That steers the LLM’s output — not by changing the prompt, which is just the five cards you played, but by steering the algorithm. I turn up the volume on the features associated with the card you drew.
Voidspace:
So would it be fair to say — as a simple analogy — that the prompt stays the same, but you’re putting it through a different lens or filter?
Marcus:
Yes, it’s a fair simplification. These models are trained for a long time, with huge computational cost, and then released with fixed weights. When you give them text, they predict what’s likely to come next.
Steering is like popping the hood and adjusting the dials — turning up the “family loyalty” dial, for example. It’s the same model, but tweaked so that when you give it the five‑card sentence, it becomes more likely to mention themes associated with the card you drew.
Voidspace:
Clever.
Marcus:
It’s a technical process — taking a tool AI labs use for research and experimenting with artistic uses of it. And it all folds into the fiction. I explain it accurately, but in character. Some people ask afterwards whether I was actually doing what I said or whether it was just scripted. I am doing it — but from the audience’s perspective, they can’t know, and maybe it doesn’t matter. But it does affect the output. I can tell which card was drawn based on the generated text.
Voidspace:
That makes perfect sense — you’ve reverse‑engineered it. I’d argue it’s like any magic trick: huge amounts of work go into creating the sense of ease. Any good art — dance, theatre, writing — feels effortless because of the vast machinery behind it. The computational processes and the study you’ve done contribute hugely. The output feels uncanny — the synchrony between randomness, the card, and the LLM.
I participated in this experience with a friend and we could both apply it to our lives. It felt uncanny, even knowing some of the process behind it.
There’s something fascinating about how tarot ties it together — or mirrors the whole piece. It combines randomness, emotion, human‑controlled symbolism, and divination. It gives that uncanny “higher forces” feeling we sometimes get when interacting with ChatGPT.
Marcus:
Yes — and with tarot, you bring your own understanding and project your own context onto the card. If you turn over The Tower — of all cards — there’s a moment in your mind that’s very interesting.
Voidspace:
Just to gloss for people unfamiliar with tarot: The Tower is the most extreme expression of doom and destruction. Many cards can be read positively or negatively, but The Tower is traditionally “oh shit!” So that moment — of all the cards in all the world, you picked that one — has emotional impact.
Marcus:
Exactly. I want to mention poker, too.
In Texas Hold’em, you have your two secret cards, and then the five community cards come out: the flop (three cards), the turn, and the river. There are moments where a single card — say the ace of diamonds — completely changes the hand. The ace of diamonds itself means nothing, but in context it can be life‑or‑death for the hand.
It’s that reveal — turning over a card and triggering a cascade of associations in your mind and your opponent’s. In poker you must hide your reaction, but the moment is still there.
In my piece, you’ve got three cards: “I’m afraid of my future because I’m not getting better,” and then you turn over the fourth card. It takes the sentence in a new direction. That moment of reveal is powerful.
Voidspace:
I love that. You can feel the energy in that moment.
Another element of sent(i)ence is the use of CAPTCHA
Marcus:
It’s a tarot‑adjacent process. After the tarot card, I type your sentence into the computer and say, “I’ll send that off and we’ll get the response in a moment.” Then my face falls and I say, “Sorry, we have a strict security policy. It’s asking me to prove that you’re human before we can continue.”
Before the machine will complete your personal statement, it asks you to prove you’re human.
Voidspace:
There’s something beautiful about that.
Marcus:
And it becomes a parallel track. You’re writing your affirmation, getting responses, constructing meaning — and in parallel, you’re subjected to a series of tests. Increasingly demanding, increasingly repetitive. You can’t get anywhere without constantly proving you’re human. Two‑factor authentication. Absurd, existential tests.
Voidspace:
And the nature of the tests is interesting — not just traffic lights. Could you outline some examples?
Marcus:
The first is a CAPTCHA. Normally you’d click on all the traffic lights. Instead, I give you a grid of images — all words. They look like Streetview signs, but they’re just words. The prompt is: “select all the words that remind you of your mother”.
Voidspace:
Nice little Blade Runner reference.
Marcus:
Yes — referencing the Voight‑Kampff test. Some words are “hospital,” “knitting,” “Scrabble,” “sailing,” and others like “sad,” “kind,” “angry,” “lonely.” You pick the ones that remind you of your mother. Technically it’s a trick — anything you choose is accepted.
Voidspace:
But the point is the impact on you.
Marcus:
Yes. In terms of the real underlying technical process, this part is fictional, but the audience plays along in good faith. Then you’re accepted — until I say, with visible frustration, “Now it’s asking for two‑factor authentication,” and hand it to your partner.
They do the same test. Their associations may be different. It’s humans projecting meaning onto words. Some people have positive feelings about their mother, some negative, some complicated. Whatever “mother” means to you.
That’s the first test — familiar but twisted. It usually gets a laugh: the absurdity of a physical CAPTCHA, and the emotional twist.
They get more unhinged as they go — less like recognisable CAPTCHAs, more like: what does it mean to be human?
Voidspace:
It added so much to the piece. Genius.
Marcus:
During development, the captchas started as a lighthearted gag — “Wouldn’t it be ironic if you had to prove you’re human?” — but it grew into something as important as the affirmation itself. It’s the theatre of the piece.
Voidspace:
The dramaturgy.
Marcus:
Yes — the structure, the climax. It builds toward that.
Voidspace:
I agree. It becomes an emotional journey.
Another thing I noticed it the way that you introduce friction to the process. You could type the words into the AI and get an instant response, but instead you build friction — imaginative overlay, bureaucratic barriers, CAPTCHAs. It creates time, texture, and theatricality.
Marcus:
Definitely. At the start, I say we’re writing an affirmation that will be sent off to an AI “in the interests of efficiency.” But over the next 20 minutes, it becomes wildly inefficient — constant proof‑of‑humanity, two‑factor authentication, bureaucratic hurdles.
Voidspace:
And a little physical printer, right?
Marcus:
Yes. At the end, I give you a printout of your affirmation on a card and say you can take it to the Affirmation Application Department.
Voidspace:
So it becomes increasingly obvious how inefficient it really is.
Marcus:
Yes. Everything gets in the way of writing this personal expression of self‑identity. You’re even forced to do it two people at a time — “for efficiency.” And some people want to see what’s on the screen, but you never do. That’s deliberate. I type it on your behalf and tell you what the AI says.
The film Brazil was a strong influence. That alternate‑reality sci‑fi bureaucracy: high‑tech but clunky, physical, absurd.
Voidspace:
What I loved is that the whole process made the AI feel like an alternate‑universe machine. Not a bot you type into, but something with gears and levers and typewriter keys crunching away. It slows the process so you can appreciate what’s happening under the hood — the labour, the resource consumption, the machinery. It resists the illusion of magic.
You see the mechanism in slow motion.
Marcus:
Yes. And the mechanic of playing with the cards is meant to be a physical metaphor for the AI’s internal workings — not literal, but aesthetically resonant.
Using AI in art is contentious. Some people have strong negative opinions. I’ve had people seem interested until I mention AI, and then the shutters come down.
I understand the various reasons why. A lot of AI use is banal, and I suppose every artist using it thinks their use isn’t banal. But if you’re making art that comments on AI, I think that’s a valid use.
Voidspace:
100%. At the festival, I programmed three pieces that do this — yours, Mikey’s, and Trainwreck. For me these were valuable additions to the programme because they use AI cleverly, in bespoke ways, as cultural commentary.
Marcus:
We have to make art that comments on AI. Ignoring it isn’t a valid response. You can choose not to use AI personally, but as an artist, my practice is to look at the world we live in and our relationship to AI. Not the technology itself, but the human relationship to it.
Voidspace:
You’re interrogating it.
Marcus:
Yes. By pulling it apart, adding friction, letting you see the gears clunking. Not treating it as a magic black box, but showing the human and the AI interacting in a side-by-side, step-by-step process — almost getting in each other’s way, stepping on each other’s toes.
In sent(i)ence, one person plays a card, then the other, then one, then the other, then the AI generates something. Then one person again, then the other, then the AI. It’s not a chatbot — not “here’s a prompt, give me a response.” It’s stage‑by‑stage construction. My previous piece, Conversation(AI)symmetry, worked the same way: two humans generating something word by word, block by block, in parallel with an AI, in a deliberately clunky, friction‑filled way.
Voidspace:
Something that helps create that interrogation in both pieces is the theatricality: the performance art, the character, the world‑building around the computational mechanics. I’d love to hear your developing thoughts on the theatricality of sen(i)ence: The timing of the beats, being in character, creating a whole world around the piece — it adds so much.
Marcus:
When I went back to do my Master’s in Computational Art, I’d been working in tech for a long time and wanted to return to study. Throughout, I wondered how to bring in my filmmaking background. I struggled with it — I thought of filmmaking as a screen‑based, linear, time‑based medium, and I wasn’t sure how to incorporate that.
But as I developed sent(i)ence, I realised it wasn’t the screen‑based part I needed — it was the set design, the lighting, the scripting, the world‑building. That’s where filmmaking fed in.
And in terms of performance: I’m not an actor; I’d never acted before. I was hesitant. At first, I didn’t think of myself as playing a character — I thought, “I’m the artist, I’ll introduce the piece, it’ll be scripted because it has to be.”
But as it developed, I realised it was a character. The bank clerk at a loan appointment is playing the character of a bank clerk. A tarot reader plays the character of a tarot reader. An artist at an art show plays the character of an artist.
It also relates to AI and LLMs. LLMs are told to play a character. They’re given system prompts you never see. Anthropic even released Claude’s — essentially a 30‑page character sheet describing who Claude is. The model roleplays that character. Claude, more than ChatGPT or others, is given a specific personality.
Voidspace:
And that brings us right back to the overarching theme you’re interrogating — the nature of the “I” – in these processes, and in these machines.
Marcus:
Exactly. The text that Claude generates is written in the first person. You can ask, who is that “I” that is speaking? I use that as a launching-off point to ask a more human question: when I speak, who is the “I” that is speaking?
Voidspace:
A beautiful way to close the loop. What’s next for your practice?
Marcus:
I’d like to explore more freeform, larger immersive shows. I think sent(i)ence could be part of a world — a larger space you wander through, with one room containing the poker‑dealer setup and other rooms containing other things. In an ideal world with infinite money, that’s what I’d do.
In a world without infinite money, I think I’ll keep exploring gameful installations where people play with AI, with themselves, with language — and they have fun doing it.
Voidspace:
Do you have any advice for an aspiring creator in your field?
Marcus:
If you’re unsure of your artistic voice, unsure what you want to say — that uncertainty is the work.
You don’t need to know the answer before you make the art. What you’re creating isn’t the answer; it’s the question. The thing you’re unsure about is exactly what the piece is about. You don’t find the answer and make art about the answer — you find the question and make art about the question.
Voidspace:
The value of questioning and interrogation: that’s the perfect note to end it on. Thank you.