Gen AI and the Arts: A Conversation with Ina Conradi
by Megan Wee JX
Ina Conradi · Associate Professor, NTU School of Art, Design and Media
Written responses to The Straits Times — Impact of GenAI on the local arts scene Submitted: 1 June 2026
How should artists view Gen AI?
I’d ask them to resist the word ‘view’ — as if GenAI were a painting on a wall you either like or don’t. It is more useful to think of it as a condition you are already working inside, whether you choose to engage with it or not.
I was at AI on the Lot in Los Angeles last week — the largest annual conference for AI in the entertainment industry, 2,000 professionals from Hollywood studios, tech companies, and media startups at Culver Studios. I presented on a panel called Experience is the Interface: from screens to spaces, AI is transforming how audiences encounter stories. What struck me most was not the technology on display. It was the quality of the questions people were asking. The most interesting practitioners in that room were not asking ‘which tool should I use?’ They were asking ‘what does this make possible that wasn’t possible before?’ That is the attitude I’d encourage.
For working artists: go into the studio with it. Not to use it as a shortcut, but to find out where it resists you, where it surprises you, and what it cannot do that you still can. For aspiring artists: learn the craft first, deeply. The value of AI in any creative process scales with how much genuine artistic thinking you bring to it. Your taste, your conceptual rigour, your embodied sense of what a piece needs — none of that is replaceable, and all of it is what makes the difference.
What about the ethical arguments against GenAI in art?
The ‘stealing’ argument is worth examining carefully, because the reality is more layered than the headline suggests. Training data ethics are a genuine concern — whose work was used to build a model’s weights, under what conditions, with what consent — and I take that seriously. It is why in my studio we pay close attention to how we use these tools — prioritising workflows that involve substantial transformation and iteration, whether through ComfyUI pipelines or other platforms, rather than accepting first outputs. The provenance of the underlying model matters, and so does the creative distance between source and final work.
But the relationship between a model’s training data and any specific output is not straightforward reproduction. Working through multiple iterations — conditioning, transformation, inpainting, spatial recomposition — creates genuine creative distance from any source. By the time a piece has moved through my own workflow, my own prompting logic, my own compositional decisions, the connection to any training image is remote in the way that a painter’s connection to every artwork they have ever looked at is remote. Influence is not theft.
The position I find most defensible — and most interesting — is to train on your own archive. I am currently working in exactly that direction: taking my own earlier works and using AI to extend, enhance, and adapt them for new venues and scales. That is not appropriation. That is an art practice in dialogue with itself.
Which brings me to the second argument — that art requires human creation to evoke emotion — because that is really the more interesting question. Art history is full of works made with machines, with assistants, with processes the maker didn’t fully control — and we don’t strip them of their status as art. Vermeer, the 17th-century Dutch master whose luminous interiors are among the most reproduced paintings in the world, is the most instructive case. Art historians now widely believe he used a camera obscura — projecting the scene onto canvas before painting — to achieve his extraordinary precision. When that theory first gained serious attention it felt scandalous, like an accusation of cheating. Today nobody cares. What matters is the quality and presence of the work, not the purity of the method. The light doesn’t become less luminous because we know how he achieved it. The paintings are still the paintings. The light is still the light.
A generation later, Harold Cohen, a British artist who spent decades from the 1970s onward building AARON — one of the first computer programs to autonomously generate original drawings and paintings — exhibited its output as his own artistic practice for over forty years. At the time, that too was controversial. Now it is art history. What makes work meaningful has never been the purity of the method. It has always been the presence of intention, judgement, and something genuinely at stake.
I have seen AI-generated work move audiences, at scale, in my own practice. Echoes, Whispers and Memories — built substantially with AI-generated imagery — has been experienced by millions: at Ars Electronica in Linz, on the West Lake Media Façade in Hangzhou, and in the 270-degree immersive Sala Immersiva at MEET Digital Culture Center in Milan. The work lands. The emotion is real. Paul Schrader, whose keynote at a conference I attended in Los Angeles reflected on a lifetime of directing films, was circling this point about intention — but it applies equally to every artist working in any medium.
The AI doesn’t care about the story, the meaning, or what is at stake. The artist does. That caring is what makes it resonate.
Do you think a career in the arts is at risk because of GenAI?
I understand the anxiety. But I think it is misdirected — and the more accurate anxiety is different, and more productive.
The roles most at risk are specific, often outsourced tasks at the commercial end of the creative industries: stock illustration, template-based design, certain kinds of commercial concept art. For fine artists the threat is different, and more personal — it is the fear that what made your voice singular, your visual language recognisable, your practice worth building, can now be approximated by anyone with a good prompt. That fear is understandable. But it mistakes style for practice.
A machine can learn to approximate the surface of a visual language. It cannot replicate the decade of looking, failing, questioning, and revising that produced it. It cannot replicate the specific cultural inheritance you bring to a canvas, a screen, or a public space. It cannot replicate what is genuinely at stake for you in the work — the question that keeps you returning to the studio when there is no external reason to. That interiority is not a minor detail. It is the whole thing.
What I have come to believe — and what I saw reflected in conversations in Los Angeles last week — is that AI does not diminish the fine artist’s role. It clarifies it. It strips away the parts of practice that were always somewhat mechanical, and leaves the part that was always irreplaceable: the vision, the question, the particular way of seeing that no training data can fully contain.
My advice: don’t compete with AI on its own terms. Go deeper — into the questions only you can ask, the material only your specific history can unlock, the work only your particular stubbornness can bring into existence. That is where the practice is. That is where it has always been.
You describe AI as having three distinct roles: generative, interpretive, and as an object of study. Can you explain the last two?
As a generative tool — the familiar territory — AI translates a concept or prompt into visual, textual, or sonic output. That is where most of the public conversation stays. The other two roles are where I think the more interesting and more necessary work is happening.
As an interpretive tool, AI is used not to generate images but to interrogate existing material. In my courses, students use large language models to work through scientific papers, datasets, and complex theoretical texts — asking the model to explain, summarise, find contradictions, suggest analogies. We have also worked with scientific imagery — in one recent project, images of invisible plankton from ocean research, using AI to help identify visual patterns and suggest translations into moving image work that the human eye could never assemble from that volume of material alone. The AI here is a thinking partner, not a maker.
As an object of study — the role I think is most undertaught in Singapore right now — we turn the lens around and examine AI itself. This means asking three distinct questions that I think every student working with these tools should be able to answer.
The first is: where does it fail, and what does its training data reveal about cultural bias? Ask any mainstream image generator to produce ‘a scientist’ or ‘a CEO’ and count how many times it defaults to a white man in a lab coat or a suit. Ask it to generate ‘a traditional wedding’ and see whose tradition it assumes. The model is not neutral — it reflects the demographics and aesthetic preferences of the dataset it was trained on, which skews heavily toward Western, English-language, commercially available internet imagery. When students from Singapore, Indonesia, or India use these tools, they are working inside a system that was not built with their visual culture as the default. That is worth knowing explicitly, not just feeling vaguely.
The second question is: how does it represent uncertainty? This connects directly to my own research using art to communicate quantum physics. AI image generators are confidence machines — they always produce a complete, resolved, visually coherent image. They cannot say ‘I don’t know’ or ‘this is unclear.’ But reality — and certainly quantum reality — is full of uncertainty, ambiguity, and incompleteness. That gap — between the model’s false confidence and the genuine uncertainty of the physics — is something my students and I work with directly.
The third question is: what does synthetic imagery do to our sense of documentary truth? During the LA wildfires earlier this year, AI-generated images of burning neighbourhoods circulated on social media alongside real photographs, and some were impossible to distinguish at a glance. When students deploy AI-generated imagery in public contexts — on urban screens, in galleries, in media — they are operating in that landscape. Understanding what synthetic imagery does to trust, to witness, to the evidentiary status of images is not a theoretical concern. It is a practical and ethical one.
Most AI art education currently focuses on the first role — generative — because that is what produces visible, shareable results quickly. But the critical role — examining the system itself — requires slowing down, sitting with discomfort, and asking questions that don’t have easy answers. It is the role I am most committed to developing, because I think it is where artists can make their most distinctive and lasting contribution to how society understands and navigates AI.
What does it mean to think and create within AI-driven cultural systems?
It means understanding that you are not just choosing a tool — you are stepping into a cultural system that has preferences, biases, aesthetics, and assumptions already baked into it. Every generative model has been trained on a corpus that reflects particular choices about what was included, and why. Working thoughtfully within that system means knowing those tendencies well enough to work with them, against them, or around them — rather than accepting whatever the model offers.
In my own work, this plays out at a very physical scale. The panel I recently presented in Los Angeles, Experience is the Interface, argued that when AI moves from screens into physical space — floors that ripple around your feet, façades that wrap a city block — the interface becomes the human body itself. The audience is no longer outside the system. They are part of it. That is not a metaphor. It is a design decision that has to be made by a human being who understands what is at stake in that encounter.
In the classroom, a student who types a prompt and accepts the first output isn’t creating within an AI-driven cultural system — they’re being carried by one. A student who iterates, argues back, applies their own aesthetic judgement at every stage, and understands the cultural implications of what they’re producing — that student is directing the system. That is the distinction I try to teach.
What are the dangers of the blanket term ‘AI art’?
Getting this vocabulary right is not an academic exercise — it has real consequences for working artists. Calling everything ‘AI art’ is a little like calling everything made with a brush ‘brush art’ — it tells you nothing about the work, and it flattens profound differences in skill, intention, and labour into a single unhelpful label.
Consider the difference between these scenarios: an artist who uses AI at the brainstorming stage to surface visual connections, then builds the final work entirely by hand — as with my 2023 animated film Moirai, traditional 3D animation frame by frame, with AI used only in early moodboarding; an artist who uses AI to generate a palette of imagery they then edit, composite, and transform; and an artist who types a prompt and publishes the first output. All three might be called ‘AI art.’ They represent profoundly different relationships to authorship, skill, and creative labour.
The art world has navigated versions of this before. When photography emerged, the question of whether it was ‘real art’ consumed decades of critical energy — energy that could have been spent looking more carefully at what photographers were actually doing and how differently they were doing it. We are repeating that pattern now.
One of the things my students consistently report surprises them most is how much work serious AI-assisted practice actually requires. They come in expecting a shortcut and discover instead a new kind of rigour — hundreds of iterations, each one a decision, each one requiring them to know clearly enough what they want to recognise when they are getting closer to it or further away. The tool does not do the wanting. The artist does. And the wanting, it turns out, is most of the work.
Should there be thresholds for acceptable GenAI usage in art?
I don’t think the question of ‘acceptable thresholds’ is the right frame. It implies a purity spectrum — more AI is worse, less AI is better — and I don’t believe that. A work with 90% AI-generated imagery can be deeply meaningful if the artistic vision and conceptual rigour are present. A work with 1% AI involvement can be cynical and empty. Percentage is not the measure.
What matters to me, both ethically and aesthetically, is: Is there genuine artistic intention driving this work? Are choices being made, or accepted by default? Is the work asking a real question? Has the artist thought about the provenance of the training data?
On the specific question of public deployment — galleries, public commissions, published work — I do think the standards are different and the disclosure obligations are higher. Audiences have a right to know what they are looking at and what kind of labour produced it. A recent and instructive example is Hell Grind — billed as the world’s first fully AI-generated feature film, made by a team of 15 people in fourteen days for $500,000, and screened at an industry event in Cannes during the 2026 festival. Its producers described it as having ‘premiered at Cannes’ — which was technically true of the city, but not of the festival itself. That blurring — deliberate or not — is precisely the problem. The ethics of the private studio and the ethics of public presentation are not the same thing, and conflating them leads to bad conversations and worse policy.
When did you launch your AI art courses at NTU, and how have they evolved?
Both courses — DM2012 Explorations in AI-Generated Art (undergraduate) and AP7055 Art in the Age of the Creative Machine (postgraduate) — launched in 2022 as part of the NTU NISTH AI in Art Studio research project, and have run every semester since. They were among the earliest dedicated AI art courses at any university in Southeast Asia.
Looking back, the conceptual foundation was already Jungian before we named it as such. Carl Jung described a practice he called Active Imagination — consciously engaging with the unconscious while remaining in dialogue rather than seeking control. It describes remarkably well what serious AI studio practice actually is. You enter a system you don’t fully understand, you stay curious rather than directive, and you allow the unexpected to surface as material rather than suppressing it as error.
The earliest iterations of the courses explored exactly that territory — dreams, memory, and the unconscious as subject matter, students learning to work with generative AI the way an analyst works with a dream: not to master it, but to be in genuine conversation with it. Over time that inquiry expanded — from dreams and memory into ocean science, quantum physics, and the mathematics of entanglement. Students from these courses have since gone on to graduate programmes at Carnegie Mellon University, New York University, Parsons, and Pratt.
Who takes these courses, and what is the atmosphere in the classroom?
Both courses consistently fill now — but that was not always the case, and I think the honest version of this answer has to include where we started.
In the first semesters, students would arrive with questions I could not answer. I was not standing at the front of the room as an expert — I was there as an artist who was genuinely, almost overwhelmingly excited by something I did not yet fully understand. I am trained as a weaver and tapestry maker — physically demanding handwork that occupied my twenties and into my mid-thirties, thread by thread, before I moved into media art. That background never left me. And for me, typing a prompt and receiving an image — no matter how low-resolution, how choppy, how undefined — felt like a miracle. The same instinct that makes you hold your breath when the first threads of a pattern begin to emerge on a loom.
Not everyone shared that excitement. Some students walked out in protest — they had serious ethical objections, genuine artistic convictions about what making meant, and they were not wrong to raise them. I gave them the freedom to leave and to question. I think that freedom mattered. The ones who stayed did so by choice, not by compliance — and some of them made work in those early semesters that still brings tears, not for technical achievement but for the stories they found the courage to tell through this strange new medium.
Today the mix includes students from engineering, social sciences, business, and the sciences alongside ADM students, each bringing a different kind of scepticism and a different kind of wonder. The non-arts students often have sharper instincts for AI’s technical limits; the arts students often have sharper instincts for its aesthetic and ethical dimensions. They teach each other.
In 2023 we became the first university from Singapore to participate in the Ars Electronica Campus Exhibition — alongside 56 international universities, before an audience of 88,000 visitors and 1,542 artists, scientists, and designers from 88 countries. The exhibition, Butterfly’s Dreams: The New Aesthetic of AI in Artistic Practice, was presented under the festival theme Who Owns the Truth? — which could not have been more precisely on point for what we were doing in the studio.
How can arts education help students develop the language to question what they are making?
Arts education has always been, at its core, about teaching people to ask harder questions of the work in front of them — including their own. What has changed is the urgency, and the specific vocabulary needed.
Gen AI and the Arts: A Conversation with Ina Conradi
Ina Conradi · Associate Professor, NTU School of Art, Design and Media
Written responses to The Straits Times — Impact of GenAI on the local arts scene Submitted: 1 June 2026
How should artists view Gen AI?
I’d ask them to resist the word ‘view’ — as if GenAI were a painting on a wall you either like or don’t. It is more useful to think of it as a condition you are already working inside, whether you choose to engage with it or not.
I was at AI on the Lot in Los Angeles last week — the largest annual conference for AI in the entertainment industry, 2,000 professionals from Hollywood studios, tech companies, and media startups at Culver Studios. I presented on a panel called Experience is the Interface: from screens to spaces, AI is transforming how audiences encounter stories. What struck me most was not the technology on display. It was the quality of the questions people were asking. The most interesting practitioners in that room were not asking ‘which tool should I use?’ They were asking ‘what does this make possible that wasn’t possible before?’ That is the attitude I’d encourage.
For working artists: go into the studio with it. Not to use it as a shortcut, but to find out where it resists you, where it surprises you, and what it cannot do that you still can. For aspiring artists: learn the craft first, deeply. The value of AI in any creative process scales with how much genuine artistic thinking you bring to it. Your taste, your conceptual rigour, your embodied sense of what a piece needs — none of that is replaceable, and all of it is what makes the difference.
What about the ethical arguments against GenAI in art?
The ‘stealing’ argument is worth examining carefully, because the reality is more layered than the headline suggests. Training data ethics are a genuine concern — whose work was used to build a model’s weights, under what conditions, with what consent — and I take that seriously. It is why in my studio we pay close attention to how we use these tools — prioritising workflows that involve substantial transformation and iteration, whether through ComfyUI pipelines or other platforms, rather than accepting first outputs. The provenance of the underlying model matters, and so does the creative distance between source and final work.
But the relationship between a model’s training data and any specific output is not straightforward reproduction. Working through multiple iterations — conditioning, transformation, inpainting, spatial recomposition — creates genuine creative distance from any source. By the time a piece has moved through my own workflow, my own prompting logic, my own compositional decisions, the connection to any training image is remote in the way that a painter’s connection to every artwork they have ever looked at is remote. Influence is not theft.
The position I find most defensible — and most interesting — is to train on your own archive. I am currently working in exactly that direction: taking my own earlier works and using AI to extend, enhance, and adapt them for new venues and scales. That is not appropriation. That is an art practice in dialogue with itself.
Which brings me to the second argument — that art requires human creation to evoke emotion — because that is really the more interesting question. Art history is full of works made with machines, with assistants, with processes the maker didn’t fully control — and we don’t strip them of their status as art. Vermeer, the 17th-century Dutch master whose luminous interiors are among the most reproduced paintings in the world, is the most instructive case. Art historians now widely believe he used a camera obscura — projecting the scene onto canvas before painting — to achieve his extraordinary precision. When that theory first gained serious attention it felt scandalous, like an accusation of cheating. Today nobody cares. What matters is the quality and presence of the work, not the purity of the method. The light doesn’t become less luminous because we know how he achieved it. The paintings are still the paintings. The light is still the light.
A generation later, Harold Cohen, a British artist who spent decades from the 1970s onward building AARON — one of the first computer programs to autonomously generate original drawings and paintings — exhibited its output as his own artistic practice for over forty years. At the time, that too was controversial. Now it is art history. What makes work meaningful has never been the purity of the method. It has always been the presence of intention, judgement, and something genuinely at stake.
I have seen AI-generated work move audiences, at scale, in my own practice. Echoes, Whispers and Memories — built substantially with AI-generated imagery — has been experienced by millions: at Ars Electronica in Linz, on the West Lake Media Façade in Hangzhou, and in the 270-degree immersive Sala Immersiva at MEET Digital Culture Center in Milan. The work lands. The emotion is real. Paul Schrader, whose keynote at a conference I attended in Los Angeles reflected on a lifetime of directing films, was circling this point about intention — but it applies equally to every artist working in any medium.
The AI doesn’t care about the story, the meaning, or what is at stake. The artist does. That caring is what makes it resonate.
Do you think a career in the arts is at risk because of GenAI?
I understand the anxiety. But I think it is misdirected — and the more accurate anxiety is different, and more productive.
The roles most at risk are specific, often outsourced tasks at the commercial end of the creative industries: stock illustration, template-based design, certain kinds of commercial concept art. For fine artists the threat is different, and more personal — it is the fear that what made your voice singular, your visual language recognisable, your practice worth building, can now be approximated by anyone with a good prompt. That fear is understandable. But it mistakes style for practice.
A machine can learn to approximate the surface of a visual language. It cannot replicate the decade of looking, failing, questioning, and revising that produced it. It cannot replicate the specific cultural inheritance you bring to a canvas, a screen, or a public space. It cannot replicate what is genuinely at stake for you in the work — the question that keeps you returning to the studio when there is no external reason to. That interiority is not a minor detail. It is the whole thing.
What I have come to believe — and what I saw reflected in conversations in Los Angeles last week — is that AI does not diminish the fine artist’s role. It clarifies it. It strips away the parts of practice that were always somewhat mechanical, and leaves the part that was always irreplaceable: the vision, the question, the particular way of seeing that no training data can fully contain.
My advice: don’t compete with AI on its own terms. Go deeper — into the questions only you can ask, the material only your specific history can unlock, the work only your particular stubbornness can bring into existence. That is where the practice is. That is where it has always been.
You describe AI as having three distinct roles: generative, interpretive, and as an object of study. Can you explain the last two?
As a generative tool — the familiar territory — AI translates a concept or prompt into visual, textual, or sonic output. That is where most of the public conversation stays. The other two roles are where I think the more interesting and more necessary work is happening.
As an interpretive tool, AI is used not to generate images but to interrogate existing material. In my courses, students use large language models to work through scientific papers, datasets, and complex theoretical texts — asking the model to explain, summarise, find contradictions, suggest analogies. We have also worked with scientific imagery — in one recent project, images of invisible plankton from ocean research, using AI to help identify visual patterns and suggest translations into moving image work that the human eye could never assemble from that volume of material alone. The AI here is a thinking partner, not a maker.
As an object of study — the role I think is most undertaught in Singapore right now — we turn the lens around and examine AI itself. This means asking three distinct questions that I think every student working with these tools should be able to answer.
The first is: where does it fail, and what does its training data reveal about cultural bias? Ask any mainstream image generator to produce ‘a scientist’ or ‘a CEO’ and count how many times it defaults to a white man in a lab coat or a suit. Ask it to generate ‘a traditional wedding’ and see whose tradition it assumes. The model is not neutral — it reflects the demographics and aesthetic preferences of the dataset it was trained on, which skews heavily toward Western, English-language, commercially available internet imagery. When students from Singapore, Indonesia, or India use these tools, they are working inside a system that was not built with their visual culture as the default. That is worth knowing explicitly, not just feeling vaguely.
The second question is: how does it represent uncertainty? This connects directly to my own research using art to communicate quantum physics. AI image generators are confidence machines — they always produce a complete, resolved, visually coherent image. They cannot say ‘I don’t know’ or ‘this is unclear.’ But reality — and certainly quantum reality — is full of uncertainty, ambiguity, and incompleteness. That gap — between the model’s false confidence and the genuine uncertainty of the physics — is something my students and I work with directly.
The third question is: what does synthetic imagery do to our sense of documentary truth? During the LA wildfires earlier this year, AI-generated images of burning neighbourhoods circulated on social media alongside real photographs, and some were impossible to distinguish at a glance. When students deploy AI-generated imagery in public contexts — on urban screens, in galleries, in media — they are operating in that landscape. Understanding what synthetic imagery does to trust, to witness, to the evidentiary status of images is not a theoretical concern. It is a practical and ethical one.
Most AI art education currently focuses on the first role — generative — because that is what produces visible, shareable results quickly. But the critical role — examining the system itself — requires slowing down, sitting with discomfort, and asking questions that don’t have easy answers. It is the role I am most committed to developing, because I think it is where artists can make their most distinctive and lasting contribution to how society understands and navigates AI.
What does it mean to think and create within AI-driven cultural systems?
It means understanding that you are not just choosing a tool — you are stepping into a cultural system that has preferences, biases, aesthetics, and assumptions already baked into it. Every generative model has been trained on a corpus that reflects particular choices about what was included, and why. Working thoughtfully within that system means knowing those tendencies well enough to work with them, against them, or around them — rather than accepting whatever the model offers.
In my own work, this plays out at a very physical scale. The panel I recently presented in Los Angeles, Experience is the Interface, argued that when AI moves from screens into physical space — floors that ripple around your feet, façades that wrap a city block — the interface becomes the human body itself. The audience is no longer outside the system. They are part of it. That is not a metaphor. It is a design decision that has to be made by a human being who understands what is at stake in that encounter.
In the classroom, a student who types a prompt and accepts the first output isn’t creating within an AI-driven cultural system — they’re being carried by one. A student who iterates, argues back, applies their own aesthetic judgement at every stage, and understands the cultural implications of what they’re producing — that student is directing the system. That is the distinction I try to teach.
What are the dangers of the blanket term ‘AI art’?
Getting this vocabulary right is not an academic exercise — it has real consequences for working artists. Calling everything ‘AI art’ is a little like calling everything made with a brush ‘brush art’ — it tells you nothing about the work, and it flattens profound differences in skill, intention, and labour into a single unhelpful label.
Consider the difference between these scenarios: an artist who uses AI at the brainstorming stage to surface visual connections, then builds the final work entirely by hand — as with my 2023 animated film Moirai, traditional 3D animation frame by frame, with AI used only in early moodboarding; an artist who uses AI to generate a palette of imagery they then edit, composite, and transform; and an artist who types a prompt and publishes the first output. All three might be called ‘AI art.’ They represent profoundly different relationships to authorship, skill, and creative labour.
The art world has navigated versions of this before. When photography emerged, the question of whether it was ‘real art’ consumed decades of critical energy — energy that could have been spent looking more carefully at what photographers were actually doing and how differently they were doing it. We are repeating that pattern now.
One of the things my students consistently report surprises them most is how much work serious AI-assisted practice actually requires. They come in expecting a shortcut and discover instead a new kind of rigour — hundreds of iterations, each one a decision, each one requiring them to know clearly enough what they want to recognise when they are getting closer to it or further away. The tool does not do the wanting. The artist does. And the wanting, it turns out, is most of the work.
Should there be thresholds for acceptable GenAI usage in art?
I don’t think the question of ‘acceptable thresholds’ is the right frame. It implies a purity spectrum — more AI is worse, less AI is better — and I don’t believe that. A work with 90% AI-generated imagery can be deeply meaningful if the artistic vision and conceptual rigour are present. A work with 1% AI involvement can be cynical and empty. Percentage is not the measure.
What matters to me, both ethically and aesthetically, is: Is there genuine artistic intention driving this work? Are choices being made, or accepted by default? Is the work asking a real question? Has the artist thought about the provenance of the training data?
On the specific question of public deployment — galleries, public commissions, published work — I do think the standards are different and the disclosure obligations are higher. Audiences have a right to know what they are looking at and what kind of labour produced it. A recent and instructive example is Hell Grind — billed as the world’s first fully AI-generated feature film, made by a team of 15 people in fourteen days for $500,000, and screened at an industry event in Cannes during the 2026 festival. Its producers described it as having ‘premiered at Cannes’ — which was technically true of the city, but not of the festival itself. That blurring — deliberate or not — is precisely the problem. The ethics of the private studio and the ethics of public presentation are not the same thing, and conflating them leads to bad conversations and worse policy.
When did you launch your AI art courses at NTU, and how have they evolved?
Both courses — DM2012 Explorations in AI-Generated Art (undergraduate) and AP7055 Art in the Age of the Creative Machine (postgraduate) — launched in 2022 as part of the NTU NISTH AI in Art Studio research project, and have run every semester since. They were among the earliest dedicated AI art courses at any university in Southeast Asia.
Looking back, the conceptual foundation was already Jungian before we named it as such. Carl Jung described a practice he called Active Imagination — consciously engaging with the unconscious while remaining in dialogue rather than seeking control. It describes remarkably well what serious AI studio practice actually is. You enter a system you don’t fully understand, you stay curious rather than directive, and you allow the unexpected to surface as material rather than suppressing it as error.
The earliest iterations of the courses explored exactly that territory — dreams, memory, and the unconscious as subject matter, students learning to work with generative AI the way an analyst works with a dream: not to master it, but to be in genuine conversation with it. Over time that inquiry expanded — from dreams and memory into ocean science, quantum physics, and the mathematics of entanglement. Students from these courses have since gone on to graduate programmes at Carnegie Mellon University, New York University, Parsons, and Pratt.
Who takes these courses, and what is the atmosphere in the classroom?
Both courses consistently fill now — but that was not always the case, and I think the honest version of this answer has to include where we started.
In the first semesters, students would arrive with questions I could not answer. I was not standing at the front of the room as an expert — I was there as an artist who was genuinely, almost overwhelmingly excited by something I did not yet fully understand. I am trained as a weaver and tapestry maker — physically demanding handwork that occupied my twenties and into my mid-thirties, thread by thread, before I moved into media art. That background never left me. And for me, typing a prompt and receiving an image — no matter how low-resolution, how choppy, how undefined — felt like a miracle. The same instinct that makes you hold your breath when the first threads of a pattern begin to emerge on a loom.
Not everyone shared that excitement. Some students walked out in protest — they had serious ethical objections, genuine artistic convictions about what making meant, and they were not wrong to raise them. I gave them the freedom to leave and to question. I think that freedom mattered. The ones who stayed did so by choice, not by compliance — and some of them made work in those early semesters that still brings tears, not for technical achievement but for the stories they found the courage to tell through this strange new medium.
Today the mix includes students from engineering, social sciences, business, and the sciences alongside ADM students, each bringing a different kind of scepticism and a different kind of wonder. The non-arts students often have sharper instincts for AI’s technical limits; the arts students often have sharper instincts for its aesthetic and ethical dimensions. They teach each other.
In 2023 we became the first university from Singapore to participate in the Ars Electronica Campus Exhibition — alongside 56 international universities, before an audience of 88,000 visitors and 1,542 artists, scientists, and designers from 88 countries. The exhibition, Butterfly’s Dreams: The New Aesthetic of AI in Artistic Practice, was presented under the festival theme Who Owns the Truth? — which could not have been more precisely on point for what we were doing in the studio.
How can arts education help students develop the language to question what they are making?
Arts education has always been, at its core, about teaching people to ask harder questions of the work in front of them — including their own. What has changed is the urgency, and the specific vocabulary needed.
In practice, we don’t develop this through reading lists or lectures on theory — we develop it through making and looking. We look closely at what AI actually does: where it fails, what its aesthetic defaults reveal, whose visual culture it centres and whose it ignores. And we ask students constantly — in crits, in feedback, in studio conversation — to articulate why. Why this image. Why this choice. What are you actually trying to say, and is this saying it? That question, repeated enough times, becomes a habit of mind. It is the most transferable thing we teach.
What arts education uniquely offers, in this moment, is exactly the capacity to sit with ambiguity and resist the first answer the machine gives you. The bottleneck in AI-driven creative work is not access to tools — students have that within minutes. It is the quality of judgement being applied to what the tools produce. That judgement is slow to build, impossible to download, and entirely what we are here to develop.
Has the definition of art — or creativity itself — evolved in the age of AI?
Creativity was always more complicated and collaborative than our mythology of the lone genius allowed. Artists have always worked within systems, traditions, constraints, and tools that exceeded their full understanding or control. AI makes that more visible — more uncomfortably visible, for some — but it doesn’t change the fundamental nature of the thing.
What AI does is separate the generative from the intentional in ways that are newly legible. A model can generate; it cannot intend. It can produce; it cannot care about what it produces. The creative act, as I understand it, is the arc from something mattering to you — a question, an unease, a vision — through the making of something, to an audience encountering it and finding that it matters to them too. AI can assist enormously with the middle part. The first and last parts remain stubbornly human.
The Aztec Sun Stone is six hundred years old and we still go to look at it — not because the stone is timeless, but because the question it was asking is still ours: how do we understand our place in the cosmos? That question does not age. The same standard applies now. The medium becomes obsolete. The question, with luck, does not.
How is the role of the artist evolving — and what is the value of the ‘human touch’?
Let me gently push back on the phrase ‘human touch’ — not because the concern behind it isn’t real, but because I think it is the wrong frame. It implies that what artists contribute is a kind of warmth, a finishing flourish applied on top of what the machine produces. That is not what artistic practice is, and it was never what it was.
Every technology that entered the studio brought the same fear. Photography was supposed to kill painting — it didn’t, it freed it. Cinema was supposed to kill theatre — it didn’t, it changed what theatre was for. Digital tools were supposed to make every designer interchangeable — they didn’t, because the tool was never the differentiator.
What artists do — what they have always done — is not touch. It is decision. It is the choice of what question to ask, what to make visible, what cultural inheritance to draw on, what to risk saying in public. Medieval workshop masters didn’t paint every passage of their altarpieces — their apprentices did. Disney’s Mulan required over 600 animators in coordinated industrial workflow; nobody argues it lacks emotional truth. What AI changes is the scale of assistance available — a workflow that once required hundreds can now be held by a few. The vision is no less singular for that.
The question I find more useful than ‘what is the human touch?’ is: what is the human frame? My own practice is built entirely on this: a decade of finding connections between worlds that are not supposed to speak to each other — quantum physics and the spiral logic of Mesoamerican cosmology, the binary structure of Southeast Asian ikat weaving and the mathematics of superposition, the science of entropy and the experience of cultural rupture. These are not illustrations of science, and they are not decoration applied to physics. They are genuine discoveries — moments where two seemingly unrelated systems turn out to be asking the same question in different languages. AI can generate imagery once you have found that connection. It cannot do the finding.
That finding — the willingness to put a real question into the world and care deeply about the answer — is what no model has, and what every artist worth the name does have. It is not a touch. It is a disposition. And it is, I think, more durable than any technology. None of those discoveries came from a prompt. They came from a life.
What is your advice for aspiring artists who are pessimistic about the future?
I want to acknowledge the feeling first. The anxiety is real, and it would be condescending to wave it away. Some traditional entry-level roles are disappearing, and nobody can tell you with certainty what the landscape looks like in ten years.
But here is what I keep coming back to: the artists who are thriving right now — including many of my former students — are thriving not because they have mastered the latest tools, but because they have something genuine to say and the skill to say it across whatever tools are available. That combination is durable. Tool proficiency alone is not.
I don’t think we will see a clean split between ‘AI art’ and ‘purist art’ — that binary is already too simple to describe what’s actually happening. What I think we will see is a growing sophistication in how audiences read AI-generated work. The work that endures will be work that has something to say, made by someone who cared deeply about saying it — regardless of the tools used. That has always been the standard. AI changes the means. It doesn’t change what we ask of the work.
My practical advice: build your practice around questions, not techniques. The techniques you master today will look quaint in ten years — the models, the platforms, the pipelines will all be unrecognisable. But the question you are asking — rooted in your specific background, your cultural inheritance, your particular way of seeing — that is yours in a way no tool can replicate, and no update can replace. The medium becomes obsolete. The question, with luck, does not.
Ina Conradi is Associate Professor at the NTU School of Art, Design and Media, Singapore. Her research spans media architecture, AI-driven art, and decolonial approaches to scientific visualisation.
Gen AI and the Arts: A Conversation with Ina Conradi
Ina Conradi · Associate Professor, NTU School of Art, Design and Media
How should artists view Gen AI?
I’d ask them to resist the word ‘view’ — as if GenAI were a painting on a wall you either like or don’t. It is more useful to think of it as a condition you are already working inside, whether you choose to engage with it or not.
I was at AI on the Lot in Los Angeles last week — the largest annual conference for AI in the entertainment industry, 2,000 professionals from Hollywood studios, tech companies, and media startups at Culver Studios. I presented on a panel called Experience is the Interface: from screens to spaces, AI is transforming how audiences encounter stories. What struck me most was not the technology on display. It was the quality of the questions people were asking. The most interesting practitioners in that room were not asking ‘which tool should I use?’ They were asking ‘what does this make possible that wasn’t possible before?’ That is the attitude I’d encourage.
For working artists: go into the studio with it. Not to use it as a shortcut, but to find out where it resists you, where it surprises you, and what it cannot do that you still can. For aspiring artists: learn the craft first, deeply. The value of AI in any creative process scales with how much genuine artistic thinking you bring to it. Your taste, your conceptual rigour, your embodied sense of what a piece needs — none of that is replaceable, and all of it is what makes the difference.
What about the ethical arguments against GenAI in art?
The ‘stealing’ argument is worth examining carefully, because the reality is more layered than the headline suggests. Training data ethics are a genuine concern — whose work was used to build a model’s weights, under what conditions, with what consent — and I take that seriously. It is why in my studio we pay close attention to how we use these tools — prioritising workflows that involve substantial transformation and iteration, whether through ComfyUI pipelines or other platforms, rather than accepting first outputs. The provenance of the underlying model matters, and so does the creative distance between source and final work.
But the relationship between a model’s training data and any specific output is not straightforward reproduction. Working through multiple iterations — conditioning, transformation, inpainting, spatial recomposition — creates genuine creative distance from any source. By the time a piece has moved through my own workflow, my own prompting logic, my own compositional decisions, the connection to any training image is remote in the way that a painter’s connection to every artwork they have ever looked at is remote. Influence is not theft.
The position I find most defensible — and most interesting — is to train on your own archive. I am currently working in exactly that direction: taking my own earlier works and using AI to extend, enhance, and adapt them for new venues and scales. That is not appropriation. That is an art practice in dialogue with itself.
Which brings me to the second argument — that art requires human creation to evoke emotion — because that is really the more interesting question. Art history is full of works made with machines, with assistants, with processes the maker didn’t fully control — and we don’t strip them of their status as art. Vermeer, the 17th-century Dutch master whose luminous interiors are among the most reproduced paintings in the world, is the most instructive case. Art historians now widely believe he used a camera obscura — projecting the scene onto canvas before painting — to achieve his extraordinary precision. When that theory first gained serious attention it felt scandalous, like an accusation of cheating. Today nobody cares. What matters is the quality and presence of the work, not the purity of the method. The light doesn’t become less luminous because we know how he achieved it. The paintings are still the paintings. The light is still the light.
A generation later, Harold Cohen, a British artist who spent decades from the 1970s onward building AARON — one of the first computer programs to autonomously generate original drawings and paintings — exhibited its output as his own artistic practice for over forty years. At the time, that too was controversial. Now it is art history. What makes work meaningful has never been the purity of the method. It has always been the presence of intention, judgement, and something genuinely at stake.
I have seen AI-generated work move audiences, at scale, in my own practice. Echoes, Whispers and Memories — built substantially with AI-generated imagery — has been experienced by millions: at Ars Electronica in Linz, on the West Lake Media Façade in Hangzhou, and in the 270-degree immersive Sala Immersiva at MEET Digital Culture Center in Milan. The work lands. The emotion is real. Paul Schrader, whose keynote at a conference I attended in Los Angeles reflected on a lifetime of directing films, was circling this point about intention — but it applies equally to every artist working in any medium.
The AI doesn’t care about the story, the meaning, or what is at stake. The artist does. That caring is what makes it resonate.
Do you think a career in the arts is at risk because of GenAI?
I understand the anxiety. But I think it is misdirected — and the more accurate anxiety is different, and more productive.
The roles most at risk are specific, often outsourced tasks at the commercial end of the creative industries: stock illustration, template-based design, certain kinds of commercial concept art. For fine artists the threat is different, and more personal — it is the fear that what made your voice singular, your visual language recognisable, your practice worth building, can now be approximated by anyone with a good prompt. That fear is understandable. But it mistakes style for practice.
A machine can learn to approximate the surface of a visual language. It cannot replicate the decade of looking, failing, questioning, and revising that produced it. It cannot replicate the specific cultural inheritance you bring to a canvas, a screen, or a public space. It cannot replicate what is genuinely at stake for you in the work — the question that keeps you returning to the studio when there is no external reason to. That interiority is not a minor detail. It is the whole thing.
What I have come to believe — and what I saw reflected in conversations in Los Angeles last week — is that AI does not diminish the fine artist’s role. It clarifies it. It strips away the parts of practice that were always somewhat mechanical, and leaves the part that was always irreplaceable: the vision, the question, the particular way of seeing that no training data can fully contain.
My advice: don’t compete with AI on its own terms. Go deeper — into the questions only you can ask, the material only your specific history can unlock, the work only your particular stubbornness can bring into existence. That is where the practice is. That is where it has always been.
You describe AI as having three distinct roles: generative, interpretive, and as an object of study. Can you explain the last two?
As a generative tool — the familiar territory — AI translates a concept or prompt into visual, textual, or sonic output. That is where most of the public conversation stays. The other two roles are where I think the more interesting and more necessary work is happening.
As an interpretive tool, AI is used not to generate images but to interrogate existing material. In my courses, students use large language models to work through scientific papers, datasets, and complex theoretical texts — asking the model to explain, summarise, find contradictions, suggest analogies. We have also worked with scientific imagery — in one recent project, images of invisible plankton from ocean research, using AI to help identify visual patterns and suggest translations into moving image work that the human eye could never assemble from that volume of material alone. The AI here is a thinking partner, not a maker.
As an object of study — the role I think is most undertaught in Singapore right now — we turn the lens around and examine AI itself. This means asking three distinct questions that I think every student working with these tools should be able to answer.
The first is: where does it fail, and what does its training data reveal about cultural bias? Ask any mainstream image generator to produce ‘a scientist’ or ‘a CEO’ and count how many times it defaults to a white man in a lab coat or a suit. Ask it to generate ‘a traditional wedding’ and see whose tradition it assumes. The model is not neutral — it reflects the demographics and aesthetic preferences of the dataset it was trained on, which skews heavily toward Western, English-language, commercially available internet imagery. When students from Singapore, Indonesia, or India use these tools, they are working inside a system that was not built with their visual culture as the default. That is worth knowing explicitly, not just feeling vaguely.
The second question is: how does it represent uncertainty? This connects directly to my own research using art to communicate quantum physics. AI image generators are confidence machines — they always produce a complete, resolved, visually coherent image. They cannot say ‘I don’t know’ or ‘this is unclear.’ But reality — and certainly quantum reality — is full of uncertainty, ambiguity, and incompleteness. That gap — between the model’s false confidence and the genuine uncertainty of the physics — is something my students and I work with directly.
The third question is: what does synthetic imagery do to our sense of documentary truth? During the LA wildfires earlier this year, AI-generated images of burning neighbourhoods circulated on social media alongside real photographs, and some were impossible to distinguish at a glance. When students deploy AI-generated imagery in public contexts — on urban screens, in galleries, in media — they are operating in that landscape. Understanding what synthetic imagery does to trust, to witness, to the evidentiary status of images is not a theoretical concern. It is a practical and ethical one.
Most AI art education currently focuses on the first role — generative — because that is what produces visible, shareable results quickly. But the critical role — examining the system itself — requires slowing down, sitting with discomfort, and asking questions that don’t have easy answers. It is the role I am most committed to developing, because I think it is where artists can make their most distinctive and lasting contribution to how society understands and navigates AI.
What does it mean to think and create within AI-driven cultural systems?
It means understanding that you are not just choosing a tool — you are stepping into a cultural system that has preferences, biases, aesthetics, and assumptions already baked into it. Every generative model has been trained on a corpus that reflects particular choices about what was included, and why. Working thoughtfully within that system means knowing those tendencies well enough to work with them, against them, or around them — rather than accepting whatever the model offers.
In my own work, this plays out at a very physical scale. The panel I recently presented in Los Angeles, Experience is the Interface, argued that when AI moves from screens into physical space — floors that ripple around your feet, façades that wrap a city block — the interface becomes the human body itself. The audience is no longer outside the system. They are part of it. That is not a metaphor. It is a design decision that has to be made by a human being who understands what is at stake in that encounter.
In the classroom, a student who types a prompt and accepts the first output isn’t creating within an AI-driven cultural system — they’re being carried by one. A student who iterates, argues back, applies their own aesthetic judgement at every stage, and understands the cultural implications of what they’re producing — that student is directing the system. That is the distinction I try to teach.
What are the dangers of the blanket term ‘AI art’?
Getting this vocabulary right is not an academic exercise — it has real consequences for working artists. Calling everything ‘AI art’ is a little like calling everything made with a brush ‘brush art’ — it tells you nothing about the work, and it flattens profound differences in skill, intention, and labour into a single unhelpful label.
Consider the difference between these scenarios: an artist who uses AI at the brainstorming stage to surface visual connections, then builds the final work entirely by hand — as with my 2023 animated film Moirai, traditional 3D animation frame by frame, with AI used only in early moodboarding; an artist who uses AI to generate a palette of imagery they then edit, composite, and transform; and an artist who types a prompt and publishes the first output. All three might be called ‘AI art.’ They represent profoundly different relationships to authorship, skill, and creative labour.
The art world has navigated versions of this before. When photography emerged, the question of whether it was ‘real art’ consumed decades of critical energy — energy that could have been spent looking more carefully at what photographers were actually doing and how differently they were doing it. We are repeating that pattern now.
One of the things my students consistently report surprises them most is how much work serious AI-assisted practice actually requires. They come in expecting a shortcut and discover instead a new kind of rigour — hundreds of iterations, each one a decision, each one requiring them to know clearly enough what they want to recognise when they are getting closer to it or further away. The tool does not do the wanting. The artist does. And the wanting, it turns out, is most of the work.
Should there be thresholds for acceptable GenAI usage in art?
I don’t think the question of ‘acceptable thresholds’ is the right frame. It implies a purity spectrum — more AI is worse, less AI is better — and I don’t believe that. A work with 90% AI-generated imagery can be deeply meaningful if the artistic vision and conceptual rigour are present. A work with 1% AI involvement can be cynical and empty. Percentage is not the measure.
What matters to me, both ethically and aesthetically, is: Is there genuine artistic intention driving this work? Are choices being made, or accepted by default? Is the work asking a real question? Has the artist thought about the provenance of the training data?
On the specific question of public deployment — galleries, public commissions, published work — I do think the standards are different and the disclosure obligations are higher. Audiences have a right to know what they are looking at and what kind of labour produced it. A recent and instructive example is Hell Grind — billed as the world’s first fully AI-generated feature film, made by a team of 15 people in fourteen days for $500,000, and screened at an industry event in Cannes during the 2026 festival. Its producers described it as having ‘premiered at Cannes’ — which was technically true of the city, but not of the festival itself. That blurring — deliberate or not — is precisely the problem. The ethics of the private studio and the ethics of public presentation are not the same thing, and conflating them leads to bad conversations and worse policy.
When did you launch your AI art courses at NTU, and how have they evolved?
Both courses — DM2012 Explorations in AI-Generated Art (undergraduate) and AP7055 Art in the Age of the Creative Machine (postgraduate) — launched in 2022 as part of the NTU NISTH AI in Art Studio research project, and have run every semester since. They were among the earliest dedicated AI art courses at any university in Southeast Asia.
Looking back, the conceptual foundation was already Jungian before we named it as such. Carl Jung described a practice he called Active Imagination — consciously engaging with the unconscious while remaining in dialogue rather than seeking control. It describes remarkably well what serious AI studio practice actually is. You enter a system you don’t fully understand, you stay curious rather than directive, and you allow the unexpected to surface as material rather than suppressing it as error.
The earliest iterations of the courses explored exactly that territory — dreams, memory, and the unconscious as subject matter, students learning to work with generative AI the way an analyst works with a dream: not to master it, but to be in genuine conversation with it. Over time that inquiry expanded — from dreams and memory into ocean science, quantum physics, and the mathematics of entanglement. Students from these courses have since gone on to graduate programmes at Carnegie Mellon University, New York University, Parsons, and Pratt.
Who takes these courses, and what is the atmosphere in the classroom?
Both courses consistently fill now — but that was not always the case, and I think the honest version of this answer has to include where we started.
In the first semesters, students would arrive with questions I could not answer. I was not standing at the front of the room as an expert — I was there as an artist who was genuinely, almost overwhelmingly excited by something I did not yet fully understand. I am trained as a weaver and tapestry maker — physically demanding handwork that occupied my twenties and into my mid-thirties, thread by thread, before I moved into media art. That background never left me. And for me, typing a prompt and receiving an image — no matter how low-resolution, how choppy, how undefined — felt like a miracle. The same instinct that makes you hold your breath when the first threads of a pattern begin to emerge on a loom.
Not everyone shared that excitement. Some students walked out in protest — they had serious ethical objections, genuine artistic convictions about what making meant, and they were not wrong to raise them. I gave them the freedom to leave and to question. I think that freedom mattered. The ones who stayed did so by choice, not by compliance — and some of them made work in those early semesters that still brings tears, not for technical achievement but for the stories they found the courage to tell through this strange new medium.
Today the mix includes students from engineering, social sciences, business, and the sciences alongside ADM students, each bringing a different kind of scepticism and a different kind of wonder. The non-arts students often have sharper instincts for AI’s technical limits; the arts students often have sharper instincts for its aesthetic and ethical dimensions. They teach each other.
In 2023 we became the first university from Singapore to participate in the Ars Electronica Campus Exhibition — alongside 56 international universities, before an audience of 88,000 visitors and 1,542 artists, scientists, and designers from 88 countries. The exhibition, Butterfly’s Dreams: The New Aesthetic of AI in Artistic Practice, was presented under the festival theme Who Owns the Truth? — which could not have been more precisely on point for what we were doing in the studio.
How can arts education help students develop the language to question what they are making?
Arts education has always been, at its core, about teaching people to ask harder questions of the work in front of them — including their own. What has changed is the urgency, and the specific vocabulary needed.
Gen AI and the Arts: A Conversation with Ina Conradi
Ina Conradi · Associate Professor, NTU School of Art, Design and Media
Written responses to The Straits Times — Impact of GenAI on the local arts scene Submitted: 1 June 2026
How should artists view Gen AI?
I’d ask them to resist the word ‘view’ — as if GenAI were a painting on a wall you either like or don’t. It is more useful to think of it as a condition you are already working inside, whether you choose to engage with it or not.
I was at AI on the Lot in Los Angeles last week — the largest annual conference for AI in the entertainment industry, 2,000 professionals from Hollywood studios, tech companies, and media startups at Culver Studios. I presented on a panel called Experience is the Interface: from screens to spaces, AI is transforming how audiences encounter stories. What struck me most was not the technology on display. It was the quality of the questions people were asking. The most interesting practitioners in that room were not asking ‘which tool should I use?’ They were asking ‘what does this make possible that wasn’t possible before?’ That is the attitude I’d encourage.
For working artists: go into the studio with it. Not to use it as a shortcut, but to find out where it resists you, where it surprises you, and what it cannot do that you still can. For aspiring artists: learn the craft first, deeply. The value of AI in any creative process scales with how much genuine artistic thinking you bring to it. Your taste, your conceptual rigour, your embodied sense of what a piece needs — none of that is replaceable, and all of it is what makes the difference.
What about the ethical arguments against GenAI in art?
The ‘stealing’ argument is worth examining carefully, because the reality is more layered than the headline suggests. Training data ethics are a genuine concern — whose work was used to build a model’s weights, under what conditions, with what consent — and I take that seriously. It is why in my studio we pay close attention to how we use these tools — prioritising workflows that involve substantial transformation and iteration, whether through ComfyUI pipelines or other platforms, rather than accepting first outputs. The provenance of the underlying model matters, and so does the creative distance between source and final work.
But the relationship between a model’s training data and any specific output is not straightforward reproduction. Working through multiple iterations — conditioning, transformation, inpainting, spatial recomposition — creates genuine creative distance from any source. By the time a piece has moved through my own workflow, my own prompting logic, my own compositional decisions, the connection to any training image is remote in the way that a painter’s connection to every artwork they have ever looked at is remote. Influence is not theft.
The position I find most defensible — and most interesting — is to train on your own archive. I am currently working in exactly that direction: taking my own earlier works and using AI to extend, enhance, and adapt them for new venues and scales. That is not appropriation. That is an art practice in dialogue with itself.
Which brings me to the second argument — that art requires human creation to evoke emotion — because that is really the more interesting question. Art history is full of works made with machines, with assistants, with processes the maker didn’t fully control — and we don’t strip them of their status as art. Vermeer, the 17th-century Dutch master whose luminous interiors are among the most reproduced paintings in the world, is the most instructive case. Art historians now widely believe he used a camera obscura — projecting the scene onto canvas before painting — to achieve his extraordinary precision. When that theory first gained serious attention it felt scandalous, like an accusation of cheating. Today nobody cares. What matters is the quality and presence of the work, not the purity of the method. The light doesn’t become less luminous because we know how he achieved it. The paintings are still the paintings. The light is still the light.
A generation later, Harold Cohen, a British artist who spent decades from the 1970s onward building AARON — one of the first computer programs to autonomously generate original drawings and paintings — exhibited its output as his own artistic practice for over forty years. At the time, that too was controversial. Now it is art history. What makes work meaningful has never been the purity of the method. It has always been the presence of intention, judgement, and something genuinely at stake.
I have seen AI-generated work move audiences, at scale, in my own practice. Echoes, Whispers and Memories — built substantially with AI-generated imagery — has been experienced by millions: at Ars Electronica in Linz, on the West Lake Media Façade in Hangzhou, and in the 270-degree immersive Sala Immersiva at MEET Digital Culture Center in Milan. The work lands. The emotion is real. Paul Schrader, whose keynote at a conference I attended in Los Angeles reflected on a lifetime of directing films, was circling this point about intention — but it applies equally to every artist working in any medium.
The AI doesn’t care about the story, the meaning, or what is at stake. The artist does. That caring is what makes it resonate.
Do you think a career in the arts is at risk because of GenAI?
I understand the anxiety. But I think it is misdirected — and the more accurate anxiety is different, and more productive.
The roles most at risk are specific, often outsourced tasks at the commercial end of the creative industries: stock illustration, template-based design, certain kinds of commercial concept art. For fine artists the threat is different, and more personal — it is the fear that what made your voice singular, your visual language recognisable, your practice worth building, can now be approximated by anyone with a good prompt. That fear is understandable. But it mistakes style for practice.
A machine can learn to approximate the surface of a visual language. It cannot replicate the decade of looking, failing, questioning, and revising that produced it. It cannot replicate the specific cultural inheritance you bring to a canvas, a screen, or a public space. It cannot replicate what is genuinely at stake for you in the work — the question that keeps you returning to the studio when there is no external reason to. That interiority is not a minor detail. It is the whole thing.
What I have come to believe — and what I saw reflected in conversations in Los Angeles last week — is that AI does not diminish the fine artist’s role. It clarifies it. It strips away the parts of practice that were always somewhat mechanical, and leaves the part that was always irreplaceable: the vision, the question, the particular way of seeing that no training data can fully contain.
My advice: don’t compete with AI on its own terms. Go deeper — into the questions only you can ask, the material only your specific history can unlock, the work only your particular stubbornness can bring into existence. That is where the practice is. That is where it has always been.
You describe AI as having three distinct roles: generative, interpretive, and as an object of study. Can you explain the last two?
As a generative tool — the familiar territory — AI translates a concept or prompt into visual, textual, or sonic output. That is where most of the public conversation stays. The other two roles are where I think the more interesting and more necessary work is happening.
As an interpretive tool, AI is used not to generate images but to interrogate existing material. In my courses, students use large language models to work through scientific papers, datasets, and complex theoretical texts — asking the model to explain, summarise, find contradictions, suggest analogies. We have also worked with scientific imagery — in one recent project, images of invisible plankton from ocean research, using AI to help identify visual patterns and suggest translations into moving image work that the human eye could never assemble from that volume of material alone. The AI here is a thinking partner, not a maker.
As an object of study — the role I think is most undertaught in Singapore right now — we turn the lens around and examine AI itself. This means asking three distinct questions that I think every student working with these tools should be able to answer.
The first is: where does it fail, and what does its training data reveal about cultural bias? Ask any mainstream image generator to produce ‘a scientist’ or ‘a CEO’ and count how many times it defaults to a white man in a lab coat or a suit. Ask it to generate ‘a traditional wedding’ and see whose tradition it assumes. The model is not neutral — it reflects the demographics and aesthetic preferences of the dataset it was trained on, which skews heavily toward Western, English-language, commercially available internet imagery. When students from Singapore, Indonesia, or India use these tools, they are working inside a system that was not built with their visual culture as the default. That is worth knowing explicitly, not just feeling vaguely.
The second question is: how does it represent uncertainty? This connects directly to my own research using art to communicate quantum physics. AI image generators are confidence machines — they always produce a complete, resolved, visually coherent image. They cannot say ‘I don’t know’ or ‘this is unclear.’ But reality — and certainly quantum reality — is full of uncertainty, ambiguity, and incompleteness. That gap — between the model’s false confidence and the genuine uncertainty of the physics — is something my students and I work with directly.
The third question is: what does synthetic imagery do to our sense of documentary truth? During the LA wildfires earlier this year, AI-generated images of burning neighbourhoods circulated on social media alongside real photographs, and some were impossible to distinguish at a glance. When students deploy AI-generated imagery in public contexts — on urban screens, in galleries, in media — they are operating in that landscape. Understanding what synthetic imagery does to trust, to witness, to the evidentiary status of images is not a theoretical concern. It is a practical and ethical one.
Most AI art education currently focuses on the first role — generative — because that is what produces visible, shareable results quickly. But the critical role — examining the system itself — requires slowing down, sitting with discomfort, and asking questions that don’t have easy answers. It is the role I am most committed to developing, because I think it is where artists can make their most distinctive and lasting contribution to how society understands and navigates AI.
What does it mean to think and create within AI-driven cultural systems?
It means understanding that you are not just choosing a tool — you are stepping into a cultural system that has preferences, biases, aesthetics, and assumptions already baked into it. Every generative model has been trained on a corpus that reflects particular choices about what was included, and why. Working thoughtfully within that system means knowing those tendencies well enough to work with them, against them, or around them — rather than accepting whatever the model offers.
In my own work, this plays out at a very physical scale. The panel I recently presented in Los Angeles, Experience is the Interface, argued that when AI moves from screens into physical space — floors that ripple around your feet, façades that wrap a city block — the interface becomes the human body itself. The audience is no longer outside the system. They are part of it. That is not a metaphor. It is a design decision that has to be made by a human being who understands what is at stake in that encounter.
In the classroom, a student who types a prompt and accepts the first output isn’t creating within an AI-driven cultural system — they’re being carried by one. A student who iterates, argues back, applies their own aesthetic judgement at every stage, and understands the cultural implications of what they’re producing — that student is directing the system. That is the distinction I try to teach.
What are the dangers of the blanket term ‘AI art’?
Getting this vocabulary right is not an academic exercise — it has real consequences for working artists. Calling everything ‘AI art’ is a little like calling everything made with a brush ‘brush art’ — it tells you nothing about the work, and it flattens profound differences in skill, intention, and labour into a single unhelpful label.
Consider the difference between these scenarios: an artist who uses AI at the brainstorming stage to surface visual connections, then builds the final work entirely by hand — as with my 2023 animated film Moirai, traditional 3D animation frame by frame, with AI used only in early moodboarding; an artist who uses AI to generate a palette of imagery they then edit, composite, and transform; and an artist who types a prompt and publishes the first output. All three might be called ‘AI art.’ They represent profoundly different relationships to authorship, skill, and creative labour.
The art world has navigated versions of this before. When photography emerged, the question of whether it was ‘real art’ consumed decades of critical energy — energy that could have been spent looking more carefully at what photographers were actually doing and how differently they were doing it. We are repeating that pattern now.
One of the things my students consistently report surprises them most is how much work serious AI-assisted practice actually requires. They come in expecting a shortcut and discover instead a new kind of rigour — hundreds of iterations, each one a decision, each one requiring them to know clearly enough what they want to recognise when they are getting closer to it or further away. The tool does not do the wanting. The artist does. And the wanting, it turns out, is most of the work.
Should there be thresholds for acceptable GenAI usage in art?
I don’t think the question of ‘acceptable thresholds’ is the right frame. It implies a purity spectrum — more AI is worse, less AI is better — and I don’t believe that. A work with 90% AI-generated imagery can be deeply meaningful if the artistic vision and conceptual rigour are present. A work with 1% AI involvement can be cynical and empty. Percentage is not the measure.
What matters to me, both ethically and aesthetically, is: Is there genuine artistic intention driving this work? Are choices being made, or accepted by default? Is the work asking a real question? Has the artist thought about the provenance of the training data?
On the specific question of public deployment — galleries, public commissions, published work — I do think the standards are different and the disclosure obligations are higher. Audiences have a right to know what they are looking at and what kind of labour produced it. A recent and instructive example is Hell Grind — billed as the world’s first fully AI-generated feature film, made by a team of 15 people in fourteen days for $500,000, and screened at an industry event in Cannes during the 2026 festival. Its producers described it as having ‘premiered at Cannes’ — which was technically true of the city, but not of the festival itself. That blurring — deliberate or not — is precisely the problem. The ethics of the private studio and the ethics of public presentation are not the same thing, and conflating them leads to bad conversations and worse policy.
When did you launch your AI art courses at NTU, and how have they evolved?
Both courses — DM2012 Explorations in AI-Generated Art (undergraduate) and AP7055 Art in the Age of the Creative Machine (postgraduate) — launched in 2022 as part of the NTU NISTH AI in Art Studio research project, and have run every semester since. They were among the earliest dedicated AI art courses at any university in Southeast Asia.
Looking back, the conceptual foundation was already Jungian before we named it as such. Carl Jung described a practice he called Active Imagination — consciously engaging with the unconscious while remaining in dialogue rather than seeking control. It describes remarkably well what serious AI studio practice actually is. You enter a system you don’t fully understand, you stay curious rather than directive, and you allow the unexpected to surface as material rather than suppressing it as error.
The earliest iterations of the courses explored exactly that territory — dreams, memory, and the unconscious as subject matter, students learning to work with generative AI the way an analyst works with a dream: not to master it, but to be in genuine conversation with it. Over time that inquiry expanded — from dreams and memory into ocean science, quantum physics, and the mathematics of entanglement. Students from these courses have since gone on to graduate programmes at Carnegie Mellon University, New York University, Parsons, and Pratt.
Who takes these courses, and what is the atmosphere in the classroom?
Both courses consistently fill now — but that was not always the case, and I think the honest version of this answer has to include where we started.
In the first semesters, students would arrive with questions I could not answer. I was not standing at the front of the room as an expert — I was there as an artist who was genuinely, almost overwhelmingly excited by something I did not yet fully understand. I am trained as a weaver and tapestry maker — physically demanding handwork that occupied my twenties and into my mid-thirties, thread by thread, before I moved into media art. That background never left me. And for me, typing a prompt and receiving an image — no matter how low-resolution, how choppy, how undefined — felt like a miracle. The same instinct that makes you hold your breath when the first threads of a pattern begin to emerge on a loom.
Not everyone shared that excitement. Some students walked out in protest — they had serious ethical objections, genuine artistic convictions about what making meant, and they were not wrong to raise them. I gave them the freedom to leave and to question. I think that freedom mattered. The ones who stayed did so by choice, not by compliance — and some of them made work in those early semesters that still brings tears, not for technical achievement but for the stories they found the courage to tell through this strange new medium.
Today the mix includes students from engineering, social sciences, business, and the sciences alongside ADM students, each bringing a different kind of scepticism and a different kind of wonder. The non-arts students often have sharper instincts for AI’s technical limits; the arts students often have sharper instincts for its aesthetic and ethical dimensions. They teach each other.
In 2023 we became the first university from Singapore to participate in the Ars Electronica Campus Exhibition — alongside 56 international universities, before an audience of 88,000 visitors and 1,542 artists, scientists, and designers from 88 countries. The exhibition, Butterfly’s Dreams: The New Aesthetic of AI in Artistic Practice, was presented under the festival theme Who Owns the Truth? — which could not have been more precisely on point for what we were doing in the studio.
How can arts education help students develop the language to question what they are making?
Arts education has always been, at its core, about teaching people to ask harder questions of the work in front of them — including their own. What has changed is the urgency, and the specific vocabulary needed.
In practice, we don’t develop this through reading lists or lectures on theory — we develop it through making and looking. We look closely at what AI actually does: where it fails, what its aesthetic defaults reveal, whose visual culture it centres and whose it ignores. And we ask students constantly — in crits, in feedback, in studio conversation — to articulate why. Why this image. Why this choice. What are you actually trying to say, and is this saying it? That question, repeated enough times, becomes a habit of mind. It is the most transferable thing we teach.
What arts education uniquely offers, in this moment, is exactly the capacity to sit with ambiguity and resist the first answer the machine gives you. The bottleneck in AI-driven creative work is not access to tools — students have that within minutes. It is the quality of judgement being applied to what the tools produce. That judgement is slow to build, impossible to download, and entirely what we are here to develop.
Has the definition of art — or creativity itself — evolved in the age of AI?
Creativity was always more complicated and collaborative than our mythology of the lone genius allowed. Artists have always worked within systems, traditions, constraints, and tools that exceeded their full understanding or control. AI makes that more visible — more uncomfortably visible, for some — but it doesn’t change the fundamental nature of the thing.
What AI does is separate the generative from the intentional in ways that are newly legible. A model can generate; it cannot intend. It can produce; it cannot care about what it produces. The creative act, as I understand it, is the arc from something mattering to you — a question, an unease, a vision — through the making of something, to an audience encountering it and finding that it matters to them too. AI can assist enormously with the middle part. The first and last parts remain stubbornly human.
The Aztec Sun Stone is six hundred years old and we still go to look at it — not because the stone is timeless, but because the question it was asking is still ours: how do we understand our place in the cosmos? That question does not age. The same standard applies now. The medium becomes obsolete. The question, with luck, does not.
How is the role of the artist evolving — and what is the value of the ‘human touch’?
Let me gently push back on the phrase ‘human touch’ — not because the concern behind it isn’t real, but because I think it is the wrong frame. It implies that what artists contribute is a kind of warmth, a finishing flourish applied on top of what the machine produces. That is not what artistic practice is, and it was never what it was.
Every technology that entered the studio brought the same fear. Photography was supposed to kill painting — it didn’t, it freed it. Cinema was supposed to kill theatre — it didn’t, it changed what theatre was for. Digital tools were supposed to make every designer interchangeable — they didn’t, because the tool was never the differentiator.
What artists do — what they have always done — is not touch. It is decision. It is the choice of what question to ask, what to make visible, what cultural inheritance to draw on, what to risk saying in public. Medieval workshop masters didn’t paint every passage of their altarpieces — their apprentices did. Disney’s Mulan required over 600 animators in coordinated industrial workflow; nobody argues it lacks emotional truth. What AI changes is the scale of assistance available — a workflow that once required hundreds can now be held by a few. The vision is no less singular for that.
The question I find more useful than ‘what is the human touch?’ is: what is the human frame? My own practice is built entirely on this: a decade of finding connections between worlds that are not supposed to speak to each other — quantum physics and the spiral logic of Mesoamerican cosmology, the binary structure of Southeast Asian ikat weaving and the mathematics of superposition, the science of entropy and the experience of cultural rupture. These are not illustrations of science, and they are not decoration applied to physics. They are genuine discoveries — moments where two seemingly unrelated systems turn out to be asking the same question in different languages. AI can generate imagery once you have found that connection. It cannot do the finding.
That finding — the willingness to put a real question into the world and care deeply about the answer — is what no model has, and what every artist worth the name does have. It is not a touch. It is a disposition. And it is, I think, more durable than any technology. None of those discoveries came from a prompt. They came from a life.
What is your advice for aspiring artists who are pessimistic about the future?
I want to acknowledge the feeling first. The anxiety is real, and it would be condescending to wave it away. Some traditional entry-level roles are disappearing, and nobody can tell you with certainty what the landscape looks like in ten years.
But here is what I keep coming back to: the artists who are thriving right now — including many of my former students — are thriving not because they have mastered the latest tools, but because they have something genuine to say and the skill to say it across whatever tools are available. That combination is durable. Tool proficiency alone is not.
I don’t think we will see a clean split between ‘AI art’ and ‘purist art’ — that binary is already too simple to describe what’s actually happening. What I think we will see is a growing sophistication in how audiences read AI-generated work. The work that endures will be work that has something to say, made by someone who cared deeply about saying it — regardless of the tools used. That has always been the standard. AI changes the means. It doesn’t change what we ask of the work.
My practical advice: build your practice around questions, not techniques. The techniques you master today will look quaint in ten years — the models, the platforms, the pipelines will all be unrecognisable. But the question you are asking — rooted in your specific background, your cultural inheritance, your particular way of seeing — that is yours in a way no tool can replicate, and no update can replace. The medium becomes obsolete. The question, with luck, does not.
Ina Conradi is Associate Professor at the NTU School of Art, Design and Media, Singapore. Her research spans media architecture, AI-driven art, and decolonial approaches to scientific visualisation.
