THE BEHAVIOURAL SURFACE
Rethinking Artistic Agency in the Encounter with AI
We have become accustomed to asking what artificial intelligence can produce.
I am increasingly interested in a different question:
What kind of surface does AI become when an artist does not use it simply to obtain an image?
I call this a behavioural surface.
The term emerges from my own practice, which has long been concerned with surfaces, material transformation, erosion, time and the gradual emergence of meaning. I do not approach a surface primarily as a support on which an image is placed. I approach it as a condition: something upon which processes occur, something through which traces become visible.
In this sense, a wall receives time. A metal surface receives oxidation. A stone receives erosion.
AI is different.
It does not simply receive the artist's action.
It responds.
This response changes the nature of the surface and, with it, the location of artistic agency.
1. From image production to encounter
The 2024 AICA International Congress, Becoming Machine, Resisting the Artificial. Art in the Present Tense, identified a central problem of contemporary artistic practice: the partial delegation of the creative act to the machine and the resulting tension between individual agency and computer control. The congress explicitly asked to what extent humans and machines share the creative process, and how AI might alter established ideas of authorship, originality, value and intellectual property.
I propose that this question can be approached from another direction.
Instead of asking how much of the image belongs to the machine and how much belongs to the artist, we might ask:
Where does the artistic decision take place?
This question shifts attention away from the visible image and toward the sequence of decisions through which the image becomes meaningful.
In my practice, I do not use AI to arrive at a predetermined image. I enter an iterative encounter in which the system produces possibilities that I then examine, reject, redirect, alter, preserve or erase.
The process is not:
prompt → image → artwork.
It is closer to:
condition → emergence → encounter → hesitation → selection → refusal → intervention → return → recognition.
The distinction is fundamental.
The generated image is not necessarily the artwork. It is an event within the process.
2. The behavioural surface
A conventional canvas is a surface upon which the artist acts.
A photographic sensor is a surface upon which light leaves a trace.
A wall is a surface upon which time, weather, use and neglect become visible.
A behavioural surface is different because it responds to the behaviour of the artist.
It does not merely register an action. It produces a response that becomes the next condition for another action.
The artist says, in effect:
Not this.
The system responds.
The artist says:
Less.
It responds again.
A form becomes too architectural; it is reduced.
A surface becomes too smooth; it is disturbed.
A colour begins to dominate; it is withdrawn.
A detail appears unexpectedly; it is preserved.
At a certain point, the artist no longer knows exactly what the next response will be. Yet the artist remains responsible for deciding whether that response enters the work.
This is why I do not understand AI as an autonomous author.
Nor do I understand it as a neutral instrument.
It is a behavioural surface: an active field of response within which artistic judgement operates.
3. Agency after control
This distinction matters because much of the current discussion of AI art remains trapped in a binary.
Either the machine is said to make the work, in which case the artist's role appears diminished, or the artist is said to remain completely in control, in which case AI is reduced to another instrument.
Both positions seem insufficient.
The first overestimates machine autonomy.
The second underestimates the significance of encountering a system whose responses cannot be fully predetermined.
Artistic agency may therefore need to be understood not as total control, but as the capacity to construct, navigate and judge conditions of emergence.
The artist's role does not disappear when the image is no longer directly produced by the artist's hand.
It changes location.
Agency moves into selection, sequencing, refusal, intervention, timing and judgement.
The question is no longer simply:
Who produced the image?
It becomes:
Who established the conditions under which this image became worth keeping?
This does not solve the question of authorship. It makes the question more precise.
4. The artist as condition-builder
My own practice begins from a simple proposition:
I do not produce images; I construct conditions in which meaning, already present, can become visible over time.
Working with AI has not caused me to abandon this proposition.
It has extended it.
If the image is not the primary object of the practice, then AI does not need to be understood primarily as an image-making machine. It can instead become another condition within which an image, a form, a trace or an ambiguity may emerge.
This changes what it means to be an artist.
The artist becomes less a producer of predetermined appearances and more a condition-builder.
This is not a surrender of authorship.
It is a redistribution of where authorship is exercised.
The artist determines what kind of encounter is permitted, what kind of response is accepted, what is rejected, what is repeated, what is interrupted and when the process stops.
The final image may contain elements that the artist could not have predicted.
That unpredictability does not necessarily weaken authorship.
It may be precisely what makes the encounter artistically productive.
5. Against the fantasy of seamless generation
There is another reason to insist on this distinction.
The dominant rhetoric surrounding generative AI tends to celebrate immediacy:
more images, faster images, infinite variation, instant transformation.
But artistic practice does not necessarily become more meaningful as the production of images becomes faster.
My own process moves in the opposite direction.
I slow the image down.
I look again.
I remove what is too explicit.
I weaken what is too spectacular.
I allow architectural forms to become uncertain.
I reduce colour.
I preserve small material traces.
I return to the surface.
In this sense, working with AI can become an exercise in resisting the apparent inevitability of its speed.
The technological system may generate endlessly.
The artist does not have to.
The machine can produce abundance.
The artist can introduce scarcity.
The machine can multiply possibilities.
The artist can refuse them.
The machine can accelerate.
The artist can hesitate.
This hesitation is not inefficiency.
It is a form of judgement.
6. Technology, history and the right to hesitate
This is where my proposition unexpectedly encounters Paul O'Kane's argument in his 2024 AICA paper, Technologies of Romance – Moving and Still Images, Death and Looking Back.
O'Kane proposes “affirming anachronism” as a critical foil to what he describes as the emphatic technologisation of art and life. Rather than simply accepting technological novelty as progress, he proposes history as a progressive tool: a way of gaining distance from the hypnotic present and reconsidering technology within the longer history of artistic practices.
This is significant for my own practice.
I do not want to resist AI by refusing it.
Nor do I want to embrace it by celebrating its novelty.
I want to work through it slowly enough to observe what kind of artistic relationship it creates.
O'Kane's proposition that the “outmoded” might become a site of resistance is particularly productive here. If technological culture increasingly rewards immediacy, acceleration and infinite production, then hesitation, repetition, reduction and slowness may themselves become critical artistic positions.
The behavioural surface therefore does not represent an attempt to make AI more human.
It is an attempt to create enough distance from the machine to encounter what its behaviour makes possible.
7. From canvas to behavioural surface
The history of art can be understood, in part, as a history of changing relationships between artists and surfaces.
The canvas became a site for gesture.
Photography transformed the surface of representation into a site of light and exposure.
Digital imaging transformed it again into a site of information and manipulation.
Generative AI introduces another possibility.
The surface no longer simply receives the artist's action.
It acts back.
This is why I use the term behavioural surface.
It is not another name for a digital canvas.
It is a different conception of the canvas itself.
The canvas receives the artist's gesture.
The wall receives time.
The behavioural surface responds.
Once this distinction is made, AI becomes less interesting as a machine for producing images and more interesting as a new kind of artistic encounter.
8. What remains human?
The question “Will AI replace the artist?” may therefore be the wrong question.
A more productive question is:
What becomes specifically human when the production of visible images is no longer exclusively human?
My answer is not the hand.
It is not craft in the narrow sense.
It is not even intention alone.
It is judgement within uncertainty.
The artist encounters something not entirely predetermined and decides whether it matters.
The artist recognizes a possibility that did not previously exist.
The artist refuses what is seductive but empty.
The artist preserves what is barely visible.
The artist decides when enough has happened.
This is why the behavioural surface does not diminish the artist.
It makes the artist's responsibility more difficult to locate, and therefore potentially more significant.
9. Beyond “AI art”
I am therefore reluctant to describe my practice simply as “AI art”.
That designation can make the technology appear to be the subject.
It is not.
The subject remains the encounter between material, time, perception and meaning.
AI is one of the surfaces through which that encounter can now occur.
The important question is not whether the machine has become an artist.
It is whether the artist can develop a practice capable of encountering a machine without either worshipping it or reducing it to a tool.
Perhaps the task is neither to become machine nor simply to resist the artificial.
Perhaps it is to develop forms of attention capable of existing between the two.
The behavioural surface is one such attempt.
It is a surface of response, but also of refusal.
A surface of possibility, but also of limitation.
A surface where the artist does not entirely control what appears, yet remains responsible for deciding what becomes visible.
And perhaps this is where artistic agency begins again:
not with the certainty of making,
but with the responsibility of recognising.
Closing proposition
I do not use AI to obtain the image I already have in mind.
I enter a behavioural surface and allow an image to emerge through the encounter.
The work begins where control becomes insufficient, and judgement becomes necessary.
Paul O’Kane — “Technologies of Romance” (UAL Research Online)
Last updated: October 2026