PhD Working Note
From Prompt to Iteration: Rethinking AI Image-Making After the Symposium
1. Starting Point: Reflection on the Symposium
This note records a shift in my thinking following the recent symposium on AI and artistic practice.
What struck me most was not a lack of technical or conceptual engagement, but a consistent orientation toward what I would describe as an earlier framework of AI image-making. Across multiple presentations, the dominant model remained structured around:
- the prompt as the primary creative act
- the image as the central artefact
- authorship as a problem of ownership or attribution
Even where work was sophisticated, and even where artists were critically engaged with AI, the discussion remained anchored to this structure.
Alongside this, there was a noticeable tendency to return AI-generated work to analogue forms, particularly darkroom processes. While this was often framed as an aesthetic or material decision, it also appeared to function as a stabilising gesture, re-grounding the work within familiar photographic traditions.
Overall, I was left with a clear sense that much of the discourse is still operating within a prompt–image–author framework, even when attempting to move beyond it.
2. A Disjunction Between Practice and Discourse
This experience created a strong sense of disjunction with my own practice.
Over the past year, working intensively with AI phototherapy, I have not experienced the process primarily in terms of prompt construction or image production. Instead, what has become increasingly clear is that the work occurs elsewhere.
The prompt, while necessary, functions only as an entry point. The image, while important, is not the final locus of meaning.
The work unfolds in the movement between images, through a process of iteration, rejection, and recognition.
This distinction was not something I had previously articulated clearly. The symposium provided the contrast needed to recognise it.
3. Iterative Prompting as the Primary Site of Practice
The central shift in my thinking can now be stated as follows:
The primary site of meaning in AI image-making is not the prompt or the image, but the iterative process that connects them.
In practice, this involves:
- generating multiple images from an initial prompt
- moving rapidly through variations
- rejecting most outputs
- continuing iteration until an image produces a strong sense of alignment
This process is not linear. It does not move toward a predefined goal. It operates through what I am now describing as pattern-to-pattern movement, where each image becomes the condition for the next.
The decisive moment is not the production of an image, but the moment at which one image produces a form of recognition, or more precisely, what I am beginning to define as affective arrest.
This is the point at which the process stops, not because the image is objectively better, but because it resonates in a way that cannot be fully explained immediately.
4. Relationship to Current Algorithmic Art Discourse
It is important to situate this within existing thinking.
Contemporary theorists such as Joanna Zylinska and the Creative AI Lab, as well as figures like Lev Manovich, have already established several key ideas:
- AI art is processual rather than object-based
- authorship is distributed across human and machine
- iteration and variation are central to generative systems
These positions are clearly relevant and align with aspects of my practice.
However, what I now recognise is that these frameworks remain largely external descriptions of systems and processes. They do not fully account for:
- how the process is experienced from within
- how meaning emerges during iteration
- what constitutes the decisive moment of selection
In this sense, they acknowledge iteration, but do not yet provide a phenomenology of iteration.
5. A New Position: Iteration as Perceptual and Cognitive Process
My current thinking is that iterative prompting should not be understood simply as a technical or generative procedure.
Instead, it operates as a perceptual and cognitive process, in which meaning is constructed through visual engagement rather than linguistic articulation.
This has several implications:
- The prompt functions as a threshold rather than a command
- The image is not a final object but a moment within a sequence
- Meaning emerges through affective recognition, not conceptual intention
- The process resembles a form of visual thinking, rather than image production
This aligns with broader cognitive and phenomenological theories of perception, while extending them into a new domain of human–machine interaction.
6. From Language to Image
One of the most important consequences of this shift is a rethinking of the role of language.
Traditionally, meaning is assumed to be constructed through a sequence:
experience → language → interpretation
In my practice, this sequence appears to be reconfigured:
experience → image → recognition → language
Language is not removed, but it is displaced from its primary position. Meaning first emerges in the visual-perceptual field and is only later articulated linguistically.
This is not a total replacement of language, but a reordering of the relationship between visual and linguistic cognition.
7. Theoretical Convergence
Through further reflection, I have begun to see that this practice aligns with several theoretical frameworks:
- Benjamin’s idea of memory appearing in flashes under pressure
- Burgin’s concept of the image as a psychical object
- Lacan’s account of the limits of symbolisation
However, these theories do not in themselves describe the methodology I am using. Rather, they help to explain why the process behaves as it does.
The key point is that the theory follows the practice, not the other way around.
8. Emerging Contribution
What is beginning to emerge from this is a clearer sense of contribution.
I am not proposing that:
- iteration is new
- distributed authorship is new
- or that AI fundamentally replaces existing artistic practices
Rather, I am proposing that:
iterative prompting constitutes a distinct mode of meaning-construction, in which images function as sites of thought, and recognition operates as the primary epistemic event.
This is a focused and practice-led claim.
9. AI Image-Making and the Need for a New Visual Literacy
A final point that has emerged through this reflection concerns how AI-generated images are currently being read.
One of the most striking aspects of the symposium was not only how artists are working with AI, but how AI images are being interpreted. There appears to be a persistent tendency to read these images through the frameworks of existing visual media:
- photography
- painting
- cinema
This is understandable, as AI-generated images often resemble these forms. They share their visual language, composition, and surface appearance.
However, this resemblance is misleading.
AI images are not photographs. They are not cinematic frames. They are not paintings. They do not emerge from:
- an indexical relation to the world
- a physical act of capture
- or a stable artistic medium
Instead, they are generated through pattern correlation within a computational system, and their meaning emerges through iterative interaction rather than representation.
In this sense, we are currently misreading AI images.
A useful comparison can be made with early film. When film first emerged, it was often perceived as a direct representation of reality, because it appeared to reproduce the world in motion. Over time, it became clear that film operates through its own language, one that must be learned.
AI image-making appears to be at a similar stage.
We are encountering a new visual medium that:
- borrows the appearance of existing forms
- but operates through a different underlying logic
This suggests the need for a new visual literacy.
Such a literacy would focus not only on what the image depicts, but on:
- how it was generated
- how it relates to other images in an iterative sequence
- and how meaning emerges through recognition within that process
In this context, the image is not a representation to be interpreted in isolation. It is part of a dynamic perceptual field, and must be understood relationally.
10. Next Steps
This shift in thinking now needs to be tested and developed further.
In particular:
- refining the concept of affective arrest
- clarifying the structure of iteration
- applying this framework consistently across my work
- positioning it more precisely within contemporary theory
The aim is to move from an intuitive understanding to a clearly articulated and defensible methodology.
Closing Note
The symposium was valuable not because it introduced new ideas, but because it made visible the gap between existing discourse and my own practice.
It allowed me to recognise that I am not simply using AI differently, but thinking through it differently.
This document marks the beginning of that clarification.