A Second Direction of Artistic Becoming
The Reversal of the Phantom Mirror
A Second Direction of Artistic Becoming
Generative AI, Predictive Perception and the Aesthetics of the Medium
2 November 2026
1. A discovery through practice: Worlds in Motion
An unexpected development in my research has emerged through the preparation of Worlds in Motion for the AVEVA art commission.
Having completed the selection of the final artworks with the curator, Anne G, I began examining the images more closely in preparation for printing at approximately two metres square. The images had been generated in Midjourney and subsequently upscaled using Topaz Gigapixel’s Wonder 3 model.
At 500% magnification in Photoshop, I discovered something extraordinary.
What appears at ordinary viewing distance to be a coherent image, a ship, a tree, a landscape or a world in motion, reveals an entirely different visual environment under magnification. The image becomes populated by unfamiliar shapes, strange structures, extraordinary textures and forms that seem almost recognisable.
Within a ship, I find what appear to be armchairs, tables or architectural fragments. Elsewhere, I suddenly recognise a bird, a face or something resembling an octopus. Yet these forms are unstable. They appear, disappear and change according to how I look.
The experience is not simply one of discovering hidden details. It feels as though I have entered another world within the image.
What is equally remarkable is the beauty of these magnified structures. They contain extraordinary visual relationships, stylistic associations and compositions that I never consciously designed. The images appear almost blatantly collage-like, yet their apparent components do not necessarily correspond to identifiable objects or source photographs.
I began to feel that the magnification was revealing the medium itself.
This discovery has emerged specifically because of the AVEVA commission. Preparing images for large-scale exhibition has required a different kind of attention from that involved in generating and selecting them. The practical demands of printing have produced an unexpected theoretical discovery.
2. From how the image is made to what kind of image has been made
Until now, much of my investigation into Phantom Mirror has concerned the process of making images through generative AI.
My methodology involves dialogue, navigation, generation, refusal, recognition and selection. These processes establish my artistic agency through an encounter with a generative system that I cannot fully control.
But the experience of magnification introduces another question.
What kind of image has actually been made?
In traditional photography, magnification reveals details recorded from an existing scene. In generative AI imagery, there is no equivalent photographed scene. The visual structure has been computationally synthesised, and generative upscaling can introduce further details that were not explicitly present in the original image.
The resulting forms are not necessarily representations of discrete objects. They emerge from computational relationships shaped by training data, learned statistical patterns and generative processes.
A ship may contain shapes resembling furniture without the system having deliberately placed furniture inside a ship. What I perceive as an armchair may be a visual configuration that happens to support that interpretation.
The image is therefore not simply illustrating an idea. It is producing conditions in which new interpretations become possible.
3. The reversal of the Phantom Mirror
When I examine these images under magnification, I find myself searching through unfamiliar visual structures, recognising possible objects, questioning those recognitions and discovering new ones.
I began to feel, almost extraordinarily, as though I were becoming an AI myself.
This is not a claim that human perception operates in the same way as a generative image model. The computational mechanisms are fundamentally different.
Rather, I have become unusually conscious of my own perceptual inference while looking into the results of a different generative process.
The machine has produced an image through learned computational relationships. It has not necessarily recognised the octopus, the armchair or the face that I subsequently perceive. It has no demonstrated lived, embodied experience of those objects comparable to my own.
Yet I encounter its visual constructions through precisely that lived experience.
My perception searches for familiar relationships within unfamiliar forms. I recognise something, reconsider it, reject it and recognise something else.
The image has not changed, but my interpretation has.
This suggests a reversal of the Phantom Mirror.
In the original methodology, human imagination and intention encounter a generative system, producing images that acquire significance through recognition.
Now I enter an already generated image, and my own perceptual system begins producing possibilities from within its ambiguity.
The machine generates the visual field. Human perception generates provisional meanings from that field.
The mirror is therefore not an equivalence between machine and human cognition. It is an encounter between two different processes through which visual coherence is constructed.
4. Predictive perception and the ambiguity of recognition
Predictive processing provides a useful framework for considering this experience.
Human perception is not simply the passive reception of visual information. It involves the continuous interpretation of sensory input through prior experience, expectation and contextual understanding.
When confronted with ambiguous visual structures, the perceptual system attempts to organise them into recognisable forms.
This helps explain why I might suddenly see an octopus within a structure that previously appeared entirely abstract, or why another person might recognise something completely different.
The phenomenon is related to pareidolia, but my interest extends beyond identifying familiar objects in ambiguous patterns.
I am interested in the experience of perception becoming visible to itself.
As I move through these magnified images, I become aware of recognition as an unstable, ongoing activity rather than a completed event.
This raises further questions. Might mood, memory, emotional associations or unconscious expectations influence which forms become perceptually available? Could repeated encounters with the same image reveal different possibilities according to the viewer’s psychological state?
These remain questions rather than established findings. Nevertheless, they open a possible relationship with psychoanalysis and the role of unconscious association in visual experience.
5. The aesthetics of a new medium
This discovery also changes how I understand generative AI as an artistic medium.
Much contemporary debate approaches AI imagery through imitation, plagiarism, authorship and replacement. These questions remain important, but they do not exhaust the medium’s artistic possibilities.
What I am discovering through magnification concerns the distinctive visual properties of computational image synthesis.
Their collage-like appearance is particularly significant.
Traditional collage involves the deliberate assembly of fragments, often taken from identifiable sources. Generative AI can produce visual structures that resemble collage without necessarily assembling discrete fragments in that manner.
The image may appear to contain incompatible objects, styles and perspectives, yet these relationships emerge through computational synthesis rather than conventional cutting and pasting.
What might otherwise be dismissed as a computational irregularity becomes aesthetically compelling.
This connects with Norouzi and Prinz’s discussion of AI as a medium and the possibility of investigating its distinctive aesthetic capacities. It also raises questions associated with Alexandre Declos’s work on aesthetic bias: to what extent is the beauty I discover an unfamiliar aesthetic possibility, and to what extent does it reflect the learned preferences and conventions embedded within generative systems?
These are not mutually exclusive explanations.
The computational character of the image becomes available for artistic investigation.
6. A second direction of artistic becoming
Previously, I understood artistic becoming principally through the generative process itself. The artwork emerged through dialogue, variation, refusal, selection and recognition.
The image was not the straightforward execution of a predetermined idea. Its meaning developed through an encounter between my artistic intentions and the possibilities offered by the system.
But magnification has revealed another stage.
The artwork does not finish becoming when I select it.
Through upscaling, material preparation, sustained looking and encounters with viewers, it continues to disclose possibilities that were not anticipated during its making.
This suggests a second direction of artistic navigation.
First, I navigate the generative system to produce the image.
Then I navigate the generated image itself.
The first process concerns the emergence of the artwork through human-machine interaction. The second concerns the emergence of new perceptual and aesthetic possibilities through the encounter between the completed image and human perception.
Both involve discovery, uncertainty and recognition, but they operate differently.
This is what I would describe as a second direction of artistic becoming.
7. World Beneath Worlds: The Return of the Original Proposition
There is a further, unexpected significance to this discovery. My original proposal for the AVEVA commission was titled World Beneath Worlds. At the time, I understood this as an imaginative exploration of worlds concealed beneath familiar appearances.
Yet the process of preparing the finished images for exhibition has revealed another interpretation of that original proposition. Through magnification, I have discovered worlds within the images themselves, visual environments whose structures, aesthetic qualities and possibilities for recognition differ from those of the images viewed as a whole.
The second world is not simply a miniature version of the first. It possesses its own strange forms, stylistic relationships and perceptual ambiguities. What appears to be a coherent scene at ordinary scale becomes an entirely different visual environment under magnification.
The original artistic idea has therefore acquired a meaning that I neither intended nor anticipated. In a particularly striking instance of artistic becoming, the medium has returned my own proposition to me in an unexpected form.
8. Image thinking, Flusser and the recursive relationship
Joanna Zylinska and Yanai Toister’s Image Thinking after Artificial Intelligence offers a broader theoretical context for this development.
Their argument is not simply that images encourage interpretation. It concerns the changing architecture of cognition through the recursive relationships between images, language, computation and human imagination.
Their discussion of Vilém Flusser and Mira Schendel is particularly relevant because it challenges the assumption that images merely represent ideas already formed. Images can participate in the assembly and development of thought itself.
Their account of the textimagetext relationship describes a recursive process in which language produces images, images transform subsequent thought and language, and these transformations feed back into further image production.
Their discussion of cognitive hacking also raises the possibility that sustained interaction with generative systems changes our imaginative expectations and habits.
My discovery at 500% magnification should not be equated directly with this larger theory. The immediate phenomenon concerns computationally generated visual ambiguity and human perceptual inference.
However, when these perceptual discoveries begin to change how I imagine, what I recognise and what I subsequently generate, they may become part of the recursive cognitive relationship that Zylinska and Toister investigate.
The relationship between the artist, the image and the generative system may itself be changing through the encounter.
9. The Phantom Mirror in two directions
The discovery suggests that Phantom Mirror may operate through two interconnected directions.
In the first, human imagination encounters computational generation. The artist navigates possibilities offered by the system, and meaning emerges through refusal and recognition.
In the second, human perception encounters the computationally constructed image. Ambiguous visual forms invite further recognition, association and interpretation, producing meanings that were not necessarily anticipated by either the artist or the system.
The image becomes a meeting point between computational synthesis and embodied human inference.
There is an important distinction to preserve. Pareidolia, ambiguity and shifting interpretation are not unique to AI. Abstract painting, surrealism, collage and photographic grain have long invited similar perceptual experiences.
What may be distinctive here is the combination of generative synthesis, computational upscaling and the production of visual structures whose apparent detail can exceed the recognisable organisation of the image as a whole.
It will be important to investigate how much of this effect originates in Midjourney, how much is introduced through Topaz, and how their interaction shapes the final image.
Nevertheless, the central discovery remains.
Generative AI is not simply a medium through which artists produce images. Its computationally constructed visual properties can themselves become sites of artistic and perceptual discovery. The image does not merely communicate a completed thought; it can provoke thinking that neither preceded nor was fully anticipated by its making.
Through the AVEVA commission, I have encountered a new dimension of my own practice. The practical preparation of images for exhibition has revealed not only unfamiliar aesthetic qualities, but also a new awareness of my own perceptual activity.
The Phantom Mirror has, in a sense, turned around.
I began by navigating a machine to discover images.
I now find myself navigating those images to discover how I see.
And through that encounter, the artwork continues to become.
Theoretical references for further development
- Norouzi, M. and Prinz, J. (2026). Is AI a Medium?
- Declos, A. (2026). AI and Aesthetic Bias.
- Zylinska, J. and Toister, Y. (2026). Image Thinking after Artificial Intelligence.
- Flusser, V. Writings on technical images and Mira Schendel, as discussed by Zylinska and Toister.
- Merleau-Ponty, M. Phenomenology of Perception and The Visible and the Invisible.
- Friston, K. and Parr, T. Research on predictive processing and active inference.
- More Than We Can Make: Generative AI, Artistic Forms, and Artistic Becoming (2026).
Working research note. The proposed relationships between computational synthesis, predictive perception, aesthetic experience and artistic becoming are interpretative propositions arising from practice, rather than established empirical conclusions.