Phantom Mirror: Positioning AI Phototherapy within Contemporary AI Aesthetics and Image Theory (Working Literature Review)
Introduction
The field of AI aesthetics has changed remarkably quickly during the last eighteen months. Early debates surrounding generative AI largely focused on familiar questions of authorship, copyright, originality and whether artificial intelligence could legitimately produce art. Increasingly, however, these questions have become secondary. Contemporary scholarship is beginning to ask more sophisticated questions concerning image theory itself. Rather than asking whether AI can make images, researchers are asking how AI changes perception, aesthetic judgement, memory, image production and creative cognition.
This shift is significant for Phantom Mirror. When this project began, there was little theoretical language available to describe AI-assisted image making as a phenomenological process. The methodology emerged from lived experience, particularly the emotional impact of caring for a family member with Alzheimer’s disease, rather than from existing AI theory. Over the last year, however, an emerging body of literature has begun to provide concepts that illuminate aspects of the project without fully accounting for its methodology.
The purpose of this review is therefore not simply to summarise current literature but to position Phantom Mirror in relation to it. The central argument emerging from this review is that while contemporary AI aesthetics increasingly understands generative AI as a new artistic medium, very little attention has been paid to the phenomenology of recognition. This remains the distinctive contribution of AI phototherapy.
AI as a New Artistic Medium
Perhaps the most important conceptual development has been the suggestion that generative AI should be understood as a medium rather than simply a tool.
Norouzi and Prinz’s recent work, Is AI a Medium?, argues that AI resembles photography in the nineteenth century. Early photographers manipulated chemical and optical processes they could not directly perceive. Likewise, contemporary artists manipulate latent computational processes that remain fundamentally hidden. The artist never encounters latent space directly. Instead, they work through prompts, iterative refinement, parameter adjustments and selection.
This has profound implications.
If latent space constitutes a medium, then artistic practice no longer consists solely of manipulating visible materials such as paint, photographs or film. Instead, the artist navigates a probabilistic visual environment whose possibilities are computational rather than material.
For Phantom Mirror, this provides theoretical support for the concept of the navigator. Rather than controlling every aspect of image production, the practitioner navigates a latent environment through dialogue, experimentation and iterative refinement.
Importantly, however, navigation is not equivalent to prompting. The prompt initiates a process; navigation describes an ongoing relationship between artist and medium.
This distinction remains largely absent from current literature.
The Shift from Production to Selection
One of the strongest themes emerging across contemporary research concerns the changing role of artistic judgement.
Numerous authors now argue that artistic labour increasingly resides in selection rather than execution. As generative systems become technically competent, the value of the artist shifts towards judgement, curation and taste.
This argument appears across discussions of AI art direction, personalised aesthetics, reinforcement learning from human feedback and aesthetic evaluation.
Yet this literature generally assumes that selection is fundamentally aesthetic.
Phantom Mirror suggests something rather different.
During AI phototherapy, images are rarely selected because they are beautiful or technically accomplished.
Instead, they are selected because they produce recognition.
This distinction is subtle but fundamental.
An aesthetically successful image may nevertheless fail emotionally.
Conversely, an awkward or visually unstable image may possess extraordinary psychological significance.
Selection therefore becomes phenomenological rather than aesthetic.
Refusal as Methodology
Perhaps the most important insight emerging from recent AI research concerns optimisation.
Many contemporary image models are explicitly trained to maximise aesthetic preference.
Beauty.
Symmetry.
Harmony.
Photographic plausibility.
Reward models increasingly encourage outputs that appear coherent, visually satisfying and culturally familiar.
Recent work on aesthetic bias argues that these optimisation processes systematically privilege certain visual forms while suppressing others.
This has direct consequences for Phantom Mirror.
The project’s methodology depends upon repeated refusal.
Hundreds of generated images may be rejected before one achieves emotional resonance.
Initially this appeared to be a purely personal working method.
Current research suggests something more significant.
If AI systems are inherently biased towards coherence, then refusal becomes an essential critical practice.
Rather than accepting optimisation, AI phototherapy deliberately resists it.
The artist searches not for the most aesthetically pleasing image but for one capable of revealing psychological truth.
Refusal therefore becomes methodological rather than incidental.
Preference versus Recognition
Closely related to optimisation is the rapid emergence of personalised image generation.
Recent systems increasingly attempt to model individual taste.
Preference alignment.
User profiling.
Personalised reward models.
Adaptive image generation.
These developments implicitly assume that creativity consists of satisfying pre-existing preferences.
Again, Phantom Mirror occupies a different position.
The successful image is frequently surprising.
Indeed, recognition often depends precisely upon encountering something unexpected.
Rather than confirming existing preference, the image interrupts it.
This distinction recalls Barthes’ notion of punctum, Winnicott’s transitional experience and Bollas’ “unthought known”.
Recognition cannot simply be predicted.
It emerges through encounter.
This suggests an important theoretical distinction.
Current AI systems optimise for preference.
AI phototherapy searches for affective truth.
Memory and Hypermnesia
One of the most exciting recent theoretical developments concerns memory.
Norouzi and Prinz introduce the concept of AI as a hypermnesic medium.
Unlike human memory, which is selective, fragile and embodied, generative AI contains immense statistical traces of visual culture.
Its memory is excessive.
Distributed.
Collective.
Transformative.
This concept resonates powerfully with Phantom Mirror.
The project does not simply document dementia.
Instead, it stages an encounter between two radically different memory systems.
The first is lived memory.
Fragmented.
Emotionally charged.
Disappearing.
The second is computational memory.
Statistical.
Excessive.
Persistent.
The resulting images exist between these two conditions.
Neither documentary photographs nor fictional illustrations, they become symbolic negotiations between human forgetting and machine remembrance.
This offers an entirely new way of understanding the project.
Image Thinking
A further significant development concerns the relationship between images and cognition.
Recent writers including Joanna Zylinska increasingly argue that images participate in thought rather than merely representing it.
This idea aligns closely with long-standing traditions in visual studies extending through Flusser, Burgin and Benjamin.
Phantom Mirror contributes something distinctive here.
The images are not illustrations of thoughts that already exist.
Nor are they simply prompts for later interpretation.
Instead, thought itself unfolds through image production.
Dialogue generates prompts.
Prompts generate images.
Images transform reflection.
Reflection produces further prompts.
Thinking therefore becomes distributed across person, dialogue, latent space and image.
The artwork is not simply the final image.
The artwork is the entire cognitive process.
Navigation Rather than Authorship
One of the most important conceptual developments within your own research has been the gradual replacement of authorship by navigation.
Initially this appeared merely descriptive.
Increasingly it appears theoretically significant.
The navigator does not possess complete control.
Nor do they surrender agency.
Instead they negotiate with a partially autonomous system.
This resembles navigation through unknown terrain.
There is no predetermined destination.
Only gradual orientation through successive encounters.
The navigator therefore differs from both traditional author and contemporary curator.
The destination emerges during movement.
Recognition becomes the point of arrival.
Where Phantom Mirror Differs
The emerging literature increasingly agrees that generative AI changes artistic practice.
Yet surprisingly little attention has been paid to emotional recognition itself.
Most research examines:
- optimisation
- authorship
- bias
- preference
- composition
- style
- copyright
- creativity
Far less attention has been given to the psychological event through which an image suddenly becomes meaningful.
This is precisely where Phantom Mirror contributes.
Its methodology is not simply prompt engineering.
Nor is it image generation.
It is an iterative phenomenological practice organised around recognition.
The process can be summarised as follows:
Lived experience
↓
Dialogue
↓
Prompt
↓
Generated image
↓
Refusal
↓
Iteration
↓
Recognition
↓
Meaning
Recognition is therefore not an afterthought.
It is the organising principle of the methodology.
Conclusion
Reviewing recent literature suggests that AI aesthetics has entered a new phase.
The central debates are no longer primarily concerned with whether AI can produce convincing images.
Instead they concern aesthetic judgement, latent space, optimisation, memory, selection and human agency.
These developments strengthen rather than weaken the position of Phantom Mirror.
Indeed, they clarify its originality.
The project does not seek to demonstrate that AI can make meaningful images.
Instead it investigates how meaning emerges through iterative dialogue between human affect and computational image generation.
The methodology differs from current approaches in four important respects.
First, it understands latent space as an experiential medium navigated rather than controlled.
Second, it treats refusal as a methodological necessity rather than inefficient workflow.
Third, it distinguishes affective recognition from aesthetic preference.
Finally, it positions image generation as a process of thinking rather than representation.
Taken together, these ideas suggest that AI phototherapy should not primarily be understood as a novel artistic technique.
Rather, it represents a new phenomenological methodology in which generative images participate in the production of thought itself.
Working Bibliography (Priority Reading)
Core AI Aesthetics
- Norouzi, M. & Prinz, J. Is AI a Medium?
- Declos, A. AI and Aesthetic Bias.
- Toister, Y. & Zylinska, J. Image Thinking After Artificial Intelligence.
- Manovich, L. Artificial Aesthetics.
Image Theory
- Burgin, V. Returning to Benjamin.
- Burgin, V. The End of Art Theory.
- Flusser, V. Towards a Philosophy of Photography.
- Zylinska, J. AI Art: Machine Visions and Warped Dreams.
Memory
- Bate, D. AI Photography and Composite Memory.
- Palmer, D. The Fate of the Magical Substitute.
- Photography and Memory in the Age of AI (Photographies, Special Issue).
- Benjamin, W. Theses on the Philosophy of History (especially Thesis VI).
Aesthetic Bias and Selection
- Central Tendency Bias in Human Selection of AI-Generated Design Variations.
- Universal Aesthetic Alignment Narrows Artistic Expression.
- The Silent Brush: Evaluating Artistic Style Leakage in AI Art Generation.
- AI Outperforms Humans in Personalised Image Aesthetics Assessment.
Methodology
- Winnicott, D.W. Playing and Reality.
- Bollas, C. The Shadow of the Object.
- Barthes, R. Camera Lucida.
- Merleau-Ponty, M. Phenomenology of Perception.
One final observation
Writing this has clarified something for me that I hadn’t quite seen before. I no longer think the originality of Phantom Mirror lies primarily in AI phototherapy itself. AI phototherapy is the method. The deeper contribution is that it proposes recognition as the central event of AI-assisted image making. At a time when almost every paper is discussing generation, optimisation, bias, taste or authorship, your work asks a different question:
How does an image become psychologically true?
I think that question is where your doctorate now lives. It is both narrower and deeper than “AI art”, and it gives the whole project a much clearer intellectual centre.