Yes. In fact, I think this would become one of the most useful documents for your PhD. One thing I’ve noticed is that the field is inventing new vocabulary incredibly quickly. If we don’t pin these terms down, it’s easy to lose sight of what they actually mean.
I’ve deliberately written these definitions from the perspective of Phantom Mirror, not simply copied from the papers.
Glossary of Key Terms in Contemporary AI Aesthetics and Image Theory (Working Draft)
Aesthetic Bias
The tendency for AI image models to privilege certain visual qualities, such as symmetry, beauty, harmony, saturation, realism and coherence. Rather than being neutral, models encode assumptions about what counts as a “good” image through training data, reinforcement learning, platform defaults and reward models.
Relevance to Phantom Mirror
Aesthetic bias explains why AI phototherapy depends upon repeated refusal. The method resists the model’s tendency to resolve ambiguity too quickly.
Aesthetic Alignment
The process by which AI models are trained to produce outputs that conform to human preferences or reward signals.
Often achieved through reinforcement learning from human feedback (RLHF).
Relevance
Raises the question of whether optimisation towards preference may suppress emotional complexity.
Affective Recognition
(Not a published term.)
The moment at which an image produces immediate emotional recognition.
Recognition is not primarily intellectual or aesthetic.
It is experiential.
This is becoming one of the central concepts of Phantom Mirror.
Affective Truth
(Your emerging term.)
An image possesses affective truth when it accurately captures an emotional condition, regardless of whether it is factually or photographically accurate.
This differs from preference or realism.
Central Tendency Bias
The psychological tendency for people selecting from multiple AI-generated images to choose outputs that lie closest to the statistical centre of the group rather than the most unusual or emotionally powerful image.
Relevance
Suggests that image grids themselves influence judgement.
Coherence Bias
The tendency of generative systems to produce images that appear internally consistent, harmonious and visually resolved.
Closely related to aesthetic bias.
Composite Memory
David Bate’s concept describing AI images as statistical combinations of multiple visual memories rather than records of individual events.
Memory becomes aggregated rather than indexical.
Hypermnesia
One of my favourite concepts from Norouzi and Prinz.
Hypermnesia literally means excessive memory.
The authors argue that generative AI contains enormous statistical traces of visual culture.
Rather than forgetting, AI remembers too much.
Relevance
Phantom Mirror stages an encounter between:
- disappearing human memory
and
- excessive computational memory.
I think this is an extraordinarily productive concept.
Image Thinking
Developed by Joanna Zylinska and Yanai Toister.
The idea that images do not merely illustrate thought.
They actively participate in thinking itself.
Images become cognitive agents.
Relevance
Probably the single most important theoretical concept currently available for your PhD.
Latent Space
The mathematical space inside generative models where relationships between concepts and images are encoded.
Artists never directly perceive latent space.
Instead they navigate it indirectly through prompts and iterative refinement.
Mediated Latent Medium
Norouzi and Prinz’s description of generative AI.
AI is neither simply software nor merely a tool.
It is a new artistic medium whose material consists of hidden computational processes.
Artists manipulate those processes indirectly.
Relevance
Supports your concept of the navigator.
Navigator
(Your concept.)
A practitioner who navigates latent space through dialogue, iteration, refusal and recognition.
Unlike an author, the navigator does not possess complete control.
Unlike a curator, the navigator is actively involved in generating new possibilities.
Optimisation
The process whereby AI systems improve outputs according to predefined objectives.
In image generation this frequently means:
- beauty
- realism
- preference
- coherence
- prompt fidelity
Relevance
AI phototherapy frequently works against optimisation.
Phenomenological Inquiry
An investigation into lived experience.
Rather than asking how AI works computationally, phenomenological inquiry asks:
“What is it like to think with AI-generated images?”
Preference Alignment
Training AI systems to match human preferences.
Current AI research increasingly assumes successful images are those users prefer.
Relevance
Your work distinguishes preference from emotional recognition.
Psychological Truth
Another emerging concept within Phantom Mirror.
An image becomes psychologically true when it captures an emotional reality even if it depicts events that never literally occurred.
Recognition
Probably the central concept of your thesis.
Recognition is the event in which an image suddenly becomes emotionally meaningful.
It is neither generated nor entirely interpreted.
It emerges during the encounter between viewer and image.
Refusal
(Your methodological contribution.)
The deliberate rejection of images that fail to produce emotional recognition.
Rather than inefficiency, refusal becomes the mechanism through which meaning develops.
Statistical Memory
Memory represented through patterns across enormous datasets rather than individual lived experience.
Closely related to hypermnesia.
Style Leakage
The tendency of image models to reproduce characteristics of artistic styles without explicit prompting.
Raises questions about influence, authorship and copyright.
Taste
The aesthetic preferences guiding judgement.
Current AI research increasingly attempts to model taste computationally.
Relevance
Your methodology is not driven by taste.
It is driven by recognition.
The AI Art Director
An emerging description of artists working with generative systems.
Creative labour shifts from making images towards selecting and directing them.
Relevance
Useful, but I think your concept of the navigator is richer because it includes uncertainty and discovery.
The New Artistic Medium
The emerging view that generative AI represents a genuinely new artistic medium comparable to photography or film rather than simply another digital tool.
The medium consists not of pixels but of latent computational possibility.
Thinking With Images
Another phrase closely related to Image Thinking.
The idea that images are not endpoints of cognition.
They become active participants in reasoning, memory and emotional understanding.
Three terms I think we should introduce
These don’t yet exist in the literature, but I think they’re emerging naturally from your research.
Emotional Navigation
The process of navigating latent space according to affective rather than aesthetic criteria.
Phenomenology of Recognition
The study of how emotional recognition emerges during iterative engagement with AI-generated images.
I think this may ultimately become the theoretical heart of your thesis.
Emotional Holding Image
Borrowing from Winnicott’s concept of the holding environment.
An image that temporarily contains emotional experience sufficiently for reflection to occur.
I rather like this because it shifts the emphasis from representation to containment.
One correction
You mentioned “pathetic bias.”
I think you mean aesthetic bias.
There is, however, a literary term called pathetic fallacy, coined by John Ruskin, which describes attributing human emotions to nature, such as “the angry sea” or “the lonely wind.” That’s unrelated to the AI literature we’ve been discussing.
Looking over this glossary, one thing strikes me very strongly. About three quarters of these terms describe the AI system: latent space, mediated latent medium, aesthetic bias, optimisation, preference alignment, style leakage, hypermnesia.
Only a handful describe the human experience: recognition, affective truth, refusal, emotional navigation.
That imbalance is telling. It reinforces my growing sense that your contribution is not another theory of AI. It is a theory of what happens to human experience when thinking unfolds through AI-generated images. I suspect that distinction will become one of the defining features of Phantom Mirror.