Cambridge Festival 2022

The Work

This work was shown at Cambridge Festival 2022 as part of the Self & Body Lab at Anglia Ruskin University. It represents one of my earliest explorations of generative AI as a medium for scientific communication and phenomenological inquiry — an attempt to render the subjective experience of dreaming and depersonalisation into something visible.

The raw material was dream reports and first-person accounts of depersonalisation gathered through live research. The question driving the work: what does it actually feel like to be dissociated from your body, to live as though inside a dream? Could emerging AI image models translate that phenomenology into something a viewer could experience rather than just read about?

An Interdisciplinary Starting Point

The project sits at the intersection of three fields I was working across simultaneously:

Cognitive science and consciousness research — I was working as Research Assistant in the Self & Body Lab, supporting a study investigating how depersonalisation affects bodily self-experience during waking and dreaming states. The research asked fundamental questions: What is a self? How does the brain generate the conscious experience of our lives? What is the relationship between the physical body and the mental self?

Phenomenology and art — these weren't abstract questions. Participants were describing lived experience — altered states, estrangement from the body, dreamlike perception of ordinary life. Translating that into a scientific paper captures the data. Translating it into moving image gets closer to the experience itself.

Generative AI as an early adopter — in early 2022, the tools available were a long way from the polished consumer interfaces that would arrive later that year. DALL-E 2 was invitation-only. Midjourney was in closed beta. Stable Diffusion didn't yet exist publicly. The generative art community had built VQGAN+CLIP and Disco Diffusion from scratch — open-source Python scripts run in Google Colab notebooks, shared directly on Twitter and Discord.

The Tools

The process used two architectures from the community-built generative art ecosystem:

VQGAN + CLIP — created by Katherine Crowson, this pairs a Vector Quantised GAN (which generates images) with CLIP (which scores how well an image matches a text prompt). VQGAN iteratively updates the image until CLIP's confidence peaks. In practice, you write a text prompt — drawn from a dream report or a phenomenological description — and watch the system find an image that inhabits that language.

Disco Diffusion — created by Somnai_dreams, this uses diffusion processes guided by CLIP to iteratively remove noise from a random image toward a target defined by text. The backward pass through noise generates images with a distinctive dreamlike texture that turned out to be well-suited to this subject matter.

Both tools were run using custom Python scripts directly — no UI, no prompt engineer workflow, no token credits. This was before the vocabulary of "prompt engineering" existed. It was experimental, technically demanding, and entirely community-driven.

Research Context

The outputs were grounded in real data. Text prompts were constructed from participants' own language — their descriptions of what depersonalisation feels like, their recalled dream content, the phenomenological qualities they reported: unreality, detachment, the sense of watching yourself from outside. The AI models became instruments for giving those descriptions an image.

This approach anticipated what would later become a recognised practice in science communication — using generative AI to bridge the gap between quantitative findings and subjective experience. At the time, it was simply a genuine question: can this tool do something a paper cannot?

Significance

This was genuinely early adoption — not early in the sense of being among the first to open Midjourney, but early in the sense of building workflows with tools that had no UI, learning the architectures from community documentation, and applying them to a specific research and artistic problem before any mainstream cultural moment around AI image generation had arrived.

The work was shown alongside research posters from the Self & Body Lab at Cambridge Festival 2022, positioning AI art not as a novelty but as a research instrument — a way of exploring the same questions about consciousness and self that the lab was pursuing through formal experimental methods.

Reference Materials