Benjamin & Lindley, TEI 2024 · Artifact paper · Read 2026-07-31

1. What they did

Visitors cast a shadow in front of a spotlight; a camera captures it; Stable Diffusion processes it img2img; a screen shows the generated image. Four components: screen, camera, local AI, display.

The interaction is one I know intimately — a camera feeds a body into a local image-to-image model and another kind of image comes out.

2. Method

Installation art plus exhibition presentation plus critical reflection. No formal user study.

3. Findings and contribution

  • The shadow is a meaningful interface between body and AI — no prompt typing required.
  • ML uncertainty as design material, not as a bug.
  • Entoptic phenomena: visual phenomena produced by the AI’s own internal mechanism rather than by what it is looking at.
  • Rancière’s theory reframes the body–AI relationship from use to expression.
  • A usable procedure for deploying a theory at all: describe the theory → locate its A/A* and B/B* → map your own installation onto that structure.

4. What it gives me

  • A structure to imitate: installation description (what) → artistic rationale + research rationale (why) → dialogue with theory.
  • The four-component sentence pattern and the way the interaction flow is written can be lifted directly.
  • My Brutalist project is very close to this — webcam captures body and material, diffusion generates architecture. The difference is what we each attend to: they care about AI’s uncertainty, I care about the power relationship between people and buildings.
  • It separates interactivity from playfulness from gamefulness. Shadowplay has the first two and not the third — no rules, no win condition, no feedback loop. Doing the gameful version is where I go beyond it.
  • No user study → adding a light qualitative study is already a contribution.
  • The “academic argument” in an artistic statement turns out to be conceptual connection and meaning-making, not data analysis.

5. Questions it left me

  • What if the shadow were fed not to an AI but to another person’s screen? Multi-person, collaborative or adversarial.
  • My thesis could sit here: one layer of user study more than a pure artwork paper, one real installation more than a pure empirical paper.
  • My own definition — gamefulness requires a feedback mechanism — could go in the Introduction as motivation.