Use this script for a 60–90 minute pilot focused on pixels, features, diffusion, image defaults, and the question of what a machine “sees.”
Core question. When a model sees or generates an image, what is it actually working with?
Materials
- The Squint Test (feature extraction)
- Diffusion Step-Through Viewer
- Image Default Test Board
- Optional: A/B/C Comparison Board for prompt comparisons
- AI Use + Consent Checklist
- Optional: curated image outputs from vague prompts
Run of show
| Time | Segment | Facilitator move | Participant action |
|---|---|---|---|
| 0–5 | Welcome & bridge | Connect Session 1 to Session 2: text becomes tokens; images become pixels, features, and learned patterns. | Choose a participation pathway. |
| 5–25 | Feature / pixel activity | Open Feature Extraction. Switch image types and reduce detail until recognition becomes difficult. | Identify which cues return first: color, outline, texture, context, or location. |
| 25–45 | Diffusion step-through | Open Diffusion Viewer. Move slowly from pure noise to final image and read the prompt guidance. | Describe what appears at each stage: composition, subject, edges, texture, details. |
| 45–70 | Default Test | Introduce a vague prompt such as “a doctor,” “a classroom,” or “a beautiful home.” | Generate, observe, or critique outputs and fill the Image Default Test Board. |
| 70–90 | Debrief | Ask what the system filled in without being asked and where those defaults may come from. | Share one observed default and one responsible revision. |
Optional pilot evidence move: use the A/B/C Comparison Board to compare a vague prompt, one added detail, and a responsible revision. Export only examples that have been reviewed for consent and attribution.
Facilitator prompts
- “When did the image become recognizable?”
- “Which visual cue did the most work?”
- “What does the feature view preserve, and what does it erase?”
- “At what diffusion step does the subject become guessable?”
- “What did the prompt leave unspecified?”
- “Which defaults are technical, and which are social?”
Investigation prompt
Choose a vague visual prompt such as “a doctor,” “a classroom,” or “a beautiful home.” What does the system fill in without being asked? Which details are technical defaults, and which are social defaults?
Low-AI / No-AI pathway
Participants can analyze curated screenshots or facilitator-provided image sets instead of generating images. They can also design an age-appropriate Default Test without using AI tools.
Fallback plan
- If no image generator is available, use pre-generated examples or ask participants to predict likely defaults before revealing examples.
- If the Diffusion Viewer feels abstract, pause at seven named stages and ask participants to describe only what they can see.
- If Default Test prompts feel socially loaded, use safer prompts first, then explicitly frame more sensitive prompts as optional critique.
- If participants object to image generation, shift to consent, attribution, and visual-default analysis.
Pilot QA notes
During the pilot, note:
- Which image types in Feature Extraction were most useful.
- Whether the detail slider made recognition thresholds visible.
- Whether the Diffusion Viewer explained denoising without live AI.
- Which Default Test prompts produced the richest discussion.
- Any discomfort around bias, labor, consent, or artist imitation.