AI projects · Evaluate · 1-2 weeks
AI image or real photo? A perception study
These projects are not aligned to any state standards or curriculum. They are generally about learning, understanding, using, and evaluating AI.
Ages 11-14 · This project can involve human data — written permission, anonymity, and offline/local handling are required by default.
Overview
People are not always good at telling AI-generated images from real photos — and the gap changes with image type and viewer experience. You will build a small perception study with consented volunteers, a balanced image set, and a clear accuracy calculation.
Age band
Ages 11–14 (middle school). An adult approves every image.
Time needed
1–2 weeks.
AI goal
Evaluate — measure human accuracy on a fixed image set and report uncertainty from a small sample.
Materials
Free tools only; no paid accounts required.
- 10 real photos the student took or otherwise has the right to use (objects or places). Do not use photos of real people without permission.
- 10 AI images from a free image tool, same broad subjects (objects, landscapes). An adult reviews every image for safety. Do not create deepfakes or other likenesses of real people.
- Printed packets or a local slideshow (offline). Do not upload volunteer responses to this site, to a third-party form, or to an AI tool by default.
- Written consent forms; answer sheets with anonymous IDs.
- Spreadsheet for scoring.
Human-data defaults (required): A parent or guardian must okay this project before the student runs it. If it is used in class, the teacher must also okay it first. For every participant, explain what this activity is, and that they can say no or stop at any time, with no pressure from adults or classmates. If a participant is under 18, get written permission from that child's parent or guardian and the child's own yes. Use only tallies or anonymous codes (P1, P2, or Volunteer A/B). Do not record names, faces, voices, contact information, school names, or other identifiers. Do not post or share any individual's answers. If results are shared, share only group totals. Keep responses on paper or on the local device. Do not upload responses to this site, to a third-party form, or to an AI tool by default. Delete raw responses when the project ends.
Steps
- Build a balanced deck. 10 real + 10 AI. Resize so resolution does not trivially give away the answer. Shuffle and number 1–20. Keep a secret key.
- Pilot on yourself. Try the deck once; adjust any image that is unfair (for example, garbled text that screams "AI" in a set otherwise about nature).
- Recruit volunteers (aim for ≥8) with written permission. Guardians sign for minors.
- Run the test. Each volunteer marks Real / AI for each image, no talking. Time limit optional (for example 20 seconds per image) — if you use one, write it down and keep it constant.
- Score. For each volunteer, accuracy = correct / 20. Also compute accuracy on real-only and AI-only subsets (each /10).
- Aggregate. Report mean accuracy across volunteers, and the min–max range. Optional: majority vote accuracy across the deck.
- Error analysis. Which images fooled the most people? Categorize (hands, text, reflections, sky, etc.) without claiming a universal law.
- Write-up. Include consent count, anonymization method, and a limit: "This deck and these volunteers are not the whole world."
Ethics and safety note
Use only images the student made or has the right to use. No photos of real people without permission. No deepfakes or other AI likenesses of real people. No sexual or violent images. Volunteers can say no or quit at any time, with no pressure from adults or classmates. Do not publish photos of volunteers. Store answer sheets on paper or the local device, and delete raw responses when the project ends.
What done looks like
| Criterion | Not yet | Done |
|---|---|---|
| Deck | Unbalanced or unreviewed | 10/10 split, adult-approved, shuffled key |
| Consent | Informal | Written permission for every volunteer |
| Privacy | Names on posters | Anonymous IDs; local storage |
| Stats | One friend's guess | Mean + range for ≥8 volunteers (or honest smaller-N limit stated) |
Adaptations
Classroom
The teacher okays the project before students run it and holds the permission records. Keep images teacher-held. Consider staff volunteers if student consent is hard.
Homeschool
Family and neighbors with permission; keep the deck object-focused.
Weekend
Use 6+6 images and 5 volunteers; state the small-N limit loudly.
Research
Split volunteers into "timed" vs "untimed" groups if you have enough people; compare means.
Science fair
Show aggregate graphs and example images that are yours to display; no volunteer photos.