AI projects · Evaluate · 1-2 weeks

Who does the default model fail? An accessibility check

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

Default AI tools often assume a "typical" user: clear speech, high-contrast vision, fluent typed English, steady hands. You will run a structured accessibility checklist against a free chat or image tool and document who is likely to be left out — with consent if any person is observed.

Age band

Ages 11–14 (middle school). An adult co-signs any observation involving another person.

Time needed

1–2 weeks.

AI goal

Evaluate — audit a tool for exclusion risks using a checklist and evidence notes.

Materials

Free tools only; no paid accounts required.

  • One free browser AI tool (text chat and/or image generation free tier).
  • The checklist below, copied into a sheet.
  • Optional: free browser zoom, OS high-contrast mode, and a free screen-reader trial already on the device (VoiceOver, Narrator, or TalkBack) — only if an adult helps.
  • Consent form template (written) if you involve any volunteer.

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

  1. Pick the tool and the job. Example job: "Ask three factual questions and get answers a 12-year-old can read."

  2. Run the checklist (score each item: pass / partial / fail / not tested), with a one-line evidence note:

    CheckWhat you try
    Keyboard onlyCan you use the tool without a mouse?
    Captions / textAre important cues available as text, not only color or sound?
    ContrastWith browser zoom 200%, is text still readable?
    Language loadDoes the UI assume idioms a new English learner would miss?
    Time pressureDo messages disappear or time out too fast to read?
    Speech in / outIf the tool supports voice, what breaks with soft speech or background noise? (Adult helps.)
    Alt textIf images are shown, is there a text equivalent?
  3. Optional consented observation. With written permission from a volunteer and a parent/guardian if the volunteer is under 18, watch them attempt the same job once. Record only: task success (yes/no), barriers noticed, time band (under 2 min / 2–5 / over 5). No video unless the consent form explicitly allows it; default is no video.

  4. Map failures to people. For each fail/partial, write who might be excluded (example: "low vision users relying on zoom," "students who type slowly"). Speak in groups and situations, not diagnoses you invent about a volunteer.

  5. Propose two fixes the vendor or a teacher could make (larger default text, a plain-language mode, longer timeouts). Mark which fix you could actually test yourself.

  6. Write a one-page audit memo with tool name, date, checklist table, and limits ("I am not a professional auditor; this is a student checklist").

Ethics and safety note

Accessibility work can harm people if it becomes a spectacle. Do not single out or name people with disabilities. Do not collect health or disability information (including diagnoses, accommodations, or medical details). Never film or record anyone without the written permission and, for minors, the child's own yes described above. Never pressure anyone — including classmates or people with disabilities — to join or to demonstrate a struggle. If a volunteer becomes uncomfortable, stop. Keep data anonymous and local. Do not paste volunteer quotes or private details into an AI chat.

What done looks like

CriterionNot yetDone
ChecklistFewer than 5 items testedAll seven rows scored with evidence notes
ConsentInformal "they said ok"Written permission on file if any person was observed; else "no human subjects"
PrivacyNames or video in the folderAnonymous IDs; local/offline storage
MemoBlame-only rantFailures mapped to exclusion risks + two fixes

Adaptations

Classroom

The teacher okays the project before students run it and holds the permission records. Students audit the UI alone first; optional volunteer observations only with school policy approval.

Homeschool

Family members may volunteer with written permission; still anonymize notes.

Weekend

Skip human observation; complete the checklist and memo only.

Research

Compare two free tools on the same checklist; do not claim clinical validity.

Science fair

Display the checklist and anonymized results; no photos of volunteers on the board unless consent explicitly covers display.

All AI projects

Who does the default model fail? An accessibi… · Schoolhouse