AI projects · Evaluate · 1-2 weeks

Reproduce a published AI number, then say what it doesn't mean

These projects are not aligned to any state standards or curriculum. They are generally about learning, understanding, using, and evaluating AI.

Ages 11-14

Overview

News stories and blogs often quote a single AI statistic ("X% of students…", "model scores Y on Z"). You will pick one publicly checkable number, try to find the original source, reproduce the claim as far as the public data allows, and write what the number does not mean.

Age band

Ages 11–14 (middle school). An adult helps evaluate whether a source is primary.

Time needed

1–2 weeks.

AI goal

Evaluate — trace a claim to a source and separate the measured quantity from the headline.

Materials

Free tools only; no paid accounts required.

  • A web browser and a notes doc.
  • One recent public article that cites an AI-related number (school newsletter, reputable news site, or a lab blog). Prefer links you can open without a subscription wall.
  • Spreadsheet or paper for "claim → source → method → sample → date."
  • Calculator.

Rule: if you cannot open the primary source, mark the chain UNVERIFIED and switch claims — do not invent the underlying study.

Steps

  1. Capture the headline claim verbatim, with article title, publisher, and date.
  2. Find the number's home. Look for a link to a paper, dashboard, or official report. If the article has no link, search the exact statistic in quotes plus the topic. Stop after 30 minutes if nothing primary appears; pick a different article.
  3. Extract the real quantity. In the primary source, write:
    • Exact wording of the measured quantity
    • Sample size / dataset name if given
    • Date of data collection
    • Any confidence interval or margin of error if given (write "not reported" if absent)
  4. Reproduce what you can. If the source provides a public table or CSV you can open, recompute the percentage or mean yourself and show the arithmetic. If the data are not public, write "not reproducible from public files" — that is a valid result.
  5. List three non-meanings. Example pattern: "This does not mean every school district…"; "This does not mean the model is safe…"; "This does not mean the rate applies to ages 11–14…"
  6. Write a half-page brief a principal could read: claim, source, what checks out, what is unverified, what the number does not mean.

Ethics and safety note

Do not harass researchers. Do not scrape behind logins. Do not accuse authors of fraud because a number is hard to reproduce — say what is missing. Prefer public links; mark paywalled or dead links UNVERIFIED.

What done looks like

CriterionNot yetDone
ChainOnly the news articlePrimary source identified or claim abandoned as unverified
QuantityVague paraphraseExact measured quantity + date + sample notes
Reproduction"Seems right"Arithmetic shown or explicit "not publicly reproducible"
Non-meaningsNoneThree clear non-meanings

Adaptations

Classroom

Teacher pre-approves a menu of articles so every claim is age-appropriate.

Homeschool

Use one article from a library database if home links are paywalled; still demand a primary citation.

Weekend

Use a claim that already links to a public dashboard so reproduction fits one day.

Research

Compare how two outlets summarize the same primary number; note distortions.

Science fair

Poster with a photocopy/print of the primary table (fair use for education) beside your recomputation.

All AI projects

Reproduce a published AI number, then say wha… · Schoolhouse