› Grades 6–8 · ages 11–14 · Fishbowl
Should AI decide things about people?
“AI should be allowed to make decisions about people — who gets a loan, a job interview, or a warning from the police.”
A fishbowl discussion on automated decision-making, with real cases students can reason about. Half the class talks while the other half tracks claims and evidence, then they swap. The concepts — bias, explainability, human in the loop — come from what went wrong in the cases.
55 minutes · one class period · free to photocopy
Students will be able to
- Explain how a system trained on past decisions can inherit past unfairness.
- Argue for or against automated decisions in a specific domain using a real case.
- Propose a safeguard (explanation, appeal, human review) and say what it costs.
Vocabulary
- Bias (in data and models) — Systematic skew — who’s represented in training data and who gets hurt in outputs.
- Explainability — Can we understand why a model produced a specific output?
- Human in the loop — A person reviews or approves before action goes live.
- Training data — The examples a model learned from — often the whole internet, licensed corpora, or both.
- Automation bias — Trusting the machine because it’s fast and confident — even when it’s wrong.
Agenda · 55 minutes
| Minutes | Segment | What happens |
|---|---|---|
| 0–6 | Warm-up | Résumé screener case. Responsibility poll. |
| 6–14 | Cases + vocabulary | Read the three cases (hiring, loans, policing) and the five vocabulary cards. Each student writes one claim and one question. |
| 14–26 | Fishbowl round 1 | Inner circle of 6 + one empty chair discusses hiring and loans. Outer circle tallies: claims, evidence, questions, vocabulary used. |
| 26–38 | Fishbowl round 2 (swap) | New inner circle discusses policing and the safeguards question. Same tallies. |
| 38–48 | Safeguard design | Groups of four propose ONE safeguard for ONE domain and state its cost (time, money, fewer decisions). |
| 48–55 | Re-poll + exit ticket | Repeat the responsibility poll. Discuss what moved. Handout. |
Perspective — let the system decide, with audits
Humans are biased and slow; a model can be tested for bias and fixed in ways a person can't.
- You can run a million test cases on a model. You can't on a hiring manager.
- Consistent rules applied the same way to everyone is a kind of fairness.
- Speed matters: a loan decision in seconds instead of weeks helps people too.
Perspective — a person must decide, the system may advise
A decision about a person needs a reason the person can hear and challenge.
- If nobody can explain the decision, nobody can appeal it.
- The model learned from a biased past; it will repeat it faster.
- 'Human in the loop' only works if the human actually looks.
Discussion questions
- 1.openingWho was responsible for the résumé screener's bias?
- 2.coreIs a consistent unfair rule better or worse than an inconsistent fair-ish human?
- 3.coreShould you have the right to an explanation for any decision made about you? By a person, too?
- 4.coreIs a human reviewer a real safeguard if they approve 99% of what the model says?
- 5.closingPick one domain. Where exactly should the human be in the loop, and what does that cost?
How fishbowl runs
An inner circle of 5–7 students discusses while the outer circle observes with a listening task (tally claims, evidence, questions). One inner-circle chair stays empty: any observer may tap in, and someone rotates out. Swap circles halfway so everyone speaks and everyone listens.
In grades 6–8
Evidence enters the room. Students can hold a position they don't believe, and that's the point.
Preview the full pack
Cover, teacher guide, and every student handout. Print teacher pages once and student pages one per learner.
On a phone, open the PDF in a new tab — in-page previews are unreliable in mobile Safari.
Standards
- CCSS ELA SL.7.1 / SL.8.1
- CCSS ELA SL.7.3 / SL.8.3 (evaluate reasoning and evidence)
- CSTA 2-IC-20 / 2-IC-21 (bias and accessibility in computing)
- C3 Framework D2.Civ.10 (rights and responsibilities)
More for grades 6–8
Taught it? Tell us how it went: send classroom notes · AI concepts field guide · Try the Activities Pass free · Commission a full unit