Group chat explodes over a clip—half 🔥, half “fake.” This module gives vocabulary, frameworks, and debate habits so you slow down, ask better questions, and argue with evidence instead of volume.
You’re not too young to shape norms—clear teen voices get invited into real policy rooms.
Time: ~90 min read + 4–6 hours over two weeks for briefs and practice debates.
Learning Objectives
By the end of this module, you will be able to:
- Use terms: bias, privacy, surveillance, consent, transparency, accountability, deepfake, labor.
- Analyze school / creative / hiring / media dilemmas from multiple stakeholder views.
- Run a structured debate: claim, warrant, evidence, rebuttal, impact weighing.
- Spot weak rhetoric: straw man, false dilemma, fear sells, anthropomorphizing “the AI.”
- Hold a nuanced personal stance—neither hype nor blanket rejection.
- Facilitate civil discussions with ground rules and time limits.
1. Friday Assembly Scenario
New “learning safety” scanner on devices—some cheer, some whisper dystopia, teachers fear false positives.
Precision questions: What is collected? Who sees alerts? What happens after a flag? Appeals? Can students read the actual policy?
Concept: Many tools mix rules + statistics; mistakes can hit marginalized students hardest if adults skip review.
Try This! Write eight student Q&A questions—star the three that matter most to your safety and dignity.
2. Vocabulary Table
| Term | Teen-friendly |
|---|---|
| Bias | Systematic worse outcomes for some groups |
| Fairness | Competing definitions—no single math of “fair” |
| Privacy | Control over your data and story |
| Surveillance | Watch + record + often analyze |
| Consent | Informed yes |
| Transparency | Enough visibility to trust or audit |
| Accountability | Who answers when harm happens |
| Deepfake | Synthetic media mimicking real people |
| Labor | Humans training, labeling, moderating—often invisible |
Pro Tip: Define terms before debates—otherwise people talk past each other.
Activity: Pick two terms; write a personal school/social example for each.
3. Deepfakes
Harms: bullying, fraud, politics, non-consensual intimate imagery (illegal many places—never OK, never spread).
Defenses: imperfect detection, metadata ideas, media literacy, platform rules—not complete fixes.
Fun Fact: We trust faces/voices quickly—pausing one beat before sharing is civic skill.
Try This! Summarize limits of detection from one reputable lab or newspaper—two sentences what works, two what fails.
4. Bias
Models learn from history—history includes inequality. Lazy design + skewed data hurts without mustache-twirling villains.
Three questions: Who is harmed when this fails? How often, and who monitors? What mitigations (data, audits, appeals)?
Concept: Marketing “unbiased AI” should trigger gentle skepticism.
5. Privacy as Control
Not only secrecy—who sees what, when, why. Free apps monetize attention/data; school accounts may differ from personal phones.
Habits: review permissions; minimize location; “cafeteria test” before cloud-pasting sensitive stuff.
Discussion: Is “nothing to hide” a strong privacy argument? Prepare two respectful counters.
6. AI in Schools — Center Learning
For: tutoring, translation, accessibility, faster feedback when teachers are stretched.
Against: thinking substitution on graded work; tool inequity; vague integrity rules.
Middle path many teachers want: practice + bounded drafting + disclosure—not proxy brain on exams.
Try This! Five-minute educator interview: what AI use would they celebrate vs red-flag?
7. Jobs — Better Questions Than Doom
Tasks automate before whole jobs vanish; transitions uneven; new roles in oversight, design, policy, teaching.
Try This! Three questions for a politician making bold AI job claims—make them specific.
8. Debate Formats That Stay Civil
Team timing sketch: openings → constructives → short cross → rebuttals → closings on impacts, not personalities.
Values format: pick a value (justice) + criterion (protect the vulnerable)—weigh args through that lens.
Ground rules: no ad hominem; cite stats; steel-man opposing best case first.
Try This! 5-minute silly motion (“pineapple pizza”) only to practice timing—then swap to a real AI motion.
9. Fallacies in AI Discourse
| Fallacy | Example vibe |
|---|---|
| Appeal to fear | “Or else humanity ends” with no mechanism |
| False dilemma | “Open everything or lock everything” |
| Anthropomorphism | “It wants power” |
| Cherry-picked demo | Viral clip, no edge cases |
Activity: One viral AI clip—list two missing contexts.
10. Opinions Without Performing Certainty
Healthy: “Today I think X because [sources]. I’d change my mind if ___.” Unhealthy: certainty for likes.
Reflection: Half page—one school AI policy you want, one risk you accept, one line you won’t cross.
11. Climate / Energy Stakeholder
Training and serving models use electricity and cooling—ask who pays, what’s marketing vs measured, whether smaller models could suffice.
Try This! One primary source (company sustainability PDF, agency note)—three bullets: measured, uncertain, what you’d want disclosed next.
12. Data Brokers & “Free” Apps
Brokers build profiles from clicks, purchases, permissions. Minimize data; say hard no to creepy permission asks.
Activity: Phone privacy settings—revoke one thing you don’t need; two sentences on how it felt.
13. Ethics One-Pager for Interviews
Fill: issue you care about; stakeholder map; one real-world lever; your action habit.
Try This! Under 200 words—read to a friend; can they name issue, stakeholder, your move without jargon?
Practice Challenges
Motion prep with six bullets each side; disclosure debate with definitions first; facial recognition stakeholder letters; deepfake law brainstorm; fallacy headline rewrite; ten-minute facilitated discussion with posted rules; stakeholder wheel; repair story where a company actually fixed harm.
Your Challenge
Research brief (350–450 words): Pick deepfakes, school surveillance, art training data, or automation in a field you care about—neutral summary of two serious views; five sources (mix news + primary); closing paragraph: your recommendation + what would change your mind.
Discuss: Did you steel-man the side you disagree with?
Key Takeaways
- Ethics = structured thinking about power and tradeoffs.
- Deepfakes and bias are real; tech + law fixes are partial.
- Privacy is contextual control.
- Debates improve with definitions, steel-manning, and cited evidence.
Key Takeaway
- Name stakeholders before you pick teams in an argument—most fights are incomplete maps.
- Deepfakes target trust in faces/voices—slow sharing is a skill.
- Bias and surveillance need ongoing design + policy, not one-time promises.
- School AI debates should center learning, fairness, and clear appeals.
- Strong opinions include falsifiers—what would actually change your mind?
Going Further
Read one official model limitations section; join debate/MUN; pair with Modules 01 & 03; follow one careful, citing journalist—mute pure rage accounts for a week and notice mood.
Facilitator Notes (Teachers / Club)
Rotate roles; require good-faith summary of opposing view before rebuttal; model “I don’t know yet” as strength.
Disagreement isn’t disrespect—in the AI era, how we argue shapes what we decide.