AIQ
AIQ
Part of the Your AI Rules lesson guide. Teaching a different grade? 🌈 Explorer (5–7) · 🔧 Builder (8–10) · ⚡ Architect (15–18)
Open with the lesson's own framing, delivered straight rather than dumbed down — this age group responds better to being treated as capable of engaging with the real stakes, not a simplified version of them:
"You now have a comprehensive understanding of AI's capabilities, limitations, and ethical dimensions. Let's synthesize this into a personal AI ethics framework."
Ask the class to spend 60 seconds individually jotting down the single AI-related issue from this world (bias, privacy, deepfakes, environmental cost, or accountability) they found most concerning, and why. Cold-call two or three students to share. Push for specificity: if someone says "privacy," ask "privacy from what, specifically — an app knowing your location, or a company selling your data to someone else, or something else entirely?" Specificity here matters because vague answers ("AI is scary," "AI is bad") are exactly what this lesson is meant to move students past.
Then set the frame for the lesson: "Today you're not learning a new problem. You're building the framework you'll use to evaluate every AI system you interact with for the rest of your life — what questions do you ask before trusting a new AI tool, what do you refuse to accept even if it's convenient, and where do you personally draw the line?" Make clear this is a synthesis lesson, not a new-content lesson — everything they need is already in their heads from the rest of World 5; today is about organizing it into something usable.
A useful way to close the warm-up: point out that the four earlier World 5 lessons each covered one axis of "AI can go wrong" in isolation — bias, privacy, deepfakes, environment — but real products fail along several axes at once. A single facial-recognition product, for instance, can simultaneously raise a fairness concern (accuracy differing across skin tones), a privacy concern (surveillance use), and an accountability concern (who's liable for a wrongful match). "Today's lesson is where we stop treating these as separate topics and start treating them as one connected set of questions you ask about any AI system."
Students work through the lesson (World 5, Hacker mode) individually — this age band's version is denser than Builder mode and rewards a slower pace with real discussion between scenes, rather than rushing to finish.
Scene 1 — Core Principles. Fairness, Transparency, Safety, and Human Control are presented with real-world weight: fairness across race, gender, age, disability, and geography; transparency as the right to know when and why AI is deciding something about you; safety as rigorous testing before deployment in high-stakes domains like healthcare, transport, and criminal justice; and human override as non-negotiable for life-and-death decisions. Have students, in small groups, pick one principle and identify a real AI application (a resume screener, a self-driving car, a content recommendation feed, a facial recognition system) where that principle is genuinely hard to guarantee — and articulate why it's hard, not just that it is. For example: a resume screener trained on a company's past (mostly male) hires will reproduce that skew even with good intentions, which is why "fairness" for a hiring tool requires deliberately checking outcomes across groups, not just trusting the training process. This resists the temptation to treat the four principles as easy checkboxes a company can simply tick off.
A follow-up worth asking each group after they've picked their example: "If your example company genuinely wanted to fix this, what's the very first concrete step they'd take — not the end goal, the first step?" This distinguishes students who understand a principle abstractly from students who can operationalize it. For fairness, a real first step might be "measure the error rate separately across groups before deciding whether there's even a problem" — you can't fix what you haven't measured, and measurement itself is a nontrivial, resource-intensive step many companies skip.
Scene 2 — Your Choices Matter. Frame this scene as collective action, not individual virtue: one person's app choice barely moves anything, but patterns across millions of users are exactly what shapes product decisions, app store policies, and eventually regulation. Ask: "Can you name a real case where public pressure — reviews, media coverage, protests, boycotts — changed how a tech company's AI product worked?" If students are stuck, you can offer a general, defensible example: multiple facial-recognition and hiring-AI products have been pulled back, restricted, or retrained after public reporting exposed bias in them, which shows that user and press pressure is a real lever, not a symbolic one.
Scene 3 — Design Your Rules. The four concrete rules (mandatory bias testing before use, mandatory AI self-disclosure, data minimization, and non-delegable accountability) are the lesson's synthesis. Run this as a structured debate rather than a worksheet: assign each group one rule and have them argue against it for two minutes, steelmanning the hardest real objection a company lawyer or engineer might actually raise — for instance, "collecting only the minimum data can make a model less accurate, which could itself cause harm in a medical or safety application" is a real tension, not a strawman. Then have the same group argue for the rule, directly addressing the objection they just raised, rather than ignoring it. This is the kind of reasoning a genuine personal ethics framework requires: knowing the tradeoffs on both sides, not just memorizing the conclusion.
Close the activity by having each student write, individually, a two-to-three sentence personal rule that goes beyond the app's four — something specific to a technology or situation they actually encounter (school AI tools, generative AI for homework, a social platform's recommendation feed). This individual step is what turns a class discussion into something each student actually owns.
A useful framing device while circulating during Scene 3: ask each group "who benefits and who could be harmed if this rule is followed, and who benefits and who could be harmed if it's ignored?" for their assigned rule. This "who benefits, who's harmed" question is a simplified but genuinely useful version of the stakeholder analysis that real AI ethics reviews use, and it scales naturally into the more formal impact-assessment language this age band met in lesson 5.2.2. A group assigned the data-minimization rule, for instance, should be able to say something like: "If followed, users are protected if the company is ever breached, but the company might build a slightly less personalized product. If ignored, the product might feel smarter, but users are more exposed if that data ever leaks or is subpoenaed."
For a stronger class, push past the first answer on any of these — the goal is not agreement, it's for students to notice when they're relying on an assumption they haven't actually examined (for example, that "the company" is a single decision-maker, when in reality it's usually dozens of people across several teams with different incentives). Letting two students genuinely disagree and defend their positions for a minute is more valuable at this age than reaching a tidy class consensus.
Close with: "The framework you built today isn't academic — it's the actual lens you'll use every time a new AI product launches for the rest of your life. Companies, regulators, and courts are still figuring a lot of this out in real time, which means your generation's judgment on this will genuinely matter more than you might expect right now."
Extension activity: Assign a short "AI ethics case study" for homework or a follow-up class period: have each student pick one real, documented AI controversy (a biased hiring algorithm, a facial recognition misuse case, a chatbot that produced harmful content, a deepfake scandal) via a quick, teacher-approved search, and write a one-page analysis applying this lesson's four rules — which rule or rules were violated, what evidence supports that, and what a fix consistent with those rules would have looked like in practice. Have a few students present their case to the class and take questions, treating it like a short case-study presentation rather than a book report.
A grading note if you want to score this: reward students for pointing to specific evidence (a news article, a company statement, a documented outcome) over general impressions, and reward proposed fixes that are concrete enough someone could actually implement them — "the company should be more careful" is weaker than "the company should have tested the model's error rate separately for each demographic group before deployment, the way lesson 5.1.1 describes."