Teaching "Your AI Rules" to Architect mode (ages 15–18)

Hook & Warm-Up

This is the final lesson of World 5, and the age-band hook signals that explicitly. Deliver it close to verbatim, at a register that treats students as near-adults capable of engaging the real literature and debate around this topic, because that is exactly what the rest of the lesson asks of them:

"Having examined AI's technical foundations, applications, and societal implications, you're now equipped to develop a principled framework for engaging with AI as a technologist, user, and citizen. Let's synthesize."

Open with a real tension rather than a review question: "Companies building AI right now — often people not much older than you'll be when you enter the workforce — are making judgment calls about fairness, disclosure, and safety with incomplete regulation, incomplete science, and real commercial pressure bearing down on every decision. There is no settled, universally agreed answer to most of what we're about to discuss today. Your job is not to memorize 'the correct ethics' as if it were a fixed answer key — it's to build a framework you can actually reason from when you hit a case nobody has given you the answer to, because you will hit exactly that kind of case."

Ask: "Who here might end up building, deploying, regulating, or just heavily relying on AI systems in their career — which, statistically, is going to be most of you in some form, whether or not you go into a technical field? What's one decision in that future role you'd want to be equipped to make well?" Use two or three responses to set expectations for the discussion ahead, and resist the urge to correct or steer these early answers — the goal here is buy-in, not precision.

You might also note, briefly, why this lesson exists at all: none of the earlier World 5 lessons told students what to do with what they learned — they described problems (biased data, surveillance economics, deepfake detection, AI's carbon footprint, the accountability gap) without asking students to commit to a position. This lesson is the first one that asks for a stance, and stances are harder and more exposing than descriptions. Naming that explicitly tends to raise the seriousness of the room.

Main Activity

Students work through the lesson (World 5, Architect mode) independently, then reconvene for discussion — the content itself is close in substance to Hacker mode at the scene level, but the surrounding activity should push considerably further into genuine tradeoff analysis and connect explicitly to material this age band covered in more technical depth earlier in World 5: algorithmic fairness definitions, the impossibility theorem, and disparate-impact law (5.1.1), GDPR and differential privacy (5.1.2), the C2PA provenance standard (5.1.3), lifecycle carbon accounting (5.2.1), and the EU AI Act's risk tiers (5.2.2).

Scene 1 — Core Principles. Have students restate each of the four principles (Fairness, Transparency, Safety, Human Control) as a testable claim about a system, not a value statement — for example, not "AI should be fair" but "this system's error rate should not differ meaningfully across protected groups on the outcome that matters." Connect back explicitly to lesson 5.1.1's point that fairness definitions (demographic parity, equalized odds, calibration) can be mathematically incompatible with each other except in trivial cases: ask, "when this lesson says 'AI should be fair,' which formal definition of fairness is it even assuming?" There often isn't a clean answer, and that's the point — plain-language ethics principles like the ones in this lesson are a starting point for a harder conversation, not a finished specification a regulator could hand to an engineer unmodified. Push one level further if the class is ready for it: ask whether it's even coherent to demand all four principles simultaneously in every deployment context, or whether real systems necessarily trade some fairness for some safety margin, or some transparency for some competitive or security concern. Naming that tradeoffs are sometimes unavoidable — not just difficult, but genuinely unavoidable given the mathematics or the domain — is a more mature position than insisting all four can always be maximized at once.

Scene 2 — Your Choices Matter. Extend beyond individual consumer choice into the student's plausible future role: as a builder of AI systems, a person subject to them, and a citizen who votes and can support or oppose regulation. Ask each student to identify which of those three roles feels most personally relevant to their likely path, and what "your choices matter" looks like concretely in that role — for example, a future engineer choosing which bias tests to run and how strict a threshold to set before shipping; a future product manager deciding whether a disclosure banner is prominent or buried; a future voter or civil servant deciding which AI-regulation approach (EU-style risk tiers versus the US's more sector-specific approach, both covered in 5.2.2) actually protects people without freezing beneficial innovation.

Scene 3 — Design Your Rules. Treat the four stated rules as a floor, not a ceiling, and have students identify precisely what each rule leaves unresolved. For example: "AI must be tested for bias against all groups" — tested against which fairness metric, with what numerical threshold for "passing," audited by whom, and with what consequence for failure? "Someone must be responsible for every AI decision" — is that meant as legally enforceable liability (like product liability law, which assumes a foreseeable causation chain that AI systems often don't fit cleanly), or just a norm with no real enforcement teeth? Have students draft an extension to one rule that closes a gap they've identified, written in the style of an actual policy clause — specific enough that a regulator or a company's legal team could implement it, not just a restated value with no operational detail.

A useful closing move for Scene 3, if time allows: have students briefly compare their extended rule against how an actual regulation handles the same gap — for example, the EU AI Act's conformity-assessment requirement for high-risk systems is one real answer to "audited by whom," and it's worth noting where the class's own draft ends up more or less demanding than the real regulatory text. This isn't about getting the "right" answer; it's about calibrating how hard real-world AI governance actually is once you get past the level of a mission statement.

This age group can also do that comparison against a regulation that applies directly to their own school, not just a foreign one: the Philippines' Department of Education issued DepEd Order No. 003, s. 2026, "Foundational Guidelines on Artificial Intelligence in Basic Education," in February 2026, following Project AGAP.AI's January 2026 launch with the ASEAN Foundation and Google.org. Have students read AIQ's summary of what DO 003 actually requires — a mandatory Privacy Impact Assessment and AI Registry entry before a school deploys any AI tool, and added safeguards for direct AI interaction in Kindergarten through Grade 3 — and ask: is DO 003 closer to the EU's risk-tier model or the more sector-specific US approach covered in 5.2.2? What does it leave unresolved the same way this lesson's own four rules do? A student heading into policy, law, or product work in the Philippines specifically will encounter this exact document, not a hypothetical one, which makes it worth reading closely rather than treating as a footnote.

Discussion

• "The lesson says multiple ethical frameworks illuminate different aspects of AI dilemmas — pick a real AI controversy you know of and show how a purely fairness-based analysis and a purely accountability-based analysis would lead you to different recommendations for fixing it."
• "AI ethics as praxis means integrating ethical reflection into actually building and deploying systems, not treating it as a separate compliance step bolted on afterward. What would that look like concretely inside a real product team's sprint-by-sprint workflow?"
• "Regulation (like the EU AI Act) moves slower than AI capability does. Is voluntary industry self-regulation a credible substitute in the meantime, or mostly a stalling tactic that delays real accountability? Defend your answer with a specific example if you can."
• "If you were advising a startup building an AI product, which of the four rules would you expect them to resist most on commercial grounds, and what argument would you actually make to convince them to follow it anyway?"
• "Your generation is described as having both 'native intuition' about AI and responsibility for shaping it. Do you think that responsibility is fairly placed on you, given you didn't build the systems you inherited and didn't get a vote on how they were designed? Why or why not?"

These questions don't have a single correct resolution, and that's deliberate — this is the one lesson in the world where the goal is genuinely open-ended reasoning rather than converging on a stated answer. Let disagreement stand where it's well-argued, and push students to name the actual crux of their disagreement — a differing fact, a differing value, or a differing prediction about the future — rather than letting the discussion end in a vague "everyone has a point."

Quiz Walkthrough

Wrap-Up & Extension

Close with: "You've now built a framework that combines technical literacy with an explicit set of values. That combination is genuinely rare — most people in positions to make decisions about AI right now have strength in one and not the other. Whatever you end up doing with AI in your career, whether building it, regulating it, or simply living with it as an informed citizen, you're better equipped than a lot of the adults currently making consequential decisions about it."

Extension activity: Have students draft a one-page "AI Ethics Charter" as if they were submitting a genuine public comment to a real regulatory body — framing it as a proposed amendment or addendum to DepEd Order No. 003, s. 2026, the Philippines' actual foundational guidelines on AI in basic education (see AIQ's guide for what it currently covers), or using the EU AI Act's risk-tier structure (from lesson 5.2.2) as a reference model if a student prefers a comparative approach. The charter should include three to five specific, enforceable rules that go beyond this lesson's four, each with a one-sentence rationale grounded in a real, documented harm from earlier in World 5 — not a hypothetical one. Have a few students read theirs aloud and open it to brief class critique: is each rule specific enough that a regulator or company could actually implement and audit it, or does it collapse back into a vague value statement that sounds good but resolves nothing?

If your school has any connection to a local university, policy program, or tech-industry contact, this charter exercise is also a natural piece of work to actually send somewhere — even just to a class blog or a school newsletter — rather than filing it away ungraded. Treating the output as writing meant for a real audience, however small, tends to sharpen the specificity of what students produce far more than "this is for a grade" does on its own.