AIQ
AIQ
Everything a teacher needs to deliver this lesson — pick your grade's script below once you've read the background.
This lesson runs 5–15 minutes inside the app, in World 5: Is AI Fair? Where the earlier lessons in this world dealt with fairness in AI decisions and how AI collects personal data, this one turns to a different kind of harm: AI that generates media designed to look or sound like something real that never happened. Students earn the "Truth Seeker" badge for completing it.
By the end of the lesson, a student should be able to:
"Deepfake" comes from "deep learning" plus "fake" — media generated or altered by neural networks to depict people saying or doing things they never said or did. The category covers several distinct techniques the lesson touches on: face swapping (putting one person's face onto another person's body in video), voice cloning (generating new speech in someone's voice, in words they never spoke), and AI image generation (creating photorealistic images of people who don't exist at all — sites like thispersondoesnotexist.com make this vivid by generating a new invented face on every page load). None of this requires a film studio or a professional voice actor anymore; it requires training data (photos, video frames, or audio samples of a real person) and, increasingly, not very much of it.
That last point is the one most likely to surprise a class: modern voice cloning models can extract a usable "voiceprint" from as little as a few seconds of someone's recorded speech — a voicemail greeting, a social media video, a phone call — and then generate entirely new sentences in that voice, saying anything the person typing the script wants. This is not hypothetical. In one widely reported 2019 case, criminals used an AI-cloned voice that convincingly imitated a company's CEO to trick an employee into wiring $243,000 to a fraudulent account. The same technique, scaled down, powers the "grandparent scam" calls that have hit families where a cloned voice of a grandchild claims to be in trouble and asks for emergency money.
The lesson frames the danger in two layers, and it's worth keeping both in mind while teaching. The first is the obvious one: someone believes a fake. A convincing deepfake of a politician saying something inflammatory, timed right before an election, can genuinely mislead voters. The second layer is subtler and, for older students, arguably more important: the mere existence of convincing deepfakes gives bad actors a new excuse to dismiss real, damaging evidence as fake. Legal scholars call this the "liar's dividend" — someone caught on a genuine recording can now say "that's a deepfake" and get away with it, at least with part of the audience. Deepfakes don't just create new lies; they make it easier to escape the truth.
On detection: the lesson's own advice — checking for blurry edges, odd hand shapes, mismatched lighting, or unnatural blinking — was reliable a few years ago and is getting less reliable every year as generation models improve. More advanced detection tools now look for things invisible to the human eye: statistical irregularities in how pixels are compressed, or even subtle biological signals that real video preserves and synthetic video tends to lose, like the faint, rhythmic color changes in skin caused by a real pulse, or natural, involuntary blink timing. None of these tools are foolproof, and none of them stay ahead for long — every improvement in detection tends to get folded back into training better generators, which is why this lesson treats critical thinking and source-checking as the durable defense, not any single tool.
Older students also meet content provenance, a different approach that doesn't try to catch fakes after the fact at all. Standards like C2PA (Coalition for Content Provenance and Authenticity — backed by companies including Adobe, Microsoft, and the BBC) cryptographically sign metadata onto real photos and video the moment they're captured, creating a tamper-evident chain of custody from camera to publication. Rather than asking "does this look fake?" a provenance system lets you ask "can this file prove where it really came from?" Neither detection nor provenance alone solves the problem — provenance only helps for media that was signed at capture, and most content in circulation isn't — which is exactly why the lesson lands on media literacy (verify sources, question outrage-bait, ask who benefits from you believing something) as the one defense every student can use immediately, regardless of what tools exist.
No printing and no special setup. Each student (or pair, if devices are shared) needs a phone, tablet, or computer with a browser and the AIQ app loaded — the hook, the three learn scenes, the sorting practice, and the quiz all run on-device with no login required. Read the Teacher Background above once before class, especially the "liar's dividend" and content-provenance sections if you're teaching Hacker or Architect mode, since both terms appear directly in those quizzes.
If you plan to use the thispersondoesnotexist.com example mentioned in the Builder and Hacker activities below, it's worth loading it once yourself before class — it generates a new AI face every time the page refreshes, so you can't know in advance exactly what will appear, and you'll want to confirm your school's network allows the site. No other outside media or accounts are required.
Deliberately not required: showing students an actual deepfake video, audio clip, or scam call recording. Real examples circulating online aren't vetted for a classroom, and the lesson's own facts, plus the invented-face example above, are enough to teach the concept without needing to track one down. If a student brings one up unprompted, it's fine to discuss it verbally without playing it.