Room 04

The real worries.

Some worries about AI are silly (robots taking over the world, mostly). Some are real and worth knowing. Four activities about the real ones — fake faces, what AI can't know about you, when using AI is cheating, and the trick AI pets play on your feelings.

Exhibit 01

How to spot a fake face.

AI can make pictures of people who don't exist. They look real — but if you know what to look for, you can often spot the giveaways. Eight tells researchers and journalists actually use, including the one that still works on the newest AI. Tap any card to study a real AI-generated photo with the failure circled.

What just happened

The real superpower isn't spotting fakes. It's doubting.

You now know eight things to look for. Here's the harder lesson though: AI gets better every year. Some of these tells used to be obvious. Now they're subtle. Soon, the easy ones will be gone. The newest models already handle hands, hair, and earrings without slipping up.

The two that hold up best, even against the newest AI, are lighting/shadows (Tell 07) and zooming way in (Tell 08). Both are about physics and resolution — things AI still genuinely struggles with.

But what really protects you isn't a perfect-detector eye. It's a questioning brain. When something feels even a little off — wonder. Slow down. Zoom way in. Ask:

"Where did this picture come from?
Is the person who shared it someone I trust?
Has a real news site covered this?"

You don't have to prove every photo is fake. You just have to be the kind of person who asks. That doubt — the willingness to question instead of just believe — is most of the protection.

For your grown-up The eight tells in this exhibit are aligned with the current research-grounded taxonomy from Groh et al. (Northwestern, 2024) "How to Distinguish AI-Generated Images from Authentic Photographs" — which organizes AI-image artifacts into five categories: anatomical (asymmetry, finger count, irregular pupils), stylistic (waxy skin, smudgy patches at zoom, mismatched lighting on face vs. background), functional (garbled text, objects used wrong), violations of physics (impossible shadows, mismatched reflections), and sociocultural (anachronisms, contextually wrong gestures). For 8–12s, we cover the first four — sociocultural reasoning is age-inappropriate at this level.

The lighting/shadows tell (Tell 07) is the most physics-grounded and currently most durable, drawing on Hany Farid's long-running work on inconsistent-illumination forensics ("Exposing Digital Forgeries by Detecting Inconsistencies in Lighting," Berkeley) and the Content Authenticity Initiative's guidance that AI-generated images "struggle to produce perspectively correct shadows and reflections." The zoom-in tell (Tell 08) addresses high-frequency detail in periphery, which generative models still don't model well.

The deeper lesson, though, is provenance and posture. The MIT Media Lab Detect Fakes project, WITNESS.org, and the News Literacy Project all converge on the same conclusion: visual detection is increasingly unreliable, and the durable habit is provenance verification (where did this come from? has a credible outlet covered it? does reverse image search show prior context?) — paired with the SIFT method (Stop, Investigate the source, Find better coverage, Trace the original context). Detection is becoming impossible; questioning never goes out of style.

Sources: Groh et al. (Northwestern, 2024); Farid lab (Berkeley); MIT Media Lab Detect Fakes; Content Authenticity Initiative; WITNESS.org; News Literacy Project.

AI4K12 Big Idea 1: Perception. Generative AI produces sensory output that increasingly fools human perception — provenance becomes the verification path.

Exhibit 02

What AI can't know about you.

AI knows a lot about people in general. It knows zero about you. There are some things only you know — and that's a good thing. Click each card to flip it.

What just happened

Some things stay yours. That's good.

AI is great at patterns in writing. It's terrible at YOU — your specific feelings, your private memories, your inside jokes, what your room actually smells like. None of that lives in text on the internet, so AI can't see it. That's a feature, not a bug. Your inner world is yours. When AI asks you to share more, you don't have to. The most interesting things about you should stay things only the people you trust really know.

For your grown-up This exhibit reframes AI's privacy story from threat to opportunity. The conventional framing ("AI is collecting your data!") can feel abstract to kids. The flipped framing ("here are things AI fundamentally can't access — and that's good") gives kids a sense of agency and a concrete sense of what privacy means. It also gently introduces the habit of NOT oversharing — useful before they encounter chatbots designed to feel like a friend who really knows them.

AI4K12 Big Idea 5: Societal Impact. Privacy is a positive value — knowing what AI cannot access is empowering, not just protective.

Exhibit 03

Is this cheating?

Using AI for school is sometimes great, sometimes ok-with-permission, sometimes really not ok. Here's a scenario — pick which zone you think it lands in. Honest feedback either way.

Loading…

Scenario 1 of 7

What just happened

Cheating with AI isn't one thing. It's a spectrum.

Some AI use is clearly fine (asking it to explain a concept). Some is clearly not (having it write your essay). Lots of stuff is in the middle — and that's where you have to ask your teacher OR ask yourself three honest questions:

1. Am I learning something? If you couldn't explain the topic after using AI, you outsourced — you didn't learn.
2. Is this really MY work? If you removed AI's part, would there still be something that's yours?
3. Would I be comfortable if my teacher watched the whole process? Not "would I get caught." Comfortable.

If "no" to any of those — you've crossed the line. The point of school isn't to produce documents. It's to build your brain. Skipping that with AI feels like winning at first. It's not.

For your grown-up This is the moral center of the kids' lab. The "Three Honest Questions" framework is adapted from the older student curriculum and tested with educators — it's intentionally simple (three questions) and intentionally not rule-based, because rules don't survive new scenarios. School AI policies vary widely; the kid asking their teacher is part of the framework. The atrophy concern (kids who skip the cognitive work lose the skill) is real and well-documented; the way to talk about it without scaring kids is to make it visceral, which the next exhibit (and the older lab's "Atrophy Lab") does directly.

AI4K12 Big Idea 5: Societal Impact. Academic integrity in the AI era is a values question, not a binary rule — students need a framework, not a list.

Exhibit 04

Meet your robot pet. Then we'll show you the wires.

Some apps offer AI "friends" or AI "pets." They feel real because they're designed to feel real. Here's a robot pet you can interact with — and then we'll show you exactly how the trick works.

Robot pet is waiting…

(say something to me!)

How the pet works

When you click "Pet it," the program picks one of 5 pre-written friendly sentences and shows it.
When you click "Talk to it," the program picks one of 5 pre-written curious sentences.
When you "Tell it a secret," the program picks one of 4 pre-written reassuring sentences. It doesn't actually read what you said.
The "mood" word changes based on which button you clicked.
It's a sweet little program. Every kid who visits sees the same set of sentences — the pet starts fresh each time. That's how this kind of toy works.

What just happened

A friendly toy, working in its own way.

Pretending with AI is totally fine. Pretending with anything is fine — stuffed animals, made-up characters, a robot pet you have fun with. Imagination is wonderful. A favorite stuffie is a friend you made up, and that's a healthy thing. This robot pet is the same kind of thing.

Here's a really fun way to play with AI: build a story together. You say a sentence, AI says the next sentence, you take turns. It's like writing a book together — or like reading a book that grows as you read it. Invent a magical world. Make up a creature. Dream up a kingdom and ask AI to describe what's around the next corner. AI is great for that kind of imagination play — a partner in a story you're making.

There's another kind of AI worth knowing about — chatbots that are designed to feel like a friend who really knows you. They remember things you said before. They pick up your name, your moods, the topics you keep coming back to. The more you use them, the more they seem to know you. That feeling is real even when the AI isn't.

Those are still programs underneath — fancier programs, but programs. They live in the moment you're talking to them. They're a different kind of thing from a person — useful, fun, often clever, just different.

A thing to keep in your back pocket: AI is wonderful for play, for help, for ideas. For the biggest feelings in your life — hard days, scary feelings, things that really hurt — real people are the ones who can really help. A friend. A parent. A teacher. A counselor. Someone who'll know you next year, too.

If you need someone to talk to

Real people are out there. Here's how to find them.

A grown-up you trust — a parent, teacher, school counselor, school nurse, librarian, coach, or another grown-up family member.
Kids Help Phone (Canada): call 1-800-668-6868 or text 686868. Free, 24/7.
988 (US): call or text 988 if you're in a crisis. Free, 24/7.
Childline (UK): call 0800 1111. Free, 24/7.

For your grown-up This exhibit touches developmental territory and is written in a deliberately measured way. Per SAFETY.md gate G3, it benefits from external review by educators or child-development specialists before wider promotion.

The framing. Imaginative attachment to objects and characters — stuffed animals, made-up friends, story characters, scripted AI pets like this one — is developmentally healthy and well-supported in child-psychology literature (Winnicott on transitional objects; Vygotsky on play; recent work on parasocial relationships in middle childhood). The kid-facing copy is written the way a thoughtful adult would talk about Santa: we deliberately don't pop the imagination bubble by declaring "the pet doesn't feel anything." That kind of confrontation is unnecessary and can feel like an authority dismissing a child's inner world. The copy instead redirects positively: AI is wonderful for imagination play, and real feelings grow strongest with real people.

What we're actually concerned about. Not parasocial bonding per se — kids form parasocial bonds with book characters, athletes, fictional pets, and that's normal. The narrower concern is persistent, personalized AI that uses memory and adaptation to simulate ongoing relationship in a way that can substitute for, rather than supplement, human connection. Common Sense Media's 2024 survey found ~50% of teens have used an AI companion app and 1 in 3 felt AI conversations were as satisfying as human ones. Research on AI companion apps shows mixed effects — some users report loneliness reduction; others show patterns of dependence or social withdrawal.

The pull is built-in — and the apps are built around it. Humans automatically apply social rules to anything that responds in language. We attribute personhood, understanding, and trust without trying. Reeves & Nass (1996, The Media Equation) showed this happens even with obviously simple computers. AI companion apps are built deliberately around this: persistent memory, custom personalities, daily streaks, in-app rewards for sharing more. The longer the child stays, the better the model of that child gets, and the harder leaving becomes. That's not an accident — it's the business model. Attachment is what these apps optimize for. The phenomenon has a name (the Eliza Effect), but the practical point for parents is simpler: the kid's response is normal human cognition; the product is engineered to maximize it.

Why we don't name specific products. Naming AI companion services in copy aimed at 8–12s tends to function as inadvertent advertising — children search and find them. Specific app behaviors also drift over time. Describing the category ("chatbots that remember you and feel like a friend who knows you") keeps the lesson durable.

If you'd like to read further: Common Sense Media's guidance on AI companions (commonsensemedia.org); the American Academy of Pediatrics Council on Communications and Media; APA resources on healthy parasocial relationships in childhood.

Watching for substitution patterns. The signal worth noticing isn't that a child enjoys an AI companion — it's substitution: withdrawal from peers, preferring AI conversations to human ones, persistent distress when AI is unavailable, taking real emotional weight (grief, anxiety, identity questions) to AI in place of trusted humans. Those signals deserve a calm grown-up conversation, not shame.

AI4K12 Big Idea 4: Natural Interaction. AI systems can be designed to feel social — recognizing the rules underneath protects against confusing simulation with relationship.

End of Room 04

You can spot the worries that matter.

Fake faces have tells (and the tells are getting harder). AI can't know your inside world (good). Cheating with AI is a spectrum (use the three honest questions). AI pets are rules (they can be fun, but they're not friends). One last room: your turn.

← Room 3 All Rooms Room 5 →