AIQ
AIQ
Part of the When AI Isn't Fair lesson guide. Teaching a different grade? 🔧 Builder (8–10) · 💻 Hacker (11–14) · ⚡ Architect (15–18)
Gather the class in a circle on the floor, the way you would for story time. Bring Pixel along if you have a printed picture or plush — this lesson is Pixel's, and little ones respond well to hearing it as something happening to their mascot friend.
Keep your voice soft and a little concerned here — this is not a scary lesson, it's a caring one, and the tone should match Pixel's gentle, "let's figure this out together" personality rather than sounding like a warning.
Say, slowly and with a worried little frown, the way Pixel would say it:
"Sometimes A.I. isn't FAIR. 😟 If you only teach A.I. about some people, it won't know about everyone!"
Then play a very short, very physical game before you explain anything. Ask five or six children to stand up. Say: "I'm building a robot team, but I'm only allowed to pick TALL friends." Walk along the line and only tap the tallest two or three children. Have them step forward. Then stop and ask the ones left behind: "How does it feel to be left out, even though you didn't do anything wrong?" Let two or three children answer. Then ask the class: "Was that a fair way to pick a team?" They will say no immediately — that's the reaction you want.
Bring it back to Pixel: "That's exactly what happened to a robot helper. It only ever saw tall people, so when it had to pick a team, it didn't know how to pick short people fairly. Today we're going to find out why that happens, and how we can teach robot helpers to be fair to EVERYONE."
Before moving on, do a quick "thumbs" check-in: ask the class to show a thumbs up if they think it's ever okay to leave someone out just because they're shorter, or a thumbs down if they think that's never okay. This takes ten seconds and gives every child, even the quiet ones, a way to weigh in before the group discussion starts. Note out loud what you see: "I see almost all thumbs down — good, that's the fair way to think about it, and it's exactly the feeling we're going to use all lesson long."
Explorer mode does not need the words "training data" or "algorithm" — it needs the idea underneath them, told through things a five- to seven-year-old already understands: pictures, games, and guessing. Walk through the three parts of the lesson as three short "robot helper stories," pausing for reactions each time.
Before you start Story 1, remind the class of one house rule for this lesson: "We're going to talk about robot helpers making mistakes, but robot helpers aren't bad, and neither is anyone who built them by accident. We're just detectives figuring out what happened, so we can fix it." This matters because young children sometimes leap to "that's a bad robot" or "that's a bad person," and the lesson works much better if the tone stays curious instead of blaming.
Say: "Robot helpers learn kind of like you do — by looking at lots and lots of examples. If I only ever showed you pictures of orange cats, and then showed you a picture of a black cat, would you know it was a cat too?" Let children answer (many will excitedly say no, or "maybe not!"). Explain: "That's what happened to a face-recognizing robot. It was shown mostly pictures of light-skinned faces, so when it saw other skin colors, it got confused a lot more often. That's not fair to the people it couldn't recognize."
Then bring out a real die (or draw one on the board). Roll it a few times. Say: "This die is fair — every number, one through six, gets a fair turn to come up. A fair robot should work like this die: giving everyone a fair turn, no matter what they look like."
Briefly mention the third example in gentle, simple words: "One time, a computer helper kept saying 'no' to nice families who wanted to borrow money to buy a house — just because of which street they lived on. That's not fair either, because where you live doesn't say anything about whether you'd be a good neighbor."
Pause here for a quick "show me" moment: hold up two drawings ahead of time — one of an orange cat, one of a black cat — and ask, "if a robot only ever practiced with drawings like THIS one" (hold up the orange cat), "and I showed it THIS one" (hold up the black cat), "do you think it might get confused?" Nearly every child will say yes at this point, which tells you the idea has landed before you move on.
Say: "Robot helpers don't decide to be unfair on purpose — they're not being mean. They just learn from whatever they were shown, the same way you'd learn that all cats are orange if that's all you ever saw." Hold up a measuring tape. "This measuring tape doesn't care who's holding it — three inches is three inches for anybody. That's what we want from a fair robot: it shouldn't matter who you are."
Gently introduce the proxy idea at the simplest possible level: "Sometimes a robot doesn't even need to know your name to be unfair to you — it can guess a lot about you just from where you live or what school you go to, even if nobody told it to look at that."
Give one more everyday comparison before moving on, since repetition helps this age group: "It's the same reason it's not fair to guess someone is bad at sports just because of their school, or guess someone is good at math just because of their name. A robot can make that same kind of unfair guess by accident, and we have to teach it not to."
Say: "The best way to help a robot be fair is to show it pictures and examples of EVERYONE — all skin colors, all sizes, girls and boys, kids from the city and kids from the province." Ask the class to name as many different kinds of people/friends as they can (tall, short, wears glasses, uses a wheelchair, speaks Bisaya at home, etc.) and jot a few on the board — this becomes your "everyone list."
Finish with: "Grown-ups also check robot helpers with a checklist, kind of like checking your homework before you turn it in, to make sure it's being fair to everybody. And the best teams building robot helpers have all kinds of people in them, because different people notice different things."
If your "everyone list" on the board is short, prompt with a few more categories: "What about kids who wear glasses? Kids who use a wheelchair? Kids whose family speaks Ilocano at home instead of Tagalog? All of those belong on our list too, because a truly fair robot helper needs examples of every single one."
A useful closing line for this section, said slowly so it sticks: "A robot helper is only as fair as the pictures it's learned from. If we want it to be fair to everyone, we have to make sure everyone gets shown." Repeat this once more at the very end of the lesson — young children benefit enormously from hearing the "big idea" sentence more than once.
If a child gets one of these "not yet," resist saying "wrong" — echo Pixel's style instead: "Ooh, close! Let's think about it again together," and walk back through the relevant story before revealing the answer. This keeps the same no-failure language the app itself uses throughout Explorer mode.
Close with Pixel's reassurance: "Now that we know robot helpers learn from what they're shown, we can help them be fair by showing them EVERYONE — and that makes them better helpers for all of us, including you!"
Extension activity — "Teach the Robot" drawing page (10–15 minutes): Give each child a sheet of paper divided into a grid of 6–8 boxes, labeled "Pictures to Teach a Robot Helper to Be Fair." Have them draw a different kind of friend or family member in each box — different skin tones, hair, sizes, or things they use (glasses, wheelchair, hijab, etc.). Tape the finished pages up around the room as a class "training set" and let a few children explain one box each. This turns the abstract idea of training data into something they made with their own hands, and stretches the core 5–15 minute lesson into a fuller period.
If you have extra time, add a short closing circle: hold up two or three of the finished drawing pages and ask, "if a robot helper only ever saw THIS page, would it be ready to be fair to our WHOLE class?" Children will naturally say no, because they can see for themselves that one page isn't "everyone." Then hold up all the pages together and say, "but ALL of these pages together — that's what a fair robot needs to see." This closing beat ties the whole lesson back to the very first "tall friends" game and gives the class a visual, tangible sense of what "diverse training data" looks like, without ever needing to use that phrase.