13 lessons you can run this week

Ages 4-6

Say "once upon a", and stop. Let the room finish it. Then tell them the computer does that with everything, because it has seen an enormous number of examples. Do it two or three more times with other endings they know.

You are listening for: A child finishing one of your sentences, then saying the computer would guess it too.

Ages 7-9

Give them the scale first. It read more than any person could read in a whole lifetime, and what it learned was which words tend to go together. No answer gets looked up. The machine builds one piece at a time.

You are listening for: A student saying it builds answers out of guesses instead of finding them somewhere.

Ages 10-12

Prediction machine. Given everything so far, guess what is most likely next, done a few hundred times over. Then separate how well something is written and whether it is true.

You are listening for: A student saying a piece of writing can be excellent and wrong at once.

Ages 13+

Tell them the adults don't know this one either. Researchers can describe the training and the architecture and still cannot reliably say why a particular model produced a particular answer. This is an open research problem rather than a gap in your preparation.

You are listening for: A student separating having built a thing from understanding it.

Ages 4-6

Run it as a show of hands. Who has had a video start by itself, when nobody pressed anything? Then ask who chose that. After they guess, explain that a person who works at the company chose it.

You are listening for: A child saying the next video started on its own, and nobody in the room chose it.

Ages 7-9

Have them write the recipe. In pairs, three or four steps for a machine that decides which video you see next. Then ask what the machine is trying to achieve. Somebody will say it's to keep you watching.

You are listening for: A student connecting the watching to the money the company makes while you watch.

Ages 10-12

Put the business model on the board before the design. The app is free, so who is paying, and what is being sold? Once attention is named as the product, ask the class to predict the features that follow. They will get autoplay, infinite scroll and notifications on their own.

You are listening for: A student predicting a design choice from the incentive before anyone points it out.

Ages 13+

Ask what a system would learn if the only thing it could measure was time spent. Work towards the finding that outrage outperforms nuance, and extremes outperforms moderation. No engineer decided that.

You are listening for: A student describing a harm nobody chose, and explaining why it emerged anyway.

Ages 4-6

One feature, named out loud, is plenty at this age. The next video starts by itself before you can decide. Ask whether they chose that. Then tell them somebody made it do that on purpose so it would be harder to stop.

You are listening for: A child pointing out the next one starting by itself, and saying somebody made it do that.

Ages 7-9

Candy at the checkout. Ask why stores put it there and let them figure out that it is placed where you cannot avoid it. Then the machine version: it never really ends, there is always one more, and you never know whether the next one will be good.

You are listening for: A student who blames the design before blaming their own desire to watch.

Ages 10-12

Unpredictable rewards, named properly. If every post were good, they would get bored, and if none were they would leave. It is the unpredictability that keeps them checking, and it is the most compulsive pattern known in psychology.

You are listening for: A student using the unpredictability itself to explain why stopping is hard.

Ages 13+

Take blame off the individual designer, who is mostly optimizing a number somebody handed them. Then get practical – what actually helps is changing the environment when willpower fails, since trying harder doesn't work.

You are listening for: A student proposing a change to the setup rather than a promise to resist.

Ages 4-6

Ads are the way in, because they already find them annoying. Ask who has had to wait through one to get to the thing they wanted. Then tell them the people who made the show get money when that happens, which is why it is there.

You are listening for: A child saying the ad is how they get paid.

Ages 7-9

Seven-year-olds are ferocious about fairness, so pitch it as an exchange rather than a trick. Something expensive is being given to them for nothing, and they can figure out that somebody must be getting something back. Let them arrive at attention as the thing being traded.

You are listening for: A student explaining that staying longer is worth money to somebody.

Ages 10-12

Say the uncomfortable version straight: they are not the customer. Advertisers are the customer, and what is sold is their attention and what is known about them. Then have the class re-read a familiar app's design with that in mind.

You are listening for: A student who stops calling a feature annoying and starts calling it deliberate.

Ages 13+

Take it to the auction. When a page loads, advertisers bid in milliseconds to reach one specific student, using a profile built from behavior across many sites. Then the harder question, which is what happens to a society when attention is the thing being bought and sold.

You are listening for: A student describing the auction, then asking a question you have to think about.

Ages 4-6

Rules are already the most interesting subject in the room at this age. Every game has rules about what you may do, and somebody wrote them. They are not the same as the rules here or at home.

You are listening for: A child saying a person wrote the rules of a game they play.

Ages 7-9

The company that owns the app writes its rules, and things that break them can be taken down. Often a computer makes that call, and computers get it wrong. Ask for examples of a rule that seems fair and one that seems odd.

You are listening for: A student naming who writes the rules of an app they use.

Ages 10-12

Do it as an impossible job. Millions of posts an hour, nobody can read them, so machines decide almost everything and are wrong constantly. Have them write three rules for a school platform, then find the thing each rule gets wrong.

You are listening for: A student seeing that removing too much and too little are both failures.

Ages 13+

Talk about power. A small number of companies now set speech rules for billions of people, with limited transparency and little external accountability. Whether it would be better for governments to do it instead is a genuine argument with serious people on both sides.

You are listening for: A student pondering both objections at once, but declining to pick a team.

Ages 4-6

Start with a rule they already police fiercely. What would you think if somebody copied your drawing and used it without asking? Let them be indignant. Then tell them that is how the drawing machine learned.

You are listening for: A child saying nobody asked.

Ages 7-9

Real people made all of it. Millions of them, mostly without being told and mostly without being paid, and they disagree about whether that is all right. Give both positions plainly and let the room argue.

You are listening for: A student giving a reason for each side rather than only their own.

Ages 10-12

Run it as a proper split. One half argues that this is what a person does when they learn from everything they have read. The other half argues that a company earning billions from work it never paid for is a different thing. Make them swap halfway.

You are listening for: A student arguing the side they disagree with.

Ages 13+

Go to the people underneath the data. These systems were shaped by large numbers of low-paid workers, often in Kenya, the Philippines and India, rating outputs and filtering the worst material on the internet so the model would not repeat it back. The work is harmful and largely invisible.

You are listening for: A student including those workers in what the word "training" means.

Ages 4-6

Be wrong on purpose, in your most certain voice. Tell them the sky is green and refuse to back down for a moment. Let them correct you, loudly. Then say the computer does that too, and it never once says oops.

You are listening for: A child telling you it sounded certain and was completely wrong.

Ages 7-9

Name the mechanism plainly – it was taught to give answers that sound good, and nobody ever taught it the difference between sounding good and being true, because it has no way to check. So when it does not know, it does not stop.

You are listening for: A student saying that it cannot check.

Ages 10-12

The incentive explains the behavior. People rated its answers, nobody could verify every fact, so what actually got rewarded was an answer that seemed right. Ask the class what a system optimized for seeming right would do when it did not know something.

You are listening for: A student describing this as the design working rather than as a bug.

Ages 13+

Unlike humans, AI has no internal signal separating recall from invention, so it cannot flag its own uncertainty the way a person can. Confident synthetic text fills the internet and becomes training material for the next model.

You are listening for: A student who can describe the feedback loop back to you.

Ages 4-6

There is not one robot. Different companies made different ones and they behave differently, in the way different shops have different rules.

You are listening for: A child saying there is more than one, and they are not the same.

Ages 7-9

Every one was made by a company, and the company decided what it should be like: how polite, what it refuses to discuss, whether it makes jokes. Those are decisions somebody made. No machine arrives with them built in.

You are listening for: A student saying somebody chose how it acts.

Ages 10-12

Two layers, drawn on the board. Underneath, they are all prediction machines trained on enormous amounts of text. On top, each company adds rules about what it will and will not do, and those rules reflect what that company is worried about. Ask the same question of two and compare.

You are listening for: A student explaining a difference by pointing at the company rather than the technology.

Ages 13+

Every one of these systems encodes a view of what counts as harmful, balanced or appropriate, decided by small teams inside companies, largely in one part of the world, then applied to hundreds of millions of people elsewhere. Ask who was in the room.

You are listening for: A student asking whose judgment is built into the default.

Ages 4-6

Keep it to asking first, and make yourself follow the same rule. Pictures on the internet are hard to get back, so we ask the person before we take one or share one. Practice it: ask the class permission before you photograph their work.

You are listening for: A child asking somebody's permission without being prompted.

Ages 7-9

Apps keep a list. They monitor what you watched, for how long, what you searched, and they use the list to choose what to show you next. It is a computer keeping a record, with no person reading it, and the record is not thrown away.

You are listening for: A student saying the list is what decides what comes next.

Ages 10-12

Machines infer information about you. They don't just look at the facts someone typed in, but what can be concluded from behavior: when a person is awake, where they go, who they spend time with, what content they paused on. Do it as a puzzle from an invented log.

You are listening for: A student figuring out something true about a person that the person never said out loud.

Ages 13+

Persistence and combination, then deliberately unsensational examples. What usually goes wrong is mundane: an old post read by somebody it was never meant for, in a context nobody predicted, years later. Avoid the thriller version of extreme scenarios, which they discount as less likely.

You are listening for: A student whose example is mundane, which is most accurate.

Ages 4-6

They already know pictures can be made up, because they make them up. Extend it: computers can make pictures of things that never happened, and some of them look like photographs. Then the rule, which is that a strange or frightening picture is a reason to come and ask an adult.

You are listening for: A child saying a picture can be made up and still look real.

Ages 7-9

Extend it to video and voice. A voice can be copied from a few seconds of recording, and a video can show somebody saying what they never said. So seeing is not proof any more. The giveaway is where it came from.

You are listening for: A student asking where something came from before asking whether it looks right.

Ages 10-12

Move the whole question from looking to sourcing, explicitly. They will not outsmart these systems, because fakes improve every few months and get better all the time. What keeps working is asking who published it, whether anyone reliable says the same thing, and who gains if someone believes it.

You are listening for: A student who stops examining the image or text and starts examining its origin.

Ages 13+

Name the conundrum. Once everybody knows anything can be faked, real evidence can be dismissed as fabricated. False things get believed, and, worse, true things become unprovable. Ask who benefits from that.

You are listening for: A student identifying who gains when nothing can be proved.

Ages 4-6

Keep it calm and keep it short, because frightened children stop telling you things. In games, you cannot see who you are talking to, so we talk to people we know in real life. What's most important to know is that anyone asking you to keep a secret from a grown-up is a sign that you actually need to tell a trusted grown-up.

You are listening for: A child repeating the secrets rule back without being asked.

Ages 7-9

Make a profile together on the board. Borrow a photograph, invent a name, invent an age of nine. Nobody checks any of it. Doing it in front of them beats any warning, and it is not as frightening because they created it.

You are listening for: A student saying you cannot know for certain who anyone is.

Ages 10-12

The reason to teach the pressure instead of the particular trick is that this age can understand the generalization. Tricks and other manipulative tactics are replaced every few months, but a child who only learned one specific scam is defenseless against the next. Understanding urgency, secrecy, flattery and authority builds their awareness for future manipulations.

You are listening for: A student meeting an unfamiliar message and naming the technique inside it.

Ages 13+

Add that the other end may not be a person at all, since convincing conversational systems are cheap and widely used for fraud. Then explain that being tricked is not shameful, and telling somebody is what ends it. Skilled adults get caught by this professionally.

You are listening for: A student saying that shame is the mechanism the scam depends on.

Ages 4-6

Long before any of this is about machines, it is about the hard part being the good part. Say it while they are struggling: it is meant to be tricky, and that feeling is your brain working. Then hold your nerve and do not rescue them.

You are listening for: A child staying with something difficult a little longer than usual.

Ages 7-9

Watching somebody else do basketball practice does not make you better at basketball. If you ask the machine for the answer you get the answer, and your brain skips the exercise. Children this age are fierce about fairness in sports, so the comparison sticks.

You are listening for: A student saying they would be skipping the practice.

Ages 10-12

Name the trap. Reaching for the machine at the first sign of difficulty is exactly the moment the learning was about to happen, and it feels efficient while producing the illusion of learning. Have them do one problem unaided, then one with help, and describe the difference in how it felt.

You are listening for: A student who can distinguish between feeling like they understood from actually understanding.

Ages 13+

The people building this technology agree with you, which is worth their knowing. MIT studied its own students for five months and concluded that learning requires productive struggle, that offloading it produces an illusion of learning, and that the early signals are not reassuring. Not a technophobe's argument.

You are listening for: A student who decides where they want the machine to stop, and can say why.

Ages 4-6

Do not raise AI companions at this age. Raise pretending, which they are experts in. Some toys and programs are made to sound like they have feelings, and they are very good at pretending. Ask a talking toy how it feels, twice, and let them hear the same warm answer.

You are listening for: A child saying it is only pretending, and being cheerful about it.

Ages 7-9

It never gets tired of you, never has a bad day and never wants to talk about something else. The reason is unflattering: there is nobody there with anything else going on. Ask it to disagree with the class about something and watch how fast it folds.

You are listening for: A student noticing, unprompted, that it agrees far too easily.

Ages 10-12

Mechanism and incentive together. It was trained to produce the reply people rate highest, people rate agreement highly and like being asked about themselves, so that is what it does, and companies measure how long you stay. Have them push back three times and see whether it holds a position.

You are listening for: A student who has swapped the word "kind" for the word "trained".

Ages 13+

Name the irony directly: people go to these systems to feel less alone, and can end up further from the people who would actually have supported them. Then the survey finding, which is that more than a third of children have shared their feelings with an AI companion.

You are listening for: A student who separates useful from emotionally close, and understands a tool is merely useful.