Home/ Materials/ Younger learners

Ages ~6–12 · grades K–8 · teacher-facing

Adapting for younger learners

Every session's core move — predict, change one thing, compare, name what the machine did — already works with children; most of these activities began life in a K–12 classroom. What changes is the language, the pacing, and one hard rule: children never need accounts or live AI tools. The simulations carry the whole investigation.

§ A · Five ground rules

across all sessions

Unplugged first, always

Children do the activity with their own bodies and crayons before any screen appears. The tool then confirms what they already experienced.

Simulations only

Every camp tool is account-free and generates nothing live — classroom-safe by construction. Children observe and investigate; any live generation happens on the teacher's screen, if at all.

"Guesses, not knowing"

Replace mechanism vocabulary with verbs kids can test: the computer guesses the next word, fills in the picture, forgets between frames. Never "the computer thinks" or "knows."

Name it precisely — kid version

The naming rule survives intact: not "it was weird," but "it always drew ___" or "it changed ___ when I ___." Sentence stems on the board do the work.

It's a thing people made

The recurring frame for bias and authorship: a person chose the pictures it learned from, a person decides what to do with its guess — and the kids are the people who check it.

§ B · Session 1 · Text

guessing games they already know
Next-word prediction · ages 6+

The fill-in-the-blank chorus

  • Read a sentence aloud and stop — "The dog chased the ___". Collect shouted guesses; tally them on the board. That tally is the probability distribution; older students can compare it to the Prediction Game on the teacher's screen.
  • Change one word — "The hungry dog chased the ___" — and tally again. What moved? That's the whole mechanism.
Say: "The computer plays this game for every single word — and it never finds out if it was right."
Count the Next Token · ages 8+ · math tie-in

Counting is the whole trick

  • Count the Next Token is a fractions lesson in disguise: count the times each word follows "the," divide, predict. Grades 3–5 can do the arithmetic on paper from a one-page class corpus (their own sentences).
  • ELIZA becomes a card game: one child holds "if you say mom, ask about family" rule cards and answers only by rule. The class guesses the rules — then meets the simulator that does the same.
Say: "It's not magic — it's counting. You just did everything the computer does."

§ C · Session 2 · Images

draw first, then look
The Squint Test · ages 6+

The blurry guessing game

  • The Squint Test on the projector is an instant whole-class game: start at the blockiest setting and add detail until someone names the picture. Which clue gave it away — color, shape, or where things were?
  • Diffusion as an analogy kids own: a photo "developing," or squinting at fog until a shape appears. Step the viewer slowly and have them call out the moment the subject appears.
Defaults & bias · ages 8+ · the authorship conversation

Draw a doctor — before the machine does

  • The Default Test, child-safe order: kids draw first. "Draw a doctor. Draw a family. Draw a beautiful home." Put the class's drawings up — they will differ wonderfully.
  • Then show the machine's defaults in the comparison viewer and ask the question that matters: "The class drew thirty different doctors. Why does the machine keep drawing the same one?"
  • That's the bias conversation at age eight — no lecture needed: it learned from pictures people chose, and the pictures left people out. End with the fix: "what would you tell it so your doctor shows up?"
Say: "The machine isn't mean — it copies what it was shown. People picked what to show it. People can pick better."

§ D · Session 3 · Video

it's literally the telephone game
Temporal Telephone · ages 6+ · works as-is

Telephone, with drawings

  • Temporal Telephone is already a children's game — that's the point. Paper version: each child copies only the previous child's drawing; post the chain and find where the cat became a bear. That's drift, and they discovered it.
  • Run round two with the original taped to the wall (the anchor). The chain stays truer — and they've discovered why video models need reference frames.
  • Flipbooks make the frames idea physical for the youngest: a movie is just drawings that mostly agree with each other.
Say: "The machine plays telephone with itself a hundred times a second — and it forgets just like we do."
the ethics spine, in their words
Every session · 5 minutes

Whose picture is it?

  • Three questions children can hold from age six: Who made this — you, the machine, or both? Did everyone in the picture say yes to being in it? Do you want your name on it when we hang it up?
  • Practice consent on their own work first: nothing goes on the classroom wall (or in the class share) without the maker saying yes — same rule the camp uses for adults in the Consent Protocol.
  • The Classroom Activity Builder has grade bands and a no-AI option for turning any of these into a printable lesson skeleton. Ready for real training data? Machine Learning for Kids and AI for Oceans are the classroom-safe next step.
Say: "Machines don't ask permission. People do. That's a thing people are better at."
← RelatedNo-AI Pathway