Unplugged AI activities for kids

The best way for young children to understand AI is to act it out. These activities need no screen and no special equipment — just toys, paper and some floor space — and each takes about 20 minutes. Every one teaches a real idea behind AI, like following exact instructions, learning from examples or making predictions, and ends with a short game to play together afterwards.

Program the Teacher / Parent (20 minutes)

Understand that computers cannot think for themselves; they must follow exact, unambiguous step-by-step instructions called algorithms.

You need: Toy block or stuffed animal, A blindfold or closed eyes, Floor space.

  1. Choose one person to be the "Robot" (who must freeze and only move when commanded).
  2. The children are the "Programmers". Their goal is to direct the robot to walk across the room and pick up a teddy bear.
  3. The programmers can only use 4 commands: "Step forward 1 foot", "Turn 90° right", "Turn 90° left", and "Reach hand down".
  4. If the programmers say "Go get the bear!", the robot tilts its head and says: "ERROR: Command too vague! What is the first step?"
  5. Celebrate when the sequence succeeds! Discuss why computers need exact steps.
Then play Agent Miles’s Mission →

Finish My Sentence! (20 minutes)

Demystify Generative AI and LLMs: discover that language models do not understand meaning, but predict the most likely next word using patterns.

You need: Whiteboard or paper, Familiar sentence strips or nursery rhymes.

  1. Gather students in a circle. Write an incomplete sentence on the board: "I ate a slice of cheesy warm ______."
  2. Ask students to shout out what word comes next (Pizza! Bread! Pie!). Ask: How did your brain know it wasn’t "Bicycle"?
  3. Write another: "The cat chased the little gray ______." (Mouse!).
  4. Explain: A computer does this exact same game, but it counted millions of books to know that "pizza" is 95% likely after "cheesy warm".
  5. Try a silly one: "The spaceship landed on a giant scoop of ______." (Ice cream!). Notice how humans can imagine brand new things!
Then play The Guessing Game & Word Chopper →

Warm & Cold Feedback Game (20 minutes)

Understand that AI learns from examples and feedback (Reinforcement Learning) rather than pre-written rules.

You need: Stickers or paper star tokens, A pretend "puppy" or bell.

  1. One child is the "Learning Puppy" and steps outside the classroom for 10 seconds.
  2. The class chooses a secret goal (e.g., touch the yellow chair in the corner).
  3. The puppy returns. Whenever they take a step closer to the yellow chair, the class claps enthusiastically (positive reward / "treat!").
  4. If they step away, the class stops clapping and stays quiet.
  5. Within 30 seconds, the puppy finds the yellow chair through feedback alone, without anyone speaking words!
  6. Discuss: This is how AI robots learn to walk and play chess—through trials and reward stars!
Then play Puppy Trainer (Reinforcement Learning) →

The Mysterious Pet Sorter (25 minutes)

Discover that an AI can only recognize what it has seen before. If the training data is missing something, the robot makes mistakes.

You need: Toy animals of different colors, Drawing paper and crayons.

  1. Show children 5 pictures of cats, but every single picture is a black cat.
  2. Tell the class: "You are a robot trained on these 5 photos. You have learned that cats have pointy ears, whiskers, and black fur."
  3. Now hold up a picture of a ginger/orange tabby cat or a white cat. Ask the robot: "Is this a cat?"
  4. The robot will say: "Error! It is orange, so it cannot be a cat!"
  5. Ask the children: What went wrong? Did the robot break, or was the photo album missing orange cats?
  6. Teach the big idea: To make AI fair and smart, humans must give it examples of all colors, shapes, and kinds of people and things.
Then play Pixel Detective & The Orange Cat Problem →

Describe It, Draw It (20 minutes)

Discover that an AI answers the words you actually gave it, not the idea in your head — so clearer, more specific asks produce better results.

You need: Paper and pencils for every child, A simple picture the class has not seen, A folder or book to hide it behind.

  1. Pair the children up and sit them back to back. One is the Describer, one is the Drawer.
  2. Give the Describer a simple picture — a house with two windows, a tree on the left, three clouds — and hide it from the Drawer.
  3. The Describer may only use words. No pointing, no peeking, no showing.
  4. After two minutes, compare the drawing to the original. The gap between them is always funny, and always instructive.
  5. Now swap roles and run it again — but first, let the class agree on three rules for describing well (say where things are, say how big, say how many).
  6. Watch the second round come out dramatically better. Name what changed: not the drawer, the instructions.
Then play Prompt Power: The Magic Words →

Fact or Fib & the Three-Sign Sort (25 minutes)

Build one automatic, rehearsed routine for the moment an AI says something wrong, strange, or upsetting — and learn that a confident voice is not the same as a correct answer.

You need: Three large signs: STOP ✋, CHECK 🔍, ASK 🧑, Sticky notes, A postcard and a sealed envelope, True and silly statements written on cards.

  1. Read a few simple claims — some true, some silly — in a flat, completely certain voice. Ask: did I sound sure? (Very!) Does sounding sure make it true? (No!) Sort the cards into TRUE or MADE-UP and ask how the children knew.
  2. Hold up a postcard and a sealed envelope. Ask: which one can the postman read? Establish that typing to a robot is a postcard, not an envelope.
  3. Together, sort sticky notes into KEEP PRIVATE or FINE TO TYPE: my favourite dinosaur, my home address, a maths question, my school's name, a joke I made up, my password.
  4. Land the rule in the children's own words: names, places and passwords stay with us.
  5. Put the three signs — STOP, CHECK, ASK — on the wall. Read a scenario aloud and have the class point at the sign they would use first.
  6. Scenarios: the robot says something you know is wrong; the robot asks where you live; the robot says something that makes your tummy feel funny; the robot gives you an answer for homework.
  7. Finish with the line, said together: "The robot is a tool. A grown-up is a person. When something feels wrong, I go to the person."
Then play Robot Says… Is That True? →

Our Meaning Neighbourhood (20 minutes)

Understand that an AI does not store words like a dictionary — it stores them by meaning, so "dog" and "puppy" live closer together than "dog" and "rocket".

You need: Blank paper with one word drawn in the middle, Pencils or crayons, A small set of toy animals or food pictures.

  1. Draw a big house in the middle of a page and write a common word inside it — DOG, CAKE or STAR.
  2. Ask the class: who are DOG’s neighbours — the words that mean almost the same thing? Collect answers: puppy, cat, bunny, pet, fur.
  3. Add one neighbour per house drawn next to the first. The closer the meaning, the closer the house.
  4. Then add far-away words — rocket, dinner, Monday — and ask the class to draw them across the street or out of town.
  5. Repeat with CAKE (cookie, candy, honey) and STAR (moon, rocket, sun). Praise any defended answer: the point is the shared idea.
Then play The Robot’s Word Map →

Human 20 Questions, Robo-Style (25 minutes)

Discover that classification can be done faster (and taught to a robot) by asking yes/no questions that slice the possibilities in half.

You need: A paper copy of the Guess-It Tree worksheet, Animal flashcards or the zoo box from the sheet, No batteries!.

  1. Choose a child to be "IT" who secretly picks an animal from the zoo box.
  2. Instead of guessing animals one by one, the class can only ask questions that split the list roughly in half — "does it live in water?", never "is it a fox?".
  3. Run a round using the printed tree as the question list. Count how many questions it takes to find the animal.
  4. Now try the slow way: guess one animal at a time with no questions. Count how many guesses it would take to be sure.
  5. Compare the two counts and land the big idea: every good question threw half the animals away.
Then play The Guess-It Machine →

The Whisper Chain & The Phone (20 minutes)

Understand how a robot "hears": sound becomes numbers, is chopped into tiny pieces, and is matched against sounds the robot has practised — and that the same trick is how the site talks back.

You need: A cardboard "phone" or a pair of paper cups on a string, No special tech required.

  1. Play a classic whisper chain with one sentence — "Robot ears love wiggly air!" — and enjoy how garbled the end is.
  2. Then play the paper-cup phone: one child whispers a word into the cup, a partner guesses it across the string.
  3. Ask: what travelled from one cup to the other? Air wiggled! A microphone does the same, but turns the wiggle into numbers.
  4. Have a child draw the shake-lines (the "wiggle picture") of a loud word and a quiet word, then a loud quiet word.
  5. Finish: robots chop your wiggly air into tiny pieces, check each piece against sounds they have practised, and say the word they think fits.
Then play How the Robot Hears →

The Three-Hoop Sort (20 minutes)

Make the privacy pause automatic: some things are fine to share, some only with a grown-up, and names, places, photos and passwords are never typed to a robot, anywhere.

You need: The Safe to Share? worksheet, Scissors and glue or sticky notes, Three coloured hoops or circles drawn on the floor.

  1. Draw three big circles on the floor: 🟢 Safe to share, 🟡 Ask a grown-up, 🔴 Never tell a robot.
  2. Hand each child one card from the worksheet (or a sticky note) and read it aloud.
  3. One child at a time, the child decides which hoop the card belongs in and steps in with it.
  4. The class gets one gentle challenge after each place: "Could a stranger use that to find them? Is it someone else's secret?"
  5. End in a circle and repeat the rule together: names, places and passwords stay with us.
  6. Give the class the winning test: "Would I be happy if the whole playground read this card?"
Then play Safe or Not Safe? →

The Human Attention String Game (25 minutes)

Demystify the Transformer architecture by demonstrating how AI uses self-attention to process whole sentences simultaneously and connect related words.

You need: A ball of colorful yarn or string, Word cards pinned to student shirts (Elephant, Sprayed, Car, Because, It, Dirty), Flashlight / spotlight.

  1. Assign 6 students word cards to form the sentence: "The elephant sprayed the car because it was dirty."
  2. Give the child wearing "It" the ball of yarn. Ask: "Who or what is dirty?"
  3. "It" unrolls the yarn and tosses the ball directly to "Car" (holding the tight string), and then to "Dirty".
  4. Point out: the string connects "It" to "Car" across the room instantly, without touching the words in between!
  5. Contrast this with the "Snail Walk": have another student slowly step from word to word, demonstrating how older models processed text sequentially.
  6. Repeat with a new sentence: "Maya gave Leo a cupcake because she was generous."
Then play The Attention Spotlight →

The Classroom Robot Programmer (30 minutes)

Understand that AI is software written by humans through triggers (WHEN), conditional checks (IF), and actions (THEN).

You need: Large index cards labeled WHEN (Triggers), IF (Conditions), THEN (Actions), Robot headband or foil hat for the designated "robot" student, Whiteboard and markers.

  1. One student or the teacher wears the Robot Headband and stands frozen like a powered-down robot.
  2. Students work in pairs to assemble three recipe cards: WHEN (e.g. Someone claps twice), IF (e.g. Robot has hands free), THEN (e.g. Spin around and beep).
  3. The "Programmer" student holds up the code card. The class executes the trigger (clapping twice).
  4. The "Robot" checks the condition and performs the exact action!
  5. Introduce a "bug": what if the trigger is missing? The robot stays frozen! Show that computers need explicit instructions.
  6. Reinforce: robots are not magic; people write the code recipes that make them work.
Then play Code the Robot Brain →

Superhero Move vs Risky Move Roleplay (25 minutes)

Instill four protective digital habits: keeping private data secure, recognizing AI hallucinations, maintaining academic honesty, and reporting distressing outputs.

You need: Printable AI Safety Superhero Pledge sheets, Superhero badges or stickers, Chart paper with the 4 Golden Rules.

  1. Introduce the four superhero tools: The Privacy Vault (Shield), Truth Goggles (Magnifier), Brain Captain Hat (Compass), and Red Alert Bell.
  2. Read four classroom dilemma cards aloud (e.g., a chatbot asking for a password, an AI writing fake dinosaur facts, copying a poem for homework, seeing a scary response).
  3. Students strike a Superhero Pose with their arms if the move is safe, or cross their arms in an "X" if it is risky!
  4. Discuss the safest solution for each card and have students practice the sentence: "I will check with my teacher first!"
  5. Students sign their Superhero Pledge certificates and commit to the 4 Golden Rules.
Then play The AI Safety Superhero →

Teaching a class? The full lesson plans, with reflection questions and curriculum links, are on the parents and teachers page.

Open the parents and teachers page →

More guides: How to explain AI to kids · AI safety for kids: rules for your family