14 AI unplugged activities with no screen

Kids get picture cards of monkeys labeled as ones that bite or don't, then build a yes-no question tree that sorts new monkeys. It shows how a machine learns rules from labeled examples, and how it can be wrong on new ones.

1. Print the monkey cards.
2. Sort the labeled ones and look for features they share.
3. Write yes-no questions that split them.
4. Test your tree on the unlabeled cards.

Dump a bin of toys and sort them by a rule: color, size, things with wheels. Then let your child invent a secret rule for you to guess. Classifying from features is the heart of most AI.

1. Sort by one rule together.
2. Child sorts by a secret rule.
3. You guess the rule from their piles.
4. Add a tricky toy that fits two piles and talk about it.

One person is a robot that follows instructions exactly and literally. The other gives directions to make a sandwich or walk to the door. Kids learn computers do what you say, not what you mean.

1. Pick a simple task.
2. The robot follows each instruction word for word.
3. Laugh at what goes wrong.
4. Rewrite the instructions until it works.

A small board game where the 'computer' player loses pieces when it makes bad moves, and slowly stops making them. Kids watch a dumb player turn smart through wins and losses, which is reinforcement learning.

1. Print the board and game rules.
2. One player is the computer and follows the move cards.
3. Remove a move card each time the computer loses.
4. Play 10 rounds and see who wins more.

Color squares on graph paper from a list of numbers to reveal a picture. Kids discover that a computer 'sees' an image as a grid of numbers, which is how face unlock and photo search start.

1. Write a number code for a simple shape.
2. Child fills in squares from the code.
3. Swap: child codes a picture for you.

Read a sentence aloud and stop: 'Once upon a...'. Everyone guesses the next word. Talk about why 'time' is so likely. Chatbots work by predicting likely next words, at huge scale.

1. Start a familiar phrase.
2. Everyone writes their guess.
3. Count which guess was most common.
4. Try an unusual sentence where guesses split.

Two people answer written questions, one acting as a computer with a rule sheet. The judge must guess who is who. It starts a great talk about whether sounding human means thinking like one.

1. Write five questions.
2. One player answers naturally, one follows simple robot rules.
3. Judge reads both answers and guesses.
4. Ask: what gave it away?

Put objects in a bag and guess them by touch only, then by sound only. Kids learn that machines also sense the world through limited senses like cameras and microphones, and guess from partial clues.

1. Hide five objects.
2. Guess by touch, then by shaking for sound.
3. Talk about which sense was most useful.

Play twenty questions, but write down each yes-no question. Afterward, look at which questions cut the possibilities in half. That is how decision trees and early AI programs narrowed choices.

Your child is the teacher showing a stuffed animal three pictures of cats and three of dogs. Then the toy (you) has to guess a new picture, and gets it wrong if the examples were all the same color. A first lesson on training data.

1. Cut pictures from a magazine.
2. Child shows the toy a few labeled examples.
3. You, as the toy, guess using only what you were shown.
4. Talk about what more examples would fix.

Fill a bag mostly with red beads and a few blue. Kids pull a handful and predict what the whole world looks like. Then show the real mix. Learning from a lopsided sample is how AI picks up bias.

1. Fill a bag with an uneven mix.
2. Draw five and predict the rest.
3. Reveal the full bag.
4. Ask: whose pictures might be missing from an AI's training?

Write a recipe as numbered steps with no guessing allowed, then follow it exactly. Kids find the gaps. AI systems are built from precise steps and lots of data, and gaps cause errors.

1. Pick something simple like toast with jam.
2. Child writes each step.
3. Follow it literally.
4. Fix the steps together.

String beads in a pattern and ask your child what comes next. Then let them build a pattern to stump you. Spotting and continuing patterns is the simplest idea behind machine learning.

Write short cards about AI in everyday life, like a camera that only opens the door for some faces or an app suggesting videos. Sort them into fair, unfair and not sure. AI4K12 puts societal impact at the center for a reason.

1. Write eight short situations.
2. Sort them together.
3. Argue about the not-sure pile.
4. Ask: who should fix the unfair ones?