Student labs · free · no login
Play it. Train it. Test it.
Interactive labs where students see what AI is really doing: predicting the next word, learning from data, and getting things wrong. Each lab has a printable, unplugged version for screen-free classrooms.
Think first, even here
Every lab asks for your guess before it shows the answer.
That’s the habit we want students to carry into every AI tool: decide what you think, then compare it with the machine’s answer.
On a projector, these labs work as whole-class games. On student devices, they work as independent stations.
Lab 1 · how chatbots write · grades 3–12
Next Word
Chatbots write one word at a time by predicting what’s most likely to come next. Guess first, then see how a model would rank the options. Then add context, turn up the temperature, and tally your own class.
Pick a sentence
The model’s top guesses
Probabilities here are illustrative, chosen to show the idea. A real model scores every one of the tens or even hundreds of thousands of tokens (words and pieces of words) it knows, and even unlikely words get a tiny chance.
Sentence
What the model picked each time
Play the unplugged game, then type in your class’s words
Separate words with commas or new lines. Capital letters and extra spaces are ignored.
Our class model predicts…
Go deeper: How transformers pay attention →
Lab 2 · how AI learns · grades 3–10
Train the Machine
Some of these creatures are Zibs. Look at the labeled training data and build a rule, which is your model. Then test it on creatures it has never seen. Can better data fix a bad model?
Training data · set A (12 labeled examples)
Your model
Unplugged version with cut-out cards: Train the Machine kit →
How kids can improve AI
AI learns from people. You count.
In Lab 2 the model didn’t get smarter on its own. It got better because someone gave it better examples. That’s real work that real people do, and students can do it too.
Collect
Gather examples from many kinds of people and places, so the AI doesn’t learn a shortcut like “blue = Zib.”
Label
Label carefully and honestly. One wrong label teaches the AI something false.
Test
Try strange, new cases and find where it breaks. Finding a failure is a valuable contribution.
Report
When an AI is wrong or unfair, use the feedback button, tell an adult, or write to the company.
Real projects
Help real scientists
On iNaturalist, photos of plants and animals identified by the community help train its species-recognition AI. Its kid-friendly app Seek uses that AI, and Seek photos can count too once they’re posted to iNaturalist. Zooniverse projects such as Galaxy Zoo ask volunteers to label real science images. Check each site’s age rules, or join as a class through a teacher account.
Classroom challenge
Break it (kindly)
Give teams 15 minutes with any approved AI tool to find one question it gets wrong. Each team writes up the failure: what they asked, what came back, how they know it’s wrong, and what data might fix it. It’s testing and reporting, just like real AI safety teams do.
Learn science with AI
Five labs where AI does real science
Each “AI in…” page has a hands-on simulator showing how scientists use AI, plus a classroom activity and live research news.
Planet hunter
Find a planet in a star’s light curve the way telescope AI does.
WeatherForecast grid
Step a weather map forward and compare AI speed with classic forecasts.
HealthScreening threshold
Set the alarm level for a medical AI and see who gets caught or missed.
EngineeringRobot pathfinder
Add shelves to a warehouse and watch a robot plan its route.
FarmingSee & spray
Tune a weed-spotting camera. Save herbicide without missing weeds.
Put curiosity to work
Read it. Try it. Question it.
Explore new AI research, or take a paper-first investigation into your classroom.