Machine Learning for Kids: a parent's guide to the free site

Machine Learning for Kids (machinelearningforkids.co.uk) is a free website, made by Dale Lane, where children collect examples, train a computer to recognise them, and then build a game with what it learned in Scratch. You can start with "Try it now" and no account, and a parent can do a first project with a child in under an hour. This guide was written by Peabody (peabody.kids), a free reading and learning site that is free through the end of 2026. Machine Learning for Kids isn't ours, and nobody paid to be listed. Everything below was checked on the site itself in October 2026.

What is Machine Learning for Kids?

The site's own summary is three steps:

1. Collect examples of things you want the computer to recognise.
2. Use the examples to train a computer to recognise them.
3. Make a game in Scratch that uses what the computer learned.

The site's About page calls it a free tool, first made available in 2017, and says it's used by thousands of schools, code clubs and families. It's built by Dale Lane using APIs from IBM Watson, and the code is open source. It's entirely web-based, with nothing to install.

A child can teach it to:

Then they use the model in Scratch, Python or EduBlocks (Python with drag-and-drop blocks). The site also has pre-trained models for projects that would be too hard to train yourself, like face detection.

Do you need an account?

No, not to start. There are three ways in:

• Try it now. No registration. The site says you can use all its features, but projects made this way are deleted after at least four hours, and you can have only one project at a time. Before it lets you in, the site runs Cloudflare's Turnstile check to make sure you're a person. Ad blockers, strict privacy settings, VPNs and school networks can stop that check; the site's own advice is to turn off the blocker for that site or try another network.
• Unmanaged class account. For "a parent, teacher or leader of a coding club" who sets it up themselves. You create a free IBM Cloud account, make API keys for the Watson services, and enter them into the site. Text projects in a class need a "Watson Assistant API key". The site has step-by-step instructions.
• Managed class account. Dale Lane sets it up for you if you email him. The site says this is only for schools and not-for-profit student coding clubs.

Children in a class don't sign themselves up. The site tells them to ask their teacher or group leader, who creates their accounts and can reset their passwords.

For one evening at home, "Try it now" is all you need.

Where does the data go?

When you create a project, the site asks where to store it: in your web browser or in the cloud. Browser storage keeps the training examples on that computer and removes the size limit. Cloud storage lets you open the project from any computer. The worksheets for Try it now choose the browser.

We couldn't confirm from the site's pages where the training itself runs for each project type, so treat anything typed into a text project as something that may leave the computer. Keep real names, addresses and other people's details out of the examples. The site's language-model page also says: if you're under 13, only use a small language model with supervision from a trusted adult, because generative AI can produce text that isn't nice or appropriate. We agree.

What is the Machine Learning for Kids age range?

The site doesn't give one. It was designed for schools and code clubs, and it sorts its worksheets by difficulty instead of age:

• Beginner: templates with most of the Scratch or Python code already made, so a child can focus on the machine learning.
• Intermediate: templates with at least one step where the child extends the code.
• Advanced: the child writes complex Scratch or Python code themselves.

Our suggestion, from how the worksheets read: a beginner worksheet suits a child who can read instructions on their own and has used Scratch a little, often around 8 or 9, with a grown-up nearby. Younger children can do it if you read the steps aloud and drive the mouse. Intermediate and advanced worksheets, and the Python and language-model projects, suit older children and teens.

The project worksheets

The worksheets are the heart of the site. Each one is a downloadable, step-by-step guide with explanations and colour screenshots, written to last a single lesson. Each ends with a working game or app that shows a real use of machine learning. Most have teacher's notes, and some come in a full version and a shorter one. There are dozens, from Make me happy and Smart Classroom (a smart assistant in Scratch that controls virtual devices) to Titanic, Phishing and Pac-Man.

To find them, go to the site and choose Worksheets. You can filter by difficulty, by what the computer recognises, and by what you make with it.

How to do one Machine Learning for Kids project at home

This follows the site's shorter "Make me happy" worksheet, a beginner text project. You'll build a Scratch face that smiles at kind messages and looks sad at mean ones. Allow 30 to 45 minutes.

1. Go to machinelearningforkids.co.uk, click Get started, then Try it now.
2. Click "+ Add a new project". Name it "make me happy", set it to recognise text, choose to store it in your web browser, and click Create.
3. Open the project and click Train.
4. Add two labels. Click "+ Add new label" and make "kind things", then "mean things". These are the buckets.
5. Fill the buckets. Click "Add example" in each and type a compliment, then an insult. Do at least six of each. The worksheet asks for the meanest insult a child can think of; at home we'd steer to silly ones ("you smell like old socks") so nobody's feelings are the example.
6. Go back to the project, click Learn & Test, then "Train new machine learning model". Training can take a few minutes.
7. Test it. Type kind and mean messages it hasn't seen. If it gets them wrong, add more examples and train again.
8. Click Make, then Scratch 3, then Open in Scratch. New blocks from your project appear at the bottom of the block list.
9. Draw the face. Delete the cat, paint a new sprite with no mouth, duplicate the costume twice, and name the three costumes "not sure", "happy" and "sad". Draw a mouth on each.
10. Add the script shown in the worksheet, using the new "recognise text" block, then click the green flag and type a message.

Then try the worksheet's extension ideas: make the character reply, swap the face for a dog that wags its tail, or use the second new block, which gives a score for how sure the model is.

The worksheet's own explanation is worth reading aloud at the end: you taught the computer by collecting examples instead of writing rules, and that's called supervised learning, because you supervised its training.

Machine learning explained for kids

Here's a version to read to a child.

Most computer programs follow rules that a person wrote down. "If the light is red, stop." But some things are too hard to write rules for. How would you write a rule for "a kind message"? There are millions of kind things to say.

So instead, we give the computer lots of examples. "This one is kind. This one is mean." The computer looks for patterns, like which words keep turning up in the kind ones. That's called training.

Then we give it a message it has never seen, and it guesses. Sometimes it's right. Sometimes it's wrong, especially if we didn't give it enough examples, or if all our examples were alike. When it's wrong, we give it more examples and it gets better.

That's machine learning: learning from examples instead of rules. Your phone uses it to guess the next word you'll type. Email uses it to spot junk. It's only as good as the examples it learned from.

Our AI explained for 7-year-olds page says the same thing in even simpler words, with a short glossary.

Machine Learning for Kids, Teachable Machine or AI for Oceans?

All three are free. They suit different moments:

• AI for Oceans is a guided one-hour activity with videos, no coding and no account. Best first step for a class or a younger child. See our AI for Oceans guide.
• Teachable Machine is the fastest way to see a model learn from your own webcam or voice, in ten minutes. See our Teachable Machine walkthrough.
• Machine Learning for Kids is where a child builds something with the model, and it has the most projects. It asks the most of a child, and of the grown-up helping.

For more blocks-based tools, see Scratch AI extensions for kids.

Where Peabody fits

Peabody isn't a machine learning course, and it doesn't teach coding. It's a reading site with stories at a child's level, some written by an AI model and checked by safety screens, and learning games where the answers are a tap. If a project like Make me happy starts a talk about how AI writes and sorts words, the free AI story generator page explains how Peabody's stories are made. Or open the library and read together.

Using Peabody in your homeschool

Machine Learning for Kids suits a homeschooled child of about 9 and up as a weekly project for three or four weeks: one beginner worksheet per session, about 45 minutes, with you nearby for the human check and the typing. On the other days, keep reading and math going on Peabody in short pieces: a story from the library at your child's level and a math lesson part from the math paths by grade, 10 to 15 minutes each. The honest limit: Peabody teaches neither coding nor machine learning, and it can't see the Scratch game your child made, so take a screenshot for your records (homeschool record keeping has ideas). The homeschool hub has the rest.

Learn more elsewhere

• Machine Learning for Kids: the site itself; start with Get started, then Try it now.
• Teachable Machine: Google's tool for training a model with your webcam or microphone, no account.
• AI for Oceans: Code.org's one-hour activity on training data and bias.
• Hour of AI activities: short AI activities for every age, each with its own account rules.

About seventy more free AI sites for families and schools are in our guide to where to learn about AI.

Questions parents ask

Is Machine Learning for Kids free?

Yes. The site describes itself as a free tool. "Try it now" needs no account. A self-managed class account needs a free IBM Cloud account and API keys, and a managed class account is offered free to schools and not-for-profit coding clubs.

What is the Machine Learning for Kids age range?

The site doesn't give an age range. It sorts worksheets into beginner, intermediate and advanced instead. A beginner worksheet suits a child who reads on their own and knows a little Scratch, often around 8 or 9 with a grown-up nearby, and the advanced and Python projects suit teens.

Do I need an account for Machine Learning for Kids?

No. "Try it now" gives access to all features without registering, but projects are deleted after at least four hours and you can keep only one at a time. An account keeps projects longer and lets a teacher manage a class.

Does Machine Learning for Kids still use IBM Watson?

As of October 2026 the site says it's built using APIs from IBM Watson, and a self-managed class needs a Watson Assistant API key for text projects. Projects can store their training examples in the web browser instead of the cloud.

How do you explain machine learning to a kid?

Say that most programs follow rules a person wrote, but machine learning learns from examples instead. You show the computer lots of labelled examples, it looks for patterns, then it guesses about new ones. It's only as good as the examples it was given.