Train Your First AI Model with Teachable Machine (No Coding Needed)

Can you teach a computer to see? In this activity you will train a real machine learning model that recognises hand shapes for rock, paper, scissors — in about 20 minutes, with no coding at all.

You will use Teachable Machine, a free tool from Google that runs in your web browser. More importantly, you will discover how AI learns, and why it sometimes gets things wrong.

What you need: a laptop or desktop computer with a webcam, and the Chrome or Edge browser. Go to teachablemachine.withgoogle.com. No account is needed.

What is a machine learning model?

Normal programs follow rules a person writes: "if the score is 10, show 'You win'." But try writing rules for "this is a picture of a hand making scissors". What exactly makes scissors look like scissors? It's surprisingly hard to put into words.

Machine learning flips the idea around. Instead of writing rules, we give the computer lots of examples with labels (this is rock, this is paper…) and it finds the patterns itself. The result is called a model. Once trained, the model can make a guess — a prediction — about examples it has never seen.

You will do exactly that in three stages: collect examples → train → test.

Step 1: Start an image project

Open teachablemachine.withgoogle.com and click Get Started.
Choose Image Project, then Standard image model.

You will see a row of boxes on the left called classes. A class is one category you want the model to recognise. There are two to start with.

Step 2: Create your classes

Click the pencil icon next to Class 1 and rename it Rock.
Rename Class 2 to Paper.
Click Add a class and name it Scissors.
Add one more class called Nothing. This teaches the model what it looks like when no hand shape is shown — otherwise it is forced to choose rock, paper or scissors even when you are just sitting there.

Step 3: Collect training examples

For each class:

Click Webcam and allow the browser to use your camera.
Make the hand shape in front of the camera.
Press and hold the Hold to Record button. It captures many pictures per second.
While recording, move your hand around: closer, further away, a little left and right, slightly turned. Aim for 150–300 images per class.

For the Nothing class, record your face and background with no hand showing.

Why move around? If every "rock" picture shows your fist in exactly the same spot, the model might learn "rock = something in the middle of the screen" instead of learning the shape. Variety helps the model learn what actually matters. This is one of the most important lessons in AI: a model is only as good as its training data.

Step 4: Train the model

Click Train Model and wait. Keep the browser tab open while it trains.

During training, the computer looks at every picture over and over and adjusts a huge number of tiny settings inside the model until it is good at telling the classes apart. Each full pass through the pictures is called an epoch.

Step 5: Test it

When training finishes, the Preview panel turns on your webcam. Make each hand shape and watch the bars:

The bars show the model's confidence for each class — how sure it is, from 0% to 100%.
Try all three shapes. Then try something tricky: a half-open hand, two hands, a different background, or a friend's hand.

Step 6: Investigate the mistakes

Your model will probably make some mistakes. That is not a failure — it is where the real learning happens. Try these experiments:

Experiment	What you might notice	Why
Show your hand in a darker room	Confidence drops or guesses go wrong	Training pictures were all taken in bright light
Let a friend try	It may work less well for them	The model only saw your hand
Hold your hand much closer	Mistakes increase	Few training pictures were taken that close
Remove the "Nothing" class and retrain	It always picks something, even with no hand	It has no way to say "none of these"

How to fix it: add more varied examples (different light, distances, backgrounds, and other people's hands — with their permission) and train again. This loop of test → find weaknesses → improve the data → retrain is exactly what AI engineers do every day.

Step 7: Make a game with your model

Want to go further? Teachable Machine can export your model:

Click Export Model to see the options. You can upload it to get a shareable link, or download it to use in your own code (for example, in a JavaScript website).
Some free tools for young coders, such as PictoBlox and ML for Kids, let you use AI models inside Scratch-style block coding — so you could build a game where the computer plays rock, paper, scissors against your webcam hand.

Think about it: AI and fairness

Imagine a company trained a face-recognition model using photos of only one group of people. It might work brilliantly for them and badly for everyone else. That is called bias, and it happens when training data does not represent everyone who will use the system.

Your rock, paper, scissors model shows the same thing on a small scale: it works best for the hand it was trained on. Good AI builders always ask:

Who is missing from my data?
Who could be treated unfairly if my model is wrong?

Is my data safe?

According to Google's Teachable Machine site, the training happens in your web browser, and your pictures are not sent to Google's servers unless you choose to save the project to Google Drive or upload the model. Even so:

Ask an adult before using a webcam for any online tool.
Do not upload models that contain pictures of other people without their permission.

What you learned

What a machine learning model is: something that learns patterns from labelled examples.
The three stages of machine learning: collect data → train → test.
Why varied, balanced training data matters.
What confidence means, and why models make mistakes.
What bias is and why fairness matters in AI.

To understand the ideas behind this activity in more depth, read What is machine learning? Explained for kids and parents.

Frequently asked questions

Is Teachable Machine free? Yes. It is free to use in a web browser, and no account is needed.

Can I train it on sounds or poses? Yes. Teachable Machine also offers Audio Projects (for example, recognising claps and whistles) and Pose Projects (recognising body positions).

Does this count as "real" AI? Yes. Under the hood it uses a neural network — the same family of technology used in many professional image-recognition systems, just on a smaller scale.

What age is this activity suitable for? Children from about 8 can do it with an adult nearby; older children and teenagers can explore the export options and build projects with their models.