Build a Teachable Machine Image Classifier With Your Kid

Most children interact with artificial intelligence as passive consumers, whether they are asking Siri a question, watching recommended videos, or playing games with AI opponents. This passive interaction can make AI feel like magic rather than a tool built on data and math. To prepare the next generation for an AI-driven world, we must shift their perspective from consumers to creators. The best way to demystify artificial intelligence is to build a functional model from scratch, using tools that require no programming experience.

Google's Teachable Machine is a free, web-based tool that allows anyone to train a machine learning model using a webcam. By building an image classifier, children aged 8 to 16 can gain an intuitive understanding of how computers learn to see. This project requires no paid subscriptions, no specialized hardware, and can be completed on any standard laptop or Chromebook. It provides a powerful, tangible introduction to the mechanics of computer vision and supervised learning.

The Concept: How Computers See

Before opening the tool, it is helpful to explain the core concept to your child. Humans recognize objects instantly because our brains have processed millions of images since birth. A computer, however, only sees a grid of numbers representing pixel colors. To teach a computer to recognize an object, we must show it many different examples of that object from various angles, distances, and lighting conditions. This process is called supervised learning, and the examples we provide are known as training data.

Step 1: Setting Up Your Classification Lab

To begin, gather two distinct sets of objects from around your house. Excellent choices include a Lego brick versus an action figure, a coffee mug versus a water bottle, or even teaching the model to recognize different hand gestures like a fist versus an open palm. The objects should be distinct enough for a computer to differentiate easily.

Next, open a web browser on your computer and navigate directly to the Teachable Machine website at https://teachablemachine.withgoogle.com/ . Click on the Get Started button and select Image Project, followed by Standard Image Model. This will open the training interface, which is divided into three clear columns: Input, Training, and Preview.

Step 2: Gathering Your Training Data

In the first column, you will see two default categories labeled Class 1 and Class 2. Rename these classes to match your chosen objects, for example, Lego and Action Figure. Click the Webcam button under the first class. Your browser will ask for permission to access your camera.

Instruct your child to hold the first object in front of the webcam. Click and hold the Record button to capture images. Have your child slowly rotate, tilt, and move the object closer and further. Capture 100 to 150 images. This variety teaches the model to recognize the object's core features rather than just one angle. Repeat this process for the second class with the second object.

Step 3: Training Your Custom Model

Once both classes have sufficient training data, move to the middle column labeled Training. Click the Train Model button. It is vital to keep the browser tab open and active during this process, which typically takes about ten to thirty seconds.

During training, the computer analyzes pixel patterns in the captured images. It looks for commonalities within each class and differences between them. Explain that the computer is building a mathematical formula to separate the Lego images from the action figure images. Once the progress bar completes, your model is ready.

Step 4: Testing and Breaking the Model

The Preview column will show a live webcam feed with a real-time prediction bar. Have your child hold up the first object. The prediction bar should show nearly 100% confidence for that class. Switch to the second object and observe the change.

Now comes the most educational part of the project: trying to break the model. Challenge your child to find the limits of what they built. What happens if you hold the Lego very far away? What happens if you cover half of the action figure with your hand? What happens if you hold up a completely different object, like a pen? The model will still try to classify the pen as either a Lego or an action figure because those are the only two categories it knows. This demonstrates a fundamental truth about AI: a model is only as smart as the data used to train it.

Parent and Educator Sidebar

Core AI Concept: Supervised Learning

Supervised learning is an AI training method where the computer is given labeled examples (training data) and learns to associate those inputs with the correct outputs. In this project, the webcam images were the inputs, and the class names were the labels. The model learned to map new, unseen webcam frames to those labels based on the patterns it identified during training.

Conversation Starters:

• If we trained this model with only blue Legos, do you think it would recognize a red Lego? Why or why not?
• How do you think self-driving cars use this technology to recognize stop signs versus speed limit signs?
• Why is it important to have a diverse set of pictures when training an AI that helps doctors identify illnesses?

Action Steps Summary

• Navigate to https://teachablemachine.withgoogle.com/ and start a new Standard Image Project.
• Define two distinct classes and capture 100 to 150 webcam images for each object from multiple angles.
• Click Train Model and keep the browser tab active while the system processes the data.
• Test the model in the Preview panel using your objects and observe the real-time confidence scores.
• Experiment with edge cases to discover how training data limitations affect the model's accuracy.

Empower your child to become an AI creator today.

Spend twenty minutes building a classifier together, and watch their understanding shift from magic to mechanics.