Build a Teachable Machine Image Classifier With Your Kid

Introduce your child to the fundamentals of machine learning by building a custom visual AI model in under thirty minutes.

Most children today are passive consumers of artificial intelligence. They interact with recommendation algorithms on video platforms, use voice assistants to play music, and ask chatbots to help with homework. However, understanding how these systems actually work is the key to moving from a passive consumer to an active creator. By building a custom image classifier, children aged eight to sixteen can demystify the technology and gain an intuitive grasp of machine learning. This guide details a step-by-step project using Google's free Teachable Machine tool, requiring no coding experience, no paid subscriptions, and no specialized hardware.

Step 1: Choosing a Project and Setting Up

To begin, gather your child and a computer equipped with a standard webcam. Open a web browser and navigate directly to the Google Teachable Machine website at https://teachablemachine.withgoogle.com/ . Click on the "Get Started" button and select "Image Project," followed by "Standard Image Model."

Before capturing data, decide on a fun, visual classification project. Excellent options for beginners include a "Rock, Paper, Scissors" game, a classifier that distinguishes between different toy cars (such as Lego versus Hot Wheels), or a model that recognizes different family members or pets. For this guide, we will use a classic "Rock, Paper, Scissors" classifier as our primary example. This project is highly interactive and provides immediate, visual feedback that keeps children engaged throughout the process.

Step 2: Gathering and Labeling Training Data

On the screen, you will see two default classes labeled "Class 1" and "Class 2." Rename these classes to match your project. For our example, rename "Class 1" to "Rock" and "Class 2" to "Paper." Click the "Add a class" button to create a third class and name it "Scissors."

Now, it is time to gather training data. Click the "Webcam" button under the "Rock" class. Instruct your child to hold their hand in a fist (the rock gesture) in front of the webcam. Click and hold the "Hold to Record" button to capture approximately one hundred images. Encourage your child to slowly rotate their hand, move it closer and further from the camera, and change the angle. This variation is crucial because it teaches the computer to recognize the gesture under different conditions, rather than just memorizing a single static image.

Repeat this exact process for the "Paper" class (flat hand) and the "Scissors" class (two fingers extended). Ensure that you capture a similar number of images for each class to prevent the model from developing a bias toward one specific gesture.

Step 3: Training the Model

With the training data gathered, click the "Train Model" button in the middle column. This process takes place entirely within the web browser, meaning no data is sent to external servers, ensuring complete privacy.

During training, tell your child to watch the progress bar. Explain that the computer is currently looking at all the images in each class and finding the patterns that make a "Rock" different from a "Paper" or a "Scissors." It is analyzing edges, colors, shapes, and shadows to build a mathematical representation of each gesture. This step usually takes less than a minute. It is important to keep the browser tab active and open during this process to avoid interrupting the training cycle.

Step 4: Testing and Iterating

Once training is complete, the "Preview" panel on the right side of the screen will activate, showing a live feed from the webcam. Have your child make one of the hand gestures in front of the camera and watch the output bars at the bottom of the panel. The bars will show a percentage representing how confident the model is that it sees a Rock, Paper, or Scissors.

This is where the real learning happens. Challenge your child to "break" the model. What happens if they use their left hand instead of their right hand? What happens if they stand further back, or if another family member steps into the frame?

If the model misclassifies a gesture, explain that this is not a failure, but a normal part of the machine learning lifecycle. It simply means the model needs more diverse training data. Have your child return to the training columns, record additional images using their left hand or from different distances, and click "Train Model" again. This iterative feedback loop perfectly mirrors how professional AI engineers develop and refine enterprise-grade models.

Parent & Educator Sidebar

Core AI Concept: Supervised Learning This project demonstrates supervised learning, where an AI model is trained on labeled data (images that we have explicitly named "Rock," "Paper," or "Scissors"). The model learns by finding statistical patterns in the training data and then applies those patterns to classify new, unseen images.

Conversation Starters:

• "How do you think a self-driving car uses a system like this to recognize stop signs versus green lights?"
• "If we only trained our model with pictures of your hand, why might it struggle to recognize my hand?"
• "How could a recycling plant use an image classifier to sort plastic, paper, and metal automatically?"

Action Steps Summary

• Access: Go to https://teachablemachine.withgoogle.com/ on a computer with a webcam.
• Define: Create and label your classes based on your chosen project (e.g., Rock, Paper, Scissors).
• Capture: Record at least one hundred diverse webcam images for each class, changing angles and distances.
• Train: Click "Train Model" and keep the browser tab active until the process completes.
• Test: Use the live preview to test the model, identify errors, and retrain with new data to improve accuracy.

Want more engaging, family-friendly AI projects to build together?