Build a Teachable Machine Image Classifier With Your Kid

PromptHacker Issue 21 • Hands-On AI Education for Families

Most children today are passive consumers of artificial intelligence. They interact with recommendation algorithms on video platforms, ask voice assistants to play music, and use generative tools to write stories. However, very few understand the underlying mechanics of how a computer learns to perceive the physical world. This gap between consumption and comprehension can make AI feel like magic rather than mathematics.

You can demystify this technology in a single weekend afternoon. By using a free, web-based tool developed by Google, you and your child can build, train, and test a custom image classifier. There is no coding required, no paid subscription needed, and no complex software to install. All you need is a computer with a standard webcam and a few household objects. This activity shifts your child from a passive user to an active creator, building foundational intuition about how machine learning actually works.

Why Hands-On AI Beats Screen Time

When children build their own technology, their relationship with it changes. Instead of viewing AI as an infallible, all-knowing entity, they begin to see it as a system that is only as good as the data provided to it. This distinction is crucial for developing digital literacy.

By training a model to recognize the difference between a favorite toy and a household object, children experience the direct relationship between input data and output accuracy. They see firsthand how a small change in lighting, angle, or background can confuse a computer, which opens the door to deeper conversations about technology and bias.

Step 1: Accessing the Platform and Choosing Your Project

To begin, open a web browser on your computer and navigate directly to Google's Teachable Machine . Click on the "Get Started" button and select "Image Project" followed by "Standard Image Model."

Before you start capturing data, help your child choose two distinct categories of objects to classify. Excellent pairings for beginners include:

• A blue pen versus a red pen.
• A LEGO brick versus an action figure.
• A hand showing a "thumbs up" versus a "thumbs down."
• An apple versus a banana.

Step 2: Gathering Training Data

On the screen, you will see two default classes labeled "Class 1" and "Class 2." Have your child rename these classes to match the chosen objects (for example, "LEGO" and "Action Figure").

Click the webcam button under the first class. Instruct your child to hold the first object in front of the camera. Click and hold the "Record" button to capture approximately one hundred to two hundred images.

Crucial Tip: Encourage your child to slowly rotate the object, move it closer and further from the lens, and tilt it in different directions while recording. This variety ensures the model learns the features of the object itself rather than just one specific angle or background. Repeat this exact process for the second class using the second object.

Step 3: Training and Testing the Model

Once both classes have sufficient images, click the "Train Model" button in the middle column. This process takes about ten to thirty seconds. Instruct your child to keep the browser tab open and active during this phase. The computer is now analyzing the pixels in the images to find patterns that distinguish Class 1 from Class 2.

Once training is complete, the "Preview" panel on the right will activate your webcam. Now comes the testing phase. Have your child hold up one of the objects in front of the camera and watch the real-time confidence bars at the bottom of the preview panel. The bars will shift dynamically, showing the percentage of certainty the model has for each class.

To make this a true scientific experiment, try to "break" the model. What happens if you hold the object upside down? What happens if you cover half of it with your hand? What happens if you introduce a completely new object that the model has never seen before? This experimentation demonstrates the limitations of pattern recognition.

Parent & Educator Sidebar

Core AI Concept: Supervised Learning Explain to your child that this model uses supervised learning. We act as the teachers by giving the computer labeled examples (the training data). The computer does not actually "know" what a LEGO is: it simply recognizes patterns of colors, edges, and shapes associated with the label we provided.

Conversation Starters:

• "Why do you think the computer got confused when we turned the toy upside down? How could we retrain it to fix that problem?"
• "If we only trained the model with blue pens, would it recognize a red pen? What does this tell us about the importance of having diverse data?"
• "How do you think self-driving cars use this exact same technology when they are driving down a busy street?"

Action Steps Summary

• Gather Materials: Find a laptop with a webcam and select two distinct household items or toys.
• Open the Tool: Navigate to Teachable Machine and start a new standard image project.
• Capture and Train: Record at least one hundred varied images for each object class, then click "Train Model."
• Test and Discuss: Use the live preview to test the model, try to challenge its accuracy, and use the sidebar questions to discuss how machine learning works.

Want more family-friendly AI projects and executive workflows?