Build a Teachable Machine Image Classifier With Your Kid

Issue 51 • December 4, 2024

Children today are surrounded by artificial intelligence, but their relationship with it is almost entirely passive. They watch algorithms recommend videos, interact with voice assistants, and use generative tools to write stories. To truly understand the technology that will shape their careers, they need to move from consumers to builders. They need to see behind the curtain and understand how computers learn to see, hear, and make decisions.

You do not need a background in software engineering, a paid subscription, or expensive hardware to teach these concepts. Using a free tool developed by Google, you and your child (ideally between the ages of 8 and 16) can build, train, and test a fully functional computer vision model in less than twenty minutes. This hands-on project demystifies machine learning by turning it into a tangible, interactive game.

The Project: Choosing Your Classes

Before opening the software, sit down with your child and decide what you want your computer to recognize. In machine learning, the categories we want a model to identify are called "classes."

For your first project, choose two or three distinct physical objects that are easily accessible around your home. Excellent beginner ideas include:

• Lego vs. Action Figure: Teach the computer to distinguish between different types of toys.
• Apple vs. Banana: Build a healthy snack detector.
• Thumbs Up vs. Thumbs Down: Create a gesture recognition system that reads hand signals.

Once you have selected your objects, sit together at a computer or laptop equipped with a standard webcam.

Step-by-Step Build Guide

Follow these precise steps to build your image classifier:

• Access the Platform: Open your web browser and navigate directly to Google Teachable Machine . Click on the "Get Started" button, then select "Image Project" followed by "Standard Image Model."
• Define Your Classes: On the screen, you will see two default boxes labeled "Class 1" and "Class 2." Click the pencil icon next to "Class 1" and rename it to match your first object (for example, "Apple"). Rename "Class 2" to match your second object (for example, "Banana").
• Gather Training Data: Click the "Webcam" button inside the first class box. Hold your first object up to the camera. Click and hold the "Record" button to capture at least 100 images. While recording, slowly rotate the object, move it closer to the lens, and tilt it. This variety helps the computer learn what the object looks like from different angles. Repeat this exact process for the second class using your second object.
• Train Your Model: Click the blue "Train Model" button in the middle column. Keep your browser tab open and active. It will take about ten seconds for the computer to analyze the images and find the patterns that distinguish your objects.
• Test and Play: Once training is complete, a live preview window will appear on the right side of the screen. Hold up one of your objects. The model will display a real-time confidence score (from 0% to 100%) indicating which object it thinks it is seeing.

Parent & Educator Sidebar

Core AI Concept: Supervised Learning Explain to your child that they just performed supervised learning. The computer did not automatically know what an apple was. It learned by looking at examples that were labeled by a human. The quality of the model depends entirely on the variety and accuracy of the training data provided.

Conversation Starters:

• "What happens if we hold up a completely different object, like a toy car? Why does the computer guess one of our two classes instead of saying it does not know?"
• "How could we trick the computer? What if we hide half of the banana behind our hand?"
• "If we only trained the model with red apples, would it recognize a green apple? How does this show how computer bias happens?"

Iterating and Breaking the Model

The most educational part of this exercise is trying to break the model. Encourage your child to find the limits of what they built. For example, if you trained the model using your right hand for gestures, try using your left hand. Does it still work? If you change the lighting in the room, does the accuracy drop?

This experimentation teaches children that artificial intelligence is not magic: it is a mathematical pattern-matching system. If the training data is limited, the model will fail. To fix a failing model, they do not need to rewrite code. They simply need to add more diverse training images (for example, capturing images in different lighting or using different hands) and retrain.

Action Steps Summary

• Pick Your Objects: Select two distinct household items with different shapes and colors.
• Visit the Site: Go to Teachable Machine on a computer with a webcam.
• Capture and Train: Record 100 images for each object, click train, and watch the model compile.
• Discuss and Test: Use the sidebar questions to discuss how the computer processes visual information.

Want more family-friendly AI projects and educational guides?