Build a Teachable Machine Image Classifier With Your Kid

Issue 9 • April 26, 2023

Most children today are passive consumers of artificial intelligence, interacting with algorithms through video recommendations, voice assistants, and gaming filters. To prepare the next generation for a technology-driven future, we must shift their perspective from consumption to creation. By building a functional image classifier, children aged eight to sixteen can demystify how computers learn to see. Using Google's free Teachable Machine platform, you and your child can train a custom computer vision model in under fifteen minutes. This hands-on project requires no coding experience, no paid subscriptions, and no specialized hardware, making it the perfect weekend activity to introduce core computer science concepts.

What is Teachable Machine?

Google's Teachable Machine is a web-based tool that makes creating machine learning models fast, accessible, and easy for everyone. It runs entirely in the web browser, meaning your data, including webcam feeds, stays private and is processed locally on your computer.

To begin this project, open a web browser on any computer with a webcam and navigate directly to https://teachablemachine.withgoogle.com/ . Click on the "Get Started" button and select "Image Project," followed by "Standard Image Model." This setup provides a clean interface with two default classes, a training panel, and a preview panel.

The Project: Healthy Snack vs. Treat Classifier

While you can classify anything, a highly engaging project for kids is building a "Healthy Snack vs. Treat Classifier." Gather two items from your kitchen: one healthy item, such as an apple or a banana, and one treat, such as a cookie or a juice box. Your goal is to teach the computer to recognize which is which.

Step 1: Define the Classes

In the first box, labeled "Class 1," click the pencil icon to rename it. Type "Healthy Snack." In the second box, labeled "Class 2," rename it to "Treat." These classes represent the categories that your artificial intelligence model will learn to distinguish.

Step 2: Gather the Training Data

Click the "Webcam" button inside the "Healthy Snack" class box. Allow the browser to access your camera. Have your child hold the healthy snack in front of the camera. Click and hold the "Hold to Record" button.

Instruct your child to slowly rotate the snack, move it closer to the camera, and then further away. This variation helps the model learn what the object looks like from different angles. Capture at least one hundred images. Repeat this exact process for the "Treat" class, recording at least one hundred images of the cookie or juice box.

Step 3: Train the Model

Once both classes have sufficient images, click the "Train Model" button in the middle column. It is crucial to leave the browser tab open and untouched during this process. The computer is now analyzing the pixels in the images, finding patterns that distinguish the healthy snack from the treat. This process typically takes thirty to sixty seconds.

Step 4: Test and Refine

Once training is complete, the "Preview" panel on the right will activate, showing a live webcam feed. Have your child hold up the healthy snack. The progress bars at the bottom should show a high percentage, close to one hundred percent, for "Healthy Snack." Now, hold up the treat and watch the bar switch to "Treat."

To test the limits of the model, try holding up a completely different healthy snack, like a carrot. Does the model still classify it correctly? If not, discuss why. The model was only trained on an apple, so it does not know what a carrot is. You can easily add more images of different healthy snacks to the first class and retrain the model to make it smarter.

Parent & Educator Sidebar

Core AI Concept: Supervised Learning This project demonstrates supervised learning, where an AI model is trained on labeled data. The computer does not actually understand what an apple or a cookie is. Instead, it looks for mathematical patterns in the colors, shapes, and edges of the pixels associated with each label.

Conversation Starters:

• What happens if we block half of the cookie with our hand? Why does the computer get confused?
• How does the computer tell the difference between the two items if the lighting in the room changes?
• If we only train the model with red apples, will it recognize a green apple as a healthy snack? How can we fix that bias?

Action Steps Summary

• Open the Tool: Go to https://teachablemachine.withgoogle.com/ and start a new Standard Image Project.
• Label Your Classes: Create two distinct categories, such as "Healthy Snack" and "Treat."
• Record Training Data: Use your webcam to capture at least one hundred varied images for each category.
• Train and Test: Click train, wait for the process to finish, and test the model with new objects to explore how the AI makes decisions.

Empower your children to build technology rather than just consume it.