Build a Teachable Machine Image Classifier With Your Kid

Issue 27 (January 3, 2024) • Hands-On AI Projects for Ages 8 to 16

Children today are surrounded by artificial intelligence, from the recommendation algorithms on YouTube to the voice assistants in our homes. However, most kids experience AI as passive consumers, viewing it as a form of digital magic rather than a tool built on data and logic. To prepare them for a future shaped by technology, we must help them transition from consumers to creators.

You do not need a computer science degree, expensive hardware, or a paid subscription to teach your child how machine learning works. Using a free, web-based tool developed by Google, you and your child (ideal for ages 8 to 16) can build, train, and test a custom image classifier in less than twenty minutes. This hands-on project demystifies AI, showing them exactly how computers learn to see the world.

Step 1: Set Up Your Workspace

To get started, sit down with your child at a laptop or desktop computer equipped with a working webcam. Open a web browser and navigate directly to Google Teachable Machine . This platform requires no login, no email registration, and is completely free to use.

Click on the "Get Started" button, then select "Image Project" followed by "Standard Image Model". You will see a clean interface divided into three main sections: Classes (your training data), Training (the learning phase), and Preview (the testing phase).

Step 2: Choose Your Classification Project

Before recording data, decide what you want your model to recognize. Simple, high-contrast objects work best. Here are three excellent starter ideas:

• The Toy Classifier: Train the computer to distinguish between a Lego brick and an action figure.
• The Snack Classifier: Train it to recognize an apple versus a banana.
• The Gesture Classifier: Train it to recognize a thumbs-up versus a peace sign.

For this guide, we will use the Gesture Classifier as our example. Rename "Class 1" to "Thumbs Up" and "Class 2" to "Peace Sign" by clicking the pencil icon next to each label.

Step 3: Gather Your Training Data

Now, your child will act as the data scientist. Click the "Webcam" button under the "Thumbs Up" class. Have your child hold a thumbs-up gesture in front of the camera. Click and hold the "Hold to Record" button to capture about 100 to 150 images.

Encourage your child to move their hand slightly, tilt their head, and change their distance from the camera while recording. This variety is crucial. It teaches the computer to recognize the gesture from different angles, rather than just memorizing a single static image. Repeat this exact process for the "Peace Sign" class, capturing another 100 to 150 images of that gesture.

Step 4: Train and Test the Model

Once you have gathered data for both classes, click the "Train Model" button in the middle column. This process takes about ten seconds. Make sure to keep the browser tab open and active while the computer processes the images.

Once training is complete, the "Preview" panel on the right will activate your webcam. Have your child make a thumbs-up gesture in front of the camera. Watch the live bar graphs below the video feed. They will dynamically shift to show a percentage confidence level (for example, 99% Thumbs Up). Switch to a peace sign and watch the bars instantly swap.

Step 5: Break the Model (Adversarial Testing)

This is the most educational part of the exercise. Challenge your child to "trick" the computer. What happens if you hold up a thumbs-down gesture? What happens if you use your left hand instead of your right hand? What happens if you hold up a completely different object, like a mug? You will quickly see the confidence bars fluctuate wildly. Explain to your child that the computer does not actually understand what a hand is, it only understands the patterns of pixels you provided during the training step.

Parent & Educator Sidebar

Core AI Concept: Supervised Learning

Supervised learning is the process of training an AI model by giving it labeled examples (the training data). The computer looks for patterns (colors, edges, shapes) that are common to all images in a specific class. When you show it a new image, it compares the new patterns to the old ones to make a prediction.

Conversation Starters:

• "If we only trained the computer using my hand, why does it struggle to recognize your hand? How does this relate to bias in real-world AI?"
• "How do you think a self-driving car uses this technology to tell the difference between a stop sign and a speed limit sign?"

Action Steps Summary

Looking for more weekend projects to build with your family?