Build a Teachable Machine Image Classifier With Your Kid

Most children interact with artificial intelligence as passive consumers (scrolling through algorithmic feeds, playing AI-driven video games, or asking voice assistants to play music). To prepare the next generation of leaders, we must shift their relationship with technology from consumption to creation. By building a custom image classifier, children ages 8 to 16 can demystify how computers learn to see and understand the physical world. Using Google's free Teachable Machine platform, you and your child can train a machine learning model to recognize hand gestures, toys, or facial expressions in real-time. This hands-on project requires no coding experience, no paid subscriptions, and no specialized hardware (just a standard computer with a webcam).

Launching Your First AI Project

To begin, open a web browser on your computer and navigate directly to https://teachablemachine.withgoogle.com/ . This free tool, developed by Google, allows anyone to train machine learning models quickly and visually. Click on the "Get Started" button and select "Image Project," followed by "Standard Image Model." The interface is divided into three clear sections: Gather (where you input your data), Train (where the computer learns), and Preview (where you test the model). This layout mirrors the actual workflow used by professional AI engineers, making it an excellent introduction to real-world technology concepts.

Defining Your Classes and Gathering Data

The first step in supervised learning is defining what you want the computer to recognize. In machine learning, these categories are called "classes." For this project, you and your child will build a "Rock, Paper, Scissors" game referee. Rename "Class 1" to "Rock" and "Class 2" to "Paper." Click the "Add a class" button to create a third category and name it "Scissors."

Now, it is time to gather training data. Click the webcam button under the "Rock" class. Instruct your child to make a fist (the "Rock" gesture) in front of the camera. Click and hold the "Record" button to capture about one hundred images. While recording, have your child slowly move their hand closer to the camera, further away, and tilt it at different angles. This variation is crucial. Repeat this exact process for the "Paper" class (flat hand) and the "Scissors" class (two fingers extended). Explain to your child that they are acting as the teacher, providing the computer with examples of what each gesture looks like from different perspectives.

Training the Model

Once you have gathered at least one hundred images for each of the three classes, click the "Train Model" button in the middle column. This process takes about ten to thirty seconds, depending on your computer's speed. During this phase, the browser is processing the images and finding mathematical patterns that distinguish a fist from a flat hand or extended fingers.

Instruct your child to keep the browser tab open and active while the training completes. This is an excellent moment to explain that the computer is not memorizing the exact images. Instead, it is learning the unique features of each gesture (such as the round shape of a fist versus the long, separated lines of a hand showing paper).

Testing, Iterating, and Finding Edge Cases

Once training is complete, the "Preview" column on the right will activate, turning on your webcam. Have your child make one of the gestures in front of the camera. The model will instantly display its confidence level for each class (for example, "Rock: 98%").

Now comes the most educational part of the project: trying to break the model. Challenge your child to find "edge cases." What happens if they use their left hand instead of their right hand? What happens if they make the gesture in the background while you stand in the foreground? What happens if they use a different family member's hand?

If the model struggles to identify a gesture, explain that this is a data problem, not a computer problem. To fix it, simply go back to the corresponding class, record more diverse images (such as using the other hand or changing the lighting), and click "Train Model" again. This iterative process teaches children that AI systems are only as good as the data used to train them.

Parent & Educator Sidebar

Core AI Concept: Supervised Learning

Supervised learning is a type of machine learning where the model is trained on labeled data. In this project, the labeled data consists of the images categorized into "Rock," "Paper," and "Scissors." The computer analyzes these labeled examples to find patterns, which it then uses to classify new, unseen images.

Conversation Starters:

• "If we only trained the model using my hand, why might it struggle to recognize your hand? How does this relate to bias in real-world AI systems?"
• "How do you think self-driving cars use this same technology to recognize stop signs, pedestrians, and other vehicles?"
• "What other classes could we create to make a helpful tool for our home (for example, sorting recycling from trash)?"

Action Steps Summary

• Open the Tool: Go to https://teachablemachine.withgoogle.com/ and start a new Standard Image Project.
• Define Classes: Create three classes named "Rock," "Paper," and "Scissors."
• Record Training Data: Capture at least one hundred webcam images for each class, ensuring variety in angles and distances.
• Train and Test: Click "Train Model," then test the classifier in the preview window using different hands and angles.

Looking for more engaging, hands-on AI projects to explore with your family?