Build a Teachable Machine Image Classifier With Your Kid

Issue 33 • March 27, 2024

Most children interact with artificial intelligence as passive consumers, playing games or asking chatbots to write stories. To prepare the next generation of leaders for an AI-driven world, they must understand how these systems actually learn. Google Teachable Machine offers a free, highly visual, and interactive way for kids aged 8 to 16 to build their very own machine learning model in under twenty minutes. By using a standard computer webcam, children can train an image classifier to recognize different objects, hand gestures, or facial expressions. This hands-on project demystifies complex computer science concepts, transforming AI from a mysterious black box into a practical tool they can control and build.

The Project: The Object Detector

This project involves building a model that can instantly distinguish between three different items: a school notebook, a favorite toy, and an empty hand. This requires no coding, no paid accounts, and no specialized hardware. All you need is a computer with a webcam and an internet connection. By guiding your child through this process, you will help them understand the fundamentals of computer vision and supervised learning.

Step 1: Setting Up the Workspace

Begin by navigating to the Google Teachable Machine website at https://teachablemachine.withgoogle.com/ . Click on the Get Started button on the homepage. You will be presented with three project options: Image Project, Audio Project, and Pose Project. Select Image Project, and then click on Standard Image Model. This opens the training interface, which is divided into three clear sections: Classes, Training, and Preview.

Step 2: Defining the Classes

Explain to your child that computers do not know what objects are until we teach them. In machine learning, these categories are called classes. By default, the interface shows Class 1 and Class 2. Click the pencil icon next to Class 1 and rename it to Notebook. Rename Class 2 to Toy. Click the Add a class button at the bottom to create a third class, and rename it to Empty Hand.

Step 3: Gathering the Training Data

Now it is time to collect the data. Click the Webcam button inside the Notebook class box. Have your child hold a school notebook in front of the webcam. Instruct them to hold the Record button while slowly moving the notebook around, tilting it, bringing it closer to the camera, and moving it further away. Capture approximately 100 to 150 images. This variety helps the computer learn what a notebook looks like from different angles. Repeat this exact process for the Toy class using a favorite toy, and finally for the Empty Hand class, capturing images of just their hand in various positions.

Step 4: Training the Machine Learning Model

With the data collected, click the Train Model button in the middle column. A small loading wheel will appear. It is crucial to keep the browser tab open and active during this process. Explain to your child that the computer is currently analyzing thousands of pixels across all the captured images, looking for common patterns (such as colors, edges, and shapes) that distinguish the notebook from the toy and the empty hand. This training process typically takes less than thirty seconds.

Step 5: Testing and Debugging the Model

Once training is complete, the Preview panel on the right side of the screen will activate, showing a live feed from the webcam. Underneath the video feed, you will see three progress bars representing the three classes, each showing a percentage from 0 to 100. Have your child hold up the notebook. The Notebook bar should instantly jump to 100 percent. Now, have them hold up the toy, and then their empty hand, observing how the percentages shift in real time.

To deepen the learning experience, challenge your child to trick or break the model. What happens if they hold up a different book that was not part of the training data? What happens if they cover half of the toy with their hand? What happens if they turn off the lights in the room? When the model gets confused (for example, identifying a different book as a Toy with 70 percent confidence), explain that this is due to limitations in the training data. This hands-on debugging demonstrates that an AI is only as good as the data used to train it.

Parent and Educator Sidebar

Core AI Concept: Supervised Learning

Supervised learning is a type of machine learning where an algorithm learns from labeled training data. In this project, the webcam images are the training data, and the labels are Notebook, Toy, and Empty Hand. The computer does not understand what these objects are in a human sense. Instead, it analyzes the pixels to find mathematical patterns that distinguish one label from another.

Conversation Starters:

• "How do you think a self-driving car uses this exact same technology to recognize stop signs, traffic lights, and pedestrians in real time?"
• "If we only trained our model with blue notebooks, why might it fail to recognize a red notebook? How does this show how bias can get into AI systems?"
• "Why is it important for the engineers who build AI systems to use diverse and complete training data when creating tools for the real world?"

Action Steps Summary

• Open https://teachablemachine.withgoogle.com/ on a computer with a webcam.
• Create three classes: Notebook, Toy, and Empty Hand.
• Record 100 to 150 webcam images for each class at various angles.
• Click Train Model and wait for the process to complete.
• Test the model live and experiment with different objects to see how the confidence scores change.

Empower the next generation with practical AI literacy.