Build a Teachable Machine Image Classifier With Your Kid

Issue 31 • February 28, 2024

Introducing children to artificial intelligence does not require complex coding or expensive software. By using a free, web-based tool, parents and educators can help children build their very first machine learning model in under fifteen minutes. This hands-on project transforms abstract concepts like training data and neural networks into a tangible, interactive experience.

Demystifying AI for the Next Generation

Artificial intelligence is no longer a futuristic concept: it is an active part of daily life. For children aged 8 to 16, understanding how these systems work is a critical literacy skill. Rather than viewing AI as a magical black box, children should understand that it is built on data, patterns, and iterative training. Google's Teachable Machine provides an ideal, accessible entry point. It requires no programming knowledge, runs entirely in a standard web browser, and respects privacy by processing all data locally on your device. By building a custom image classifier, children gain a practical understanding of supervised learning, feature extraction, and model testing.

Step 1: Accessing the Platform

To begin, open a web browser on any computer equipped with a webcam. Navigate directly to the Teachable Machine website at https://teachablemachine.withgoogle.com/ and click the "Get Started" button. From the project creation screen, select "Image Project" and choose the "Standard Image Model" option. This opens the main workspace, which is divided into three clear sections: Classes, Training, and Preview. Explain to your child that this workspace represents the entire pipeline of a machine learning engineer.

Step 2: Defining the Classes and Gathering Data

A machine learning model needs categories to organize what it sees. In AI, these categories are called "Classes". For this project, we will build a classifier that distinguishes between three states: a "Thumbs Up" gesture, a "Thumbs Down" gesture, and a "Neutral" face.

First, rename "Class 1" to "Thumbs Up" by clicking the pencil icon. Click the "Webcam" button to activate your camera. Hold your thumb up in front of the camera and hold down the "Record" button. Move your hand slightly, change your distance, and tilt your head to capture about 150 images. This variety helps the model learn the core shape rather than just a single static image.

Second, rename "Class 2" to "Thumbs Down". Repeat the process, capturing 150 images of a thumbs-down gesture.

Third, click "Add a class" to create a third category. Name it "Neutral" and record 150 images of yourself simply looking at the camera with no hand gestures. This acts as the baseline control group.

Step 3: Training the Model

With the training data gathered, it is time to train the neural network. Click the blue "Train Model" button in the middle column. It is crucial to keep the browser tab open and active during this process. Explain to your child that the computer is currently analyzing the hundreds of images you just recorded. It is looking for common patterns, edges, colors, and shapes that distinguish a thumbs-up from a thumbs-down. The training process typically takes less than thirty seconds, during which the browser processes the mathematical relationships between the pixels in each image.

Step 4: Testing and Iterating

Once training is complete, the "Preview" panel on the right will activate, showing a live feed from your webcam. Below the video feed, you will see progress bars representing each of your three classes, along with a percentage score. This percentage is the confidence score.

Have your child make a thumbs-up gesture. Watch the "Thumbs Up" bar jump to 99% or 100% confidence. Now, make a thumbs-down gesture and observe the shift.

To deepen the learning, test the limits of the model. What happens if you make a thumbs-up gesture using your other hand? What happens if you use a toy or a drawing of a hand? If the model struggles, explain that it needs more diverse training data. You can easily add more images to any class and click "Train Model" again to improve its accuracy.

Parent and Educator Sidebar

Core AI Concept: Supervised Learning

Supervised learning is a type of machine learning where a model is trained on labeled data. In this project, the labeled data consists of the images you recorded, which were explicitly labeled as "Thumbs Up", "Thumbs Down", or "Neutral". The model learns the relationship between the input (the images) and the output (the labels) so it can make predictions on new, unseen images.

Conversation Starters:

• "What happens if we train the model using only my hand, and then you try to use it? Why might the computer get confused?" (This introduces the concept of algorithmic bias and the importance of diverse datasets).
• "How does the computer know the difference between your hand and your face? What specific patterns or shapes is it looking at?" (This introduces feature extraction and computer vision).
• "How might a self-driving car use this exact same technology to recognize stop signs, traffic lights, and pedestrians?" (This connects the simple project to real-world applications).

Action Steps Summary

• Visit https://teachablemachine.withgoogle.com/ on a computer with a webcam.
• Create three distinct classes: Thumbs Up, Thumbs Down, and Neutral.
• Record 150 sample images for each class using the webcam, ensuring diverse angles.
• Click "Train Model" and wait for the browser to process the patterns.
• Test the model in the Preview panel and discuss the core concepts using the sidebar questions.

Ready to explore more AI projects? Start building today and inspire the next generation of technology leaders.