Build a Teachable Machine Image Classifier With Your Kid

Issue 30 • February 14, 2024

Children today are surrounded by artificial intelligence, from video recommendation algorithms to voice assistants, yet they often view these tools as pure magic. To prepare the next generation for a technology-driven future, we must demystify how these systems actually work. This guide provides a hands-on, zero-cost project to build a custom image classifier with your child using Google's Teachable Machine. By training a model to recognize physical objects using a standard webcam, your child will transition from a passive consumer of technology to an active creator, gaining a foundational understanding of machine learning without writing a single line of code.

Demystifying AI for the Next Generation

For children aged 8 to 16, theoretical explanations of neural networks and training datasets can quickly become dry and unengaging. Instead, the most effective way to teach artificial intelligence is through direct, interactive experimentation. When children actively build and train a model, they realize that AI is not an all-knowing entity. Instead, they see that it is a system that learns patterns from the specific data they provide.

This project requires no paid subscriptions, no specialized hardware, and no programming experience. All that is needed is a computer or tablet with a functional webcam and an internet connection. By spending thirty minutes on this activity, you can help your child develop critical thinking skills regarding data quality, algorithmic bias, and the mechanics of modern technology.

Setting Up Your Free AI Lab

To begin, open a web browser and navigate directly to the official Google Teachable Machine website at https://teachablemachine.withgoogle.com/ on your computer. This free tool runs entirely in the browser, meaning no software installation or account creation is required, ensuring a safe and private environment for your child.

Once on the homepage, click on the Get Started button and select Image Project from the available options. Next, choose Standard Image Model. You will be presented with a clean, intuitive interface divided into three main sections: Classes, Training, and Preview. Explain to your child that these three sections represent the lifecycle of building any artificial intelligence model: gathering data, training the brain, and testing the results.

Step-by-Step: Building Your First Image Classifier

To make the project engaging, choose a fun and simple classification task. A highly successful option is teaching the computer to distinguish between two different physical objects, such as a favorite toy and a school notebook, or distinguishing between three different hand gestures like rock, paper, and scissors.

First, define the categories. In the first box, labeled Class 1, click the pencil icon and rename it to match your first object, for example, Toy Dinosaur. In the second box, labeled Class 2, rename it to match your second object, such as Notebook.

Second, gather the training data. Click the Webcam button in the Toy Dinosaur class box. Instruct your child to hold the toy dinosaur in front of the webcam. Click and hold the Record button to capture at least one hundred images. While recording, have your child slowly rotate the toy, move it closer and further from the camera, and tilt it. Explain that this variety helps the computer learn what the toy looks like from all angles. Repeat this exact process for the Notebook class, capturing another one hundred images of the notebook in various positions.

Third, train the model. Click the Train Model button in the middle column. Warn your child not to close the browser tab or look away while the computer processes the images. Within about ten seconds, the browser will finish training the model, and the live preview on the right side of the screen will activate.

Debugging and Refining the Model

Now comes the most educational phase: testing and debugging. Have your child hold the toy dinosaur up to the webcam in the Preview column. The progress bars at the bottom should show a high confidence level, close to one hundred percent, for the Toy Dinosaur class.

Next, challenge your child to trick the computer. What happens if they hold the toy dinosaur upside down? What happens if they hold up a different toy, or if you put your hand in the frame? Often, the computer will become confused and misclassify the object.

Explain that this confusion is not a computer glitch, but a reflection of the training data. If the computer only saw the toy dinosaur held right-side up in bright lighting, it will struggle to recognize it in other conditions. To fix this, click the webcam button again and record more diverse images, such as holding the toy upside down or in dimmer light. Click Train Model again and observe how the accuracy improves. This iterative process beautifully illustrates how machine learning models rely entirely on the quality and diversity of their input data.

Parent and Educator Sidebar

Core AI Concept: Supervised Learning. Explain to your child that they just performed supervised learning. This is a process where an AI model is trained using labeled data (the images labeled as Toy Dinosaur or Notebook). The computer looks for common patterns, such as colors, edges, and shapes, to distinguish between the labels.

Conversation Starters:

• If we trained this model using only red toys, do you think it would recognize a blue toy dinosaur? Why does the color of our training data matter?
• How do you think a self-driving car uses this kind of technology to recognize stop signs versus pedestrians?
• Why is it important for the people who build AI to use diverse data from many different types of people and environments?

Action Steps Summary

• Open https://teachablemachine.withgoogle.com/ in your web browser.
• Select Image Project and choose the Standard Image Model.
• Rename the classes and record one hundred webcam images for each object.
• Click Train Model and wait for the browser to process the data.
• Test the model in the Preview panel and add more diverse images to improve accuracy.

Empower the next generation of builders by demystifying technology today.