Build a Teachable Machine Image Classifier With Your Kid

Issue 10 • May 10, 2023

As an Executive, you likely discuss artificial intelligence in terms of productivity and strategic advantage. For your children, however, AI is often an invisible force behind their favorite video games and search engines. To prepare them for a future where AI is ubiquitous, we must help them transition from passive consumers of technology to active creators. One of the most effective ways to demystify how computers learn is by building a custom image classifier together. Using Google's free Teachable Machine platform, you and your child can train a real machine learning model in under fifteen minutes. This hands-on project requires no coding experience, no paid subscriptions, and no specialized hardware, making it an ideal weekend activity for children aged eight to sixteen.

Step 1: Setting Up Your AI Lab

To get started, you only need a computer with a working webcam and a few everyday household items. Gather three distinct objects that are easy to hold. Excellent choices include a colorful coffee mug, a small toy car, and a pencil. The goal is to choose items with clear visual differences in shape and color. Sit down with your child at a table where the webcam can easily capture both of you and the objects. Explain to your child that they are about to act as the teacher, and the computer is going to be the student. Just like a human student, the computer needs examples to learn what these objects look like.

Step 2: Accessing the Teachable Machine Interface

Open your web browser and navigate directly to the Google Teachable Machine website at https://teachablemachine.withgoogle.com/ . This tool runs entirely in the browser, meaning no software installation is required, and all data processing happens locally. Click on the Get Started button and select Image Project from the options, followed by Standard Image Model. You will see an interface with two default classes, a training section, and a preview panel. Explain to your child that these classes are like folders where the computer will store the examples they provide.

Step 3: Training the Model with Custom Data

Now, let your child take the lead. Have them rename Class 1 to match the first object, such as Toy Car. Click on the Webcam button within that class. Instruct your child to hold the toy car in front of the camera and press and hold the Record button. As they hold the button, they should slowly rotate the toy car, move it closer to the lens, and pull it back. This captures a wide variety of angles and lighting conditions. Aim to capture at least one hundred image samples. Explain to your child that this collection of images is called training data. Repeat this process for the second class, renaming it to match the second object, such as Coffee Mug, and recording another one hundred images. Once the data is collected, click the Train Model button. This process takes about ten to thirty seconds. Advise your child to keep the browser tab open while the computer processes the images, analyzing the patterns of pixels that define each object.

Step 4: Testing, Iterating, and Breaking 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. Have your child hold up the toy car. The interface will display a real-time confidence score, showing a percentage of how certain the model is that it sees the toy car. Watch your child's excitement as the bar jumps to one hundred percent.

Now comes the most educational part: trying to break the model. Have your child hold up a different toy car that was not used in the training data, or hold the original car at an extreme angle. Observe how the confidence score fluctuates. This is a perfect opportunity to explain that the computer does not actually know what a car is. It only recognizes the specific patterns of colors and shapes from the training images. If the computer fails to recognize the new object, have your child add more diverse images to the training set and train the model again, demonstrating the iterative nature of machine learning.

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 labels were the names of the objects, and the data was the collection of webcam images. The computer looks for mathematical patterns in the pixels to map the input (the image) to the correct output (the label). It does not possess human-like understanding, but rather excels at pattern recognition.

Conversation Starters:

• What do you think would happen if we trained the model with only blue toy cars, and then showed it a red toy car? Why might that be a problem in real-world AI systems?
• How does the computer tell the difference between your hand and the object you are holding? How could we make our training data better so the computer does not get confused by your fingers?
• Can you think of other places where image recognition is used in daily life? Think about how your phone unlocks with your face.

Action Steps Summary

• Gather three distinct household objects and sit down with your child at a computer with a webcam.
• Navigate to https://teachablemachine.withgoogle.com/ and start a Standard Image Project.
• Record at least one hundred webcam images for each object, rotating them to capture different angles.
• Click Train Model and wait for the browser to process the training data.
• Test the model using the live preview, and experiment with different objects to see how the confidence scores change.

Want more practical AI workflows for your daily routine?