Book 1 · Chapter 2 · Mission
Your First Train
Teach a computer the difference between two things you choose yourself
⏱ About 15 minutes
Teach a computer the difference between two things you choose yourself
⏱ About 15 minutes
Ready to jump in? The tool you’ll use lives at teachablemachine.withgoogle.com — free from Google, no account, no download, runs right in your browser. Read the safety note below first, then head over.
Teachable Machine is kid-safe. It is a free Google tool. You don’t need an account. You don’t need to download anything. Your photos and your trained model stay on your own computer — they are not sent to Google or anybody else. You can close the tab when you’re done and nothing follows you.
Do this one with a parent around, the way the book suggests. They’ll want to see what you train.
Teach a computer the difference between two things, using training data that you choose yourself. You pick the two things. You take the photos. You watch the AI learn from what you showed it — and then you try to fool it on purpose.
By the end, you will have done the same thing Emma Yang did when she was twelve, at a smaller scale and on your own computer. You are officially on the builder side of this book now, not the audience side.
1 minute
Open your browser and go to teachablemachine.withgoogle.com. Click the big Get Started button, then click Image Project, then click Standard.
You’ll see two boxes on the screen, called Class 1 and Class 2. That’s where your training data goes.
2 minutes
Rename Class 1 and Class 2 to the two things you want your AI to tell apart. Pick anything fun:
The more different the two things look, the easier your AI’s job will be at first. Once you get good, try two things that look similar and see how the AI does.
3 minutes
Click Webcam under Class 1. Hold up the first thing and take 20 to 30 pictures from different angles.
Variety matters more than perfection. The AI will learn whatever you show it — so show it a range.
3 minutes
Same thing for Class 2. Twenty to thirty pictures, different angles, same care.
This is your training data. These photos are what the AI is going to learn from. What you put in here determines what your AI becomes.
1 minute
Click the big Train Model button. Wait about thirty seconds. This is the training loop running in the background — input, guess, compare, adjust, over and over, at high speed. The same four steps you read about in the chapter.
Don’t close the tab while it’s training. If you do, you’ll lose your model and have to start over.
2 minutes
Hold the camera back up to one of your two things. Watch the confidence bars on the right side of the screen. The AI is guessing, right now, in real time. Move the thing around. Move your hand. Watch the bars shift.
You just taught a computer to tell two things apart. Most adults have never done this.
3 minutes
This is the most interesting part of the whole mission. Try to break your AI on purpose.
Find the moments where it gets confused. Those confused moments tell you exactly what was missing from your training data. That is not a failure. That is information.
When your kid’s AI gets confused, ask them: “What was missing from the data?” That one question is the whole shape of lifelong AI literacy. It transfers to every result any AI ever hands them for the rest of their life.
Add 5 minutes
Show a parent or a friend what you built. Let them try to fool your AI on purpose.
See if they can figure out what was missing from your training set, just by watching the confused cases. That’s what real AI testers do for a living.
Teachable Machine lets you save your trained model directly to your browser, or export it for later use. Hang onto this one.
You’ll keep using the training-loop idea for the rest of the book — Input, Guess, Compare, Adjust — so the model you just made is a real artifact of the moment you got it.
“You just did what Emma Yang did, at a smaller scale and on your own computer. Input. Guess. Compare. Adjust. Those four steps, running thousands of times, are the engine of every AI on earth. Now you’ve felt them in your own hands.”
“And you probably noticed the moments when your AI got confused. That confusion is not a failure. That confusion is actually information. A good builder looks at every mistake the AI made and asks what was missing, and then they go fix it.”
“The webcam won’t turn on.”
Your browser is asking for permission to use the camera and you probably missed the popup. Look near the top of the page for a little camera icon or a notification bar, and click “Allow.” If that doesn’t work, try a different browser like Chrome or Firefox. Teachable Machine works best in modern browsers.
“My AI gets it wrong most of the time.”
Your two things probably look too similar. Or you didn’t take enough photos. Or the photos were all from the same angle. Go back, add 10–15 more photos of each class with more variety (different angles, different backgrounds, different lighting), and train again. Most classifier problems are really training-data problems.
“Train Model is taking forever.”
Training usually takes 20–60 seconds. If it’s been longer than two minutes, your internet connection might have dropped. Refresh the page and try again. Teachable Machine runs mostly in your browser, but it needs a stable connection to load the starter model.
“I want to save my model but I don’t see how.”
Look for the “Export Model” button near the top right of the training panel. You can save it to your browser (fastest), download the files to your computer, or upload to Google Drive. For this mission, saving to your browser is plenty.
“I don’t have a webcam.”
That’s okay. Teachable Machine also lets you upload photos instead. Take 20–30 photos with a phone, email or AirDrop them to the computer, and click “Upload” instead of “Webcam” under each class.