Stage 1: AI Learns from Examples
Keep both tabs open all week. Open in a new tab — don’t use the buttons in this page to leave the course.
four empty groups in Teachable Machine and a few photos in one of them
what 'examples' and 'labels' mean for an AI
a Teachable Machine project that is ready to train tomorrow
Before campers touch their laptops, show the whole room:
- Open Teachable Machine. Point at the boxes called Class 1 and Class 2.
- Rename Class 1 to Rock. Say out loud: "This is the name tag. We call it a label."
- Click the webcam button. Take 2 photos of your rock hand. Say: "These are the examples."
- Don't click Train yet. That is tomorrow.
The big idea
An AI is a computer that learns from examples. That is the whole secret.
The AI does not have eyes. It does not have a brain. It just looks at a big pile of pictures we show it. We put a name tag on every pile — "these are rocks" or "these are scissors". After it sees enough piles, it can start guessing what new pictures show.
📷 📷 📷 📷 ← these are EXAMPLES
│
▼
"Rock" ← this is the LABEL (the name tag)
Today is not about training. Today is about making the piles. Tomorrow we let the AI practice with them.
- AI
- a computer that learns from examples
- example
- a picture we show the AI
- label
- the name tag for a group of examples
- class
- what Teachable Machine calls a group
- training
- letting the AI practice with the examples (we do this tomorrow)
Both your tabs are open from Setup. If not, click the buttons at the top of this page.
Build it
Step 1 — Open your Teachable Machine project
Click the Teachable Machine button at the top of this page. You should see two boxes: Class 1 and Class 2.
If you see a Get Started screen instead, click Image Project, then Standard image model.
Step 2 — Name your four classes
We need four name tags. Click on Class 1 and type a new name:
Rock
Click on Class 2 and type:
Paper
Now click Add a class at the bottom. Name it:
Scissors
Add one more class and name it:
Nothing
The "Nothing" class is important. It is for when no hand is in front of the camera. Without it, the AI will always pick one of the other three — even if there is no hand at all.
Step 3 — Take three photos of Rock
We are going to add examples to just one class today. The other three classes wait until tomorrow. The point is to feel the loop, not to fill the whole project.
In the Rock class:
- Click the Webcam button.
- Make a rock hand sign (a closed fist).
- Click Hold to Record for one second. You should see two or three photos appear in the Rock box.
That's it. Three photos. Don't fill up the box — we'll do that tomorrow.
Understand it
The AI is more like a photo album than a brain. It does not "know" what rock looks like. It just remembers the pictures we put in the album under the name "Rock".
When the AI sees a new picture later, it will look through its album and ask: "which group looks the most like this new one?" Whichever group wins, that's its guess.
The reason labels matter is that every picture needs a name tag. A photo album with no labels is just a pile of pictures — nobody can find anything. The AI is the same.
The reason "Nothing" is its own class is more subtle. If we only teach the AI about rock, paper, and scissors, then when we show it an empty hand, it has to pick one of those three. None of them is right, but the AI will still pick — and it will sound confident about a wrong answer. By teaching it what no hand looks like, we give it a way to say "I see nothing" on purpose.
Try this
Try this
Three short experiments. Predict before you run, then test your guess.
Tomorrow we will train the model. Imagine we only used 3 photos for Rock and 0 photos for the other classes. Predict what the AI's guesses will look like. Why?
Try renaming your "Rock" class to something silly like thing_1. Two days from now, will you remember what was in there? Labels are for you as much as the AI. Rename it back to Rock.
Stage 2 fills up the other three classes. Look at your Rock box right now. How many photos do you think it really needs before the AI can recognize a rock sign?
Test your stage
- You have four classes in Teachable Machine: Rock, Paper, Scissors, and Nothing.
- The names are spelled exactly like above (capital first letter).
- The Rock class has at least 3 photos.
- The other three classes are empty on purpose — we fill them tomorrow.
- Design check. Read your class names out loud. Could a friend looking over your shoulder tell what each class is for?
If it breaks
- My camera turned off. Teachable Machine sometimes pauses the camera. Click the webcam icon to turn it back on. If your browser blocked the camera, click the camera icon in the address bar and choose Allow.
- I accidentally trained the model. That's fine. Click Reset at the bottom of the model panel, or just ignore it — we will train fresh tomorrow.
- I added five or six classes by accident. Hover over the extra class. Click the trash can icon to remove it. Stay at four.
- My photos all look the same. That is okay for today. Tomorrow we will vary the photos on purpose.
The single biggest temptation at this age is to click Train immediately. The whole point of Stage 1 is to slow that down so kids understand the examples + labels idea before training is a thing they can do. If a camper trains early, that's okay — it doesn't break anything — but use the moment to point out that the model will be very wrong (because three of the classes are empty) and that's why we wait.
If a camper finishes in 15 minutes, push them into the medium stretch (more Rock photos, varied angles). Don't let them fill up Paper, Scissors, or Nothing today — those are the Stage 2 reward. Tomorrow's "I get to train it!" energy depends on today's restraint.
Walk the room and read class names. Look for typos like "rocks", "Rock1", or "rockpaper". Those will cause confusion in Stage 4 when we connect to RAISE.