PICTURE THIS · DAY 1 OF 5
Meet the AI: Train It, Then Break It
Teach a computer what your thumbs mean, then discover what it did not learn.
Topic: Image classification
Objective: I can train a two-class image model and test it with examples it has not seen.
Demonstration of learning: I can name one condition that made my model fail and explain what training data it needs next.
Launch
A computer does not see a thumb the way you do. It finds patterns in the examples you provide.
Predict: What could change about a thumbs-up photo without changing its meaning?
Build the first model
- Open Teachable Machine and choose Image Project → Standard image model.
- Rename the two classes Thumbs Up and Thumbs Down.
- Record at least 30 examples for each class. Change the angle, distance, hand position, and background.
- Train the model. Watch the confidence bars while you test.
Try to break it
- Turn your thumb sideways.
- Use your other hand.
- Move closer and farther away.
- Change the background or lighting.
- Ask another person to test it.
Work-time success criteria
- Both classes have at least 30 varied examples.
- The model predicts both classes.
- You recorded one failure condition.
- You can say what new examples would make the model more reliable.
Canvas check: Export the shareable model URL and submit it here. In the comment, complete: My model struggled when ______ because my training data did not include ______.
Need a walkthrough?
Watch this short classroom walkthrough, then return to your own model.
Watch the 2:24 walkthroughRAISE Playground was created by the MIT RAISE Initiative and the Personal Robots Group at the MIT Media Lab.