PICTURE THIS · DAY 3 OF 5
Be the AI: Rules, Data, and Bias
Sort by hand, write the rules you used, then see what happens when a model meets examples outside its training data.
Topic: Classification rules and dataset bias
Objective: I can classify images using visible rules and test whether those rules work on new examples.
Demonstration of learning: I can explain which pattern the model learned and why a bias-test image confused it.
Part 1 · Sort like a model
- Work with a partner. Cut out the cat/dog images and sort them into two classes.
- Write the visible rules you used. Do not use 'because I just know.'
- Trade one uncertain image with another pair and compare rules.
Part 2 · Train the same idea
- Open Teachable Machine.
- Use the provided cat/dog set to recreate the two classes.
- Train, then test with the separate bias-test photos.
- Record which image failed, the prediction, and the pattern you think caused it.
Work-time success criteria
- Your pair wrote at least two visible classification rules.
- You tested every bias-test image.
- You recorded a model failure or low-confidence prediction.
- Your explanation names a gap in the training data.
Sentence stem: The model probably learned ______, so it misclassified ______ when ______.
RAISE Playground was created by the MIT RAISE Initiative and the Personal Robots Group at the MIT Media Lab.