Can a machine learn?
Artificial Intelligence
Work at your pace. Take a break between stages; there is no countdown.
You need: paper, a pencil and this lab. A helper is welcome for reading and discussion.Connect to what you know
Your game used rules you wrote. A learning system builds a rule from labelled examples.
Understand it, step by step
- 1
Training examples pair an input with a label. Here, every circle should go in box A and every square in box B.
- 2
Our tiny classifier counts labels for a chosen feature: colour or shape. It predicts the most common label seen for that feature. This is a limited classroom model, not a general AI.
- 3
Training accuracy is not enough. Test new combinations that were not in training. If all training circles were red, colour can look useful while actually giving the wrong rule.
Training: red circle → A, blue square → B. A colour model predicts red square → A, which is wrong. A shape model predicts square → B even when the colour changes.
Pause and predict
What will a colour model do with a blue circle?