Teacher / research preview
How Machines Learn
Follow an idea from raw examples, through pattern and feedback, into a machine that predicts — and meet the same structures you already know from climate and music.
Concepts walked (7)
Relationships walked (7)
- Generalization enables Prediction Established
- Machine learning enables Generalization Established
- Machine learning depends on Pattern Established
- Model bias is derived from Training data Strongly supported
- Overfitting depends on Training data Established
- Overfitting suppresses Generalization Established
- Training data enables Machine learning Established
Thinking patterns exercised (5)
Activities & expected evidence
- Identify a missing link
A model turns training data into predictions. What has to happen in the middle for the predictions to be any good on new cases?
Expected evidence: identified a missing link
- Explain the mechanism
Explain how a model can be systematically unfair even though it is 'just maths'. What connects biased data to real harm?
Expected evidence: explained a relationship
- Predict an effect
Predict the ripple before you see the answer.
Expected evidence: predicted an effect
- Transfer to a new field
You have now met feedback three times: steadying Earth's climate, holding a living body in balance, and training a machine on its mistakes. Which of these run on the SAME feedback structure?
Expected evidence: transferred a pattern, revised a model
- Reflect
Reflect on what carried over.
Expected evidence: reflected
Transfer challenge
You have now met feedback three times: steadying Earth's climate, holding a living body in balance, and training a machine on its mistakes. Which of these run on the SAME feedback structure?
Machine learning → Balancing loop, Reinforcing loop
What carries over: In all three, a system measures the gap between where it is and where it 'should' be, then feeds that error back to shrink the gap. The loop — sense, compare, adjust — is identical, even though the parts are air, cells, or numbers.
Where the analogy breaks: A thermostat has one fixed goal set by a person. A learning machine changes its own internal rules from data and can pursue goals no one stated — which is also why its feedback can go wrong in ways a thermostat never could.