🔄 The great flip: rules vs learning
Why: See the flip that created modern AI: from writing rules to learning them from examples.
Unlocks: Features — how real things become numbers a machine can learn from.
Everything you've programmed so far follows one pattern: you write the rules, the machine applies them. Machine learning flips it: you show the machine examples with answers, and it works out the rules itself. To feel why the flip was necessary, try being the rule-writer for a spam filter:
This is the wall every rule-writer hits: for some problems the rules are unwritable. What rule detects a cat in a photo? Millions of pixels, any lighting, any pose. Humans do it instantly and cannot say how.
The flip: collect 10,000 emails already labeled spam/not-spam and let an algorithm find what actually separates them. The rules it grows are called a model — and the labeled pile it studied is the training data.
Which problem MOST needs machine learning rather than handwritten rules?
In machine learning, where do the "rules" ultimately come from?