Before the models, the metrics, and the chatbots, there is one idea everything else stands on: a computer can work out its own rules from examples, instead of following rules a person wrote. This chapter builds that idea properly — what it means, where it already surrounds you, what a "model" physically is, and the judgment call the professionals make constantly: when learning from examples is the right tool, and when it absolutely isn't.
5 topics
Machine learning has a reputation problem: it is explained either as magic ("the computer teaches itself!") or as math (a wall of formulas by page two). Both explanations lose the actual idea, which is simple enough to state in one sentence: when nobody can write the rules for a task, show the computer thousands of examples and let it find the rules itself. This chapter is that sentence, made solid.
Five topics build it. First, the idea itself — written rules versus learned rules, and which tasks belong to each. Then the map of terms — AI, ML, deep learning, generative AI — that turns the buzzword soup into four nested boxes. Then a walk through one ordinary morning to see how much of your day already runs on learned predictions. Then the word you will use most for the rest of the book: what a model physically is (a file of numbers — really). And finally the chapter's sharpest lesson, the one that sets this book's tone: when not to use machine learning at all.
The one distinction this chapter installs
Rules written by a person
"Free delivery over $30." Exact, explainable, cheap — and only possible when someone can state the rule.
Rules learned from examples
"Is this review fake?" No writable rule exists — so the computer learns the pattern from thousands of labeled examples.