Video narration
How does a spam filter catch a message it has never seen before, without being given a rule for every possible scam? It starts with training data: past messages labeled spam or not spam. From these examples, the system finds patterns-combinations of words, senders, or features that often appear in junk. Here is machine learning's key turn: learning is not memorizing every rule. It is finding patterns that help. Those patterns become a model: a learned guide for judging new mail. When a new message arrives, the model uses that guide to predict spam or probably not spam. That is a best guess, not certain understanding. It does not understand a scam's meaning or think about your inbox; it detects statistical patterns, and it can be wrong. Better examples and corrections can improve predictions. That's how: labeled examples become patterns that predict whether unfamiliar mail is junk.