What is Machine Learning?

The spam-filter story (all of ML in one example)

You cannot write spam rules — scammers change them daily. So instead: collect a million emails labeled "spam" / "not spam" and let an algorithm tune its internal knobs (parameters) until its guesses match the labels. That is training. The result generalizes — it flags tomorrow's scam it has never seen, because scam-ness has patterns.

Every machine learning system is a version of that story with different materials: clicks instead of emails, X-rays instead of spam.

Why it took over

Classic software is brittle where the world is fuzzy — handwriting, speech, images, language. Learning from examples handles fuzzy. Scale it up enough (data + compute) and you cross from "classifies photos" to "writes essays" — that crossing is generative AI, the same discipline with billions more knobs.

What it is bad at

Novelty (no examples = no pattern), explanation ("why" is statistics, not reasoning), and quiet bias — it learns the prejudices in the training examples as faithfully as the signal. Which is exactly why hallucination-checking habits apply to every ML system, not just chatbots.

Frequently asked questions

How is machine learning different from regular programming?

Regular programming is rules a human writes — "if the email contains X, block it". Machine learning flips it — you show the computer thousands of examples and it works out the rules itself, which is how it catches spam rules no human could write.

Is machine learning the same as AI?

Machine learning is a subset of AI — the part that learns from data. Modern AI you interact with (chatbots, image generators) is essentially all machine learning, specifically deep learning.

What is deep learning?

Deep learning is machine learning using large neural networks with many layers. The "depth" is what let models scale to human-level language and vision — it is the technique behind every [generative AI](/learn/what-is-generative-ai) product you have heard of.