A program that learns from examples

In ordinary programming a human writes rules: "if an email contains word X, it is spam". Machine learning (ML) flips the approach: instead of rules you give the program many examples (thousands of emails labeled "spam"/"not spam"), and it finds the patterns distinguishing one from another itself. The model learns rather than following set logic.

What the difference is

• Ordinary program: a human sets rules to the program applies them.
• ML: a human gives data (examples) to the model derives rules itself.

When ML is useful

ML is strong where rules are too complex or non-obvious to write by hand: image recognition (how do you describe "a cat" in code?), understanding text, predicting demand, recommendations. Where patterns exist but are hard to express explicitly — ML finds them in the data.

ML is not "magic" or "thinking"

It is important to understand: an ML model does not "understand" or "think" like a human. It finds statistical patterns in data and applies them. This is powerful but has limits: a model is only as good as the data it learned on, and works poorly on things very different from the training examples. Understanding this essence protects from inflated expectations.

Course rule: ML is learning from data instead of manual rules. The model finds patterns in examples but does not "think"; its quality is determined by the data.