Machine learning: how a computer learns from examples

Machine learning (ML) is an approach where a program learns a task from examples: it's shown data and the correct answers, and it works out the rule for getting the answer by itself.

Machine learning: the model learns from examples

In plain words

Try explaining in words to your little brother how to tell a cat from a dog. Hard, right? Ears, face, size - there are endless exceptions. But show him a hundred photos of cats and a hundred of dogs, and he'll learn on his own. Machine learning works the same way: examples instead of rules.

How it works

A dataset is collected: for example, thousands of emails labelled "spam" and "not spam".
The model tries to guess the answer for each example.
The program compares the guess with the correct answer and measures the error.
The model slightly adjusts its internal numbers so it makes fewer mistakes next time.
This is repeated millions of times until the model guesses well on new data too.

The training loop: guess, error, adjust

Where you'll see it

Machine learning powers email spam filters, music and video recommendations, face recognition in your gallery, keyboard suggestions, sleep scores on watches and bank systems that catch fraudsters. Most of what is called AI today is machine learning.

Does it matter when buying

You won't see this term directly in spec sheets. But it helps to understand the principle: a model is only as good as the data it was trained on. If a watch was trained mostly on adult runners, its estimates may be less accurate for a teenage swimmer. That's why I treat "smart" scores as a hint, not a verdict.

Myths

Myth: Machine learning and AI are different things. Reality: Machine learning is part of AI - and the most common part today.
Myth: More data always makes a better model. Reality: Data quality matters too. A pile of wrong examples teaches the model to be wrong.

FAQ

What is "supervised learning"?

When every example has a correct answer, like a textbook with answers at the back.

Does my phone learn from me?

Partly, yes: the keyboard remembers your words, the gallery remembers faces. Often this happens right on the device.

How much data is needed for training?

From hundreds of examples for simple tasks to trillions of words for large language models.