What Is Machine Learning? Explained Simply (With Demo)

What is machine learning? It is the technology behind your spam filter, your Netflix recommendations, the face unlock on your phone and, at a much bigger scale, tools like ChatGPT. Yet the core idea is surprisingly simple: instead of telling a computer exactly what to do, you show it lots of examples and let it work out the rules for itself.

In this guide you'll learn how that works in plain English, try training a tiny spam filter yourself, and finish with a quick quiz. No maths, no code.

What is machine learning, in one sentence?

Machine learning is a way of building software that learns from examples instead of following rules a person wrote by hand.

Think about how a child learns what a dog is. Nobody hands them a rulebook (“four legs, fur, a tail, barks, unless it's a very quiet dog…”). They simply see lots of dogs, get corrected now and then (“no, that's a cat”), and gradually build their own sense of “dog”. Machine learning works the same way, just with numbers, data and a lot more examples.

Traditional programming vs machine learning

In traditional programming, a developer writes the rules and the computer follows them: data and rules go in, answers come out. That works brilliantly for things like calculating tax or sorting a list.

But some problems are almost impossible to write rules for. What exact rules would recognise every handwritten “7”, or every spam email, including the clever new ones? In machine learning, you flip it around: you give the computer data and the answers, and it produces the rules, which we call a model.

How machine learning works, step by step

Almost every ML project follows the same basic recipe:

1. Collect examples (data). For a spam filter: thousands of emails, each marked spam or not spam.
2. Pick the clues (features). Things the model can measure, like the number of links, words in ALL CAPS, or whether the sender is in your contacts.
3. Train the model. The model makes a guess for each example, checks how wrong it was, and nudges its internal settings to be a little less wrong next time. Repeat, many times over.
4. Test it on new data. Keep some examples hidden during training, then check how well the model does on them. This is the honest score.
5. Use it to predict. Once it's good enough, the model sorts brand-new emails it has never seen.

Try it: train a tiny spam filter

Below, each dot is an email, placed by two clues: how many links it has and how many words are in ALL CAPS. Red dots are spam, green dots are not. The coloured background shows what the model currently predicts for any new email in that spot.

Click the chart to add your own examples and watch the model's decision change. This model uses a simple method called k-nearest neighbours: it labels a new email the same way as the few examples closest to it.

Notice two things. First, more (and more varied) examples make the boundary smarter. Second, a single odd example can create a strange little island when the model only looks at its one closest neighbour. That's a tiny taste of a real problem called overfitting, which we'll come back to.

The 3 types of machine learning

Machine learning comes in three main flavours, depending on what kind of examples and feedback the model gets.

1. Supervised learning: learning with answers

The model learns from labelled examples, where each example comes with the right answer. Emails marked spam or not spam, house listings with their sale prices, photos tagged “cat”. It's the most common type, and the demo above is supervised learning.

2. Unsupervised learning: finding hidden groups

Here the data has no labels. The model's job is to discover structure on its own, such as grouping customers with similar shopping habits or spotting a transaction that looks unlike all the others.

3. Reinforcement learning: learning by trial and error

The model (often called an agent) takes actions and receives rewards or penalties, like points in a game. Over many attempts it learns which actions lead to the best results. It's used for games, robotics and fine-tuning chat assistants with human feedback. If you're curious how agents act in the real world, see our guide on what an AI agent is.

Machine learning examples you already use

• Spam filters that keep junk out of your inbox.
• Recommendations for videos, songs and products based on what similar people enjoyed.
• Photo search that finds “beach” or “dog” in your gallery without tags.
• Maps that predict traffic and arrival times.
• Voice assistants that turn speech into text.
• Fraud detection that flags unusual card payments.
• Chat assistants like ChatGPT, Claude and Gemini.

Machine learning vs AI vs deep learning

These three terms get mixed up constantly. Picture three circles, one inside the other:

• Artificial intelligence (AI) is the big circle: any technique that lets computers do tasks that normally need human intelligence.
• Machine learning sits inside it: AI that learns from data instead of hand-written rules.
• Deep learning sits inside ML: machine learning with large neural networks, loosely inspired by the brain. It powers modern language models (see how ChatGPT works) and AI image generators.

The limits of machine learning

ML is powerful, but it isn't magic. A few honest limits every beginner should know:

• Garbage in, garbage out. A model only knows what its examples taught it. Incomplete or unfair data leads to wrong or biased predictions.
• Overfitting. Like a student who memorises last year's exam answers, a model can learn its training examples too precisely and then fail on anything new. Testing on unseen data catches this.
• It needs data, often a lot of it. Rare situations the model never saw can trip it up.
• It can be hard to explain. Big models often can't tell you why they made a particular prediction.

Where to go next

Now that you know how machines learn from examples, you're ready for the most famous example of all: how ChatGPT works. And once you start using AI tools, these prompt engineering tips will help you get much better answers.

Frequently asked questions