Skip to content

AI explained in plain English. Read the latest articles

AI explained Beginner

What is machine learning? Explainer with everyday examples

Machine learning is the technique that lets computers learn from examples. You see it in spam filters, recommendations and ChatGPT.

A person looks at patterns in data on a laptop, in an explanation of what machine learning is

What is machine learning if you look at it in everyday life? Your inbox filters spam without anyone ever asking you for rules, a streaming service makes recommendations, and your phone corrects your typing mistakes.

No one wrote all those rules one by one. The computer has drawn patterns from examples. After this article, you can explain machine learning to a colleague in two sentences.

What is machine learning?

Machine learning is a way of letting computers learn from examples. That helps them recognise patterns and make predictions, without every rule being programmed in advance by a person.

Think of a child learning what a cat is. You show lots of photos of cats. After a while, the child can also recognise a cat it has never seen before.

That is different from a list of fixed rules. You could say: a cat has four legs, whiskers and pointed ears. But that quickly goes wrong. Some cats are curled up, missing a leg or partly out of frame.

With machine learning, you give examples. The computer looks for features that often appear together. That is how a model is created.

A model is not a person and not a brain. It is the outcome of the learning process. The model uses what it has learned from examples to make a new judgement.

Part Traditional software Machine learning
How the rules are created A person writes the rules in advance. The system learns patterns from examples.
What it needs Clear instructions and fixed steps. Lots of good examples.
Can it be wrong Yes, if the rule does not fit the situation. Yes, because it makes a prediction.
Example If an amount is above a threshold, show a warning. Recognise whether an email is likely to be spam.

Machine learning is described as part of artificial intelligence in which a computer keeps improving at a task by learning, according to Toolshero. The key word here is learning. Not the way a human learns, but by finding patterns in data.

How does a computer learn from examples?

A computer learns by getting lots of examples, looking for patterns in them, testing the result and then using it on new examples. The outcome remains a prediction, not a certainty.

The process sounds more complicated than it is. In plain language, it has three steps.

  1. Collect data. You gather examples. For a spam filter, these are emails that are or are not spam.
  2. Train. The system looks for differences and similarities. For example, it looks at words, senders and links.
  3. Test and use. You check whether the model also makes good choices with new examples it has not seen before.

The quality of the examples matters. If the examples are messy, wrong or one-sided, the model learns the wrong patterns. Then the result may look neat, but still be wrong.

A model also does not know for sure whether something is correct. It gives an answer that seems likely. That is useful for quick decisions, but risky if you treat it as the truth.

That is why control is needed for important decisions. Think of work, money, healthcare, job applications or credit. Make sure a person can look in and step in.

What types of machine learning are there?

There are roughly three types of machine learning: learning with examples where the right answer is known, learning without answers, and learning through reward and punishment. The technical English terms exist, but you do not need to remember them to understand the idea.

The first type is learning with the right answer included. This is called supervised learning. A spam filter, for example, is shown emails that people have marked as spam, and emails that are just normal mail.

The second type is learning without pre-given answers. This is called unsupervised learning. A system can, for example, group customers by buying behaviour without anyone saying in advance what name each group should get.

The third type is learning through reward and punishment. This is called reinforcement learning. A system learns a game, for example, by trying moves and seeing whether winning or losing gets closer.

Type In plain language Everyday example
Learning with answers The system sees examples with the correct label. Recognising spam in your email.
Learning without answers The system finds groups or patterns by itself. Grouping customers by buying behaviour.
Learning through reward The system tries actions and learns from success or failure. A system learning a game by winning and losing.

In practice, tools can combine several techniques. For you as a user, what matters most is what the tool does, which data were used, and how errors are checked.

Where do you encounter machine learning every day?

You come across machine learning in spam filters, recommendations, speech recognition, navigation, fraud detection at banks and photo apps. Often you do not even notice it, because it is simply built into an existing app.

A well-known example is your email programme. Many email programmes can move messages into your inbox or Junk folder. We do not need to know exactly how each company does this technically to understand the basic idea: the system uses patterns to recognise spam.

  • Spam filters. Your email programme estimates whether a message is advertising, a scam or ordinary mail.
  • Recommendations. Streaming services suggest things based on what you watched or listened to before.
  • Typing mistakes on your phone. Your phone suggests corrections when you type a word wrongly.
  • Navigation. Route planners use patterns and current information to suggest a route.
  • Fraud alerts at your bank. A bank can spot unusual payment behaviour and ask for extra checks.
  • Photo apps. A photo app can group faces, animals or places.

So machine learning is not just something for new AI chatbots. It has been part of everyday digital services for years. Often it is invisible until it makes a mistake.

That mistake may be small, such as a strange music suggestion. But sometimes it is annoying, for example if a real customer email ends up in the Junk folder.

What is the difference between machine learning and AI or ChatGPT?

AI is the umbrella term, machine learning is one way to build AI, and ChatGPT is a language model trained with machine learning. You can think of them as circles that fit inside one another.

The largest circle is AI. If you want the broad basics, read What is AI. Inside that sits machine learning: systems that learn from examples.

Inside that also sit forms of generative AI. That is AI that can create new text, images or other content. ChatGPT is an example of a language model.

But machine learning is broader than ChatGPT. A spam filter, route planner or fraud alert can also use machine learning. They do not necessarily produce long texts, but they do make predictions.

There is also AI with fixed rules. In that case, the system does not learn from examples, but follows pre-set rules. That can be useful when the situation is simple and easy to describe.

So the short version is: AI is the big field. Machine learning is an important way to build AI. ChatGPT is one well-known application within that larger whole.

What are the risks and limitations of machine learning?

A model can make mistakes and turn out unfair if the training data are one-sided or wrong. The result is a probability, not the truth.

One risk is bias. That means a system picks up skewed or unfair patterns. If job application software learns from a one-sided past, for example, it can repeat old choices instead of judging fairly.

Another risk is that people trust the computer too much. Because advice looks neat, it can seem reliable very quickly. Yet the model can miss something, misread something or sound too confident.

With text AI, this can also lead to made-up or wrong answers. Read more about that in AI mistakes and hallucinations. The lesson is the same: checking is still needed.

Privacy is also an issue. The GDPR applies as soon as personal data are used to train or run models. Personal data are data that relate directly or indirectly to a person.

There is also the European AI Act. It defines an AI system as a machine-based system that operates with varying levels of autonomy and infers from input how to generate output such as predictions, content, recommendations or decisions, according to the explanation by Orrick and the official text at EUR-Lex.

For employees and business owners, the main rule is simple. If a tool decides something about people, money or opportunities, you should be extra critical. Ask how errors are spotted and who is ultimately responsible.

What can you do with it as an employee or business owner?

You do not need to build machine learning yourself to use it well. It does help if you recognise where it is used and what questions you should ask.

Use this checklist when you look at a new tool. That could be an email tool, accounting package, advertising platform or AI assistant.

  • What was it trained on? Ask what kind of examples were used.
  • Does the provider use my data for training? Also ask whether you can switch that off.
  • How accurate is it? Ask who checks errors and what happens when something goes wrong.
  • Does it make decisions about people? Think of hiring or credit. Then make sure there is human oversight.
  • Where is the data stored? Ask whether the data are in the EU or outside it.

For use, you usually do not need to programme. If you want to build something yourself, technical knowledge is needed. But in many cases you use ready-made tools that already do the job.

What is machine learning in practice?

A spam filter shows very clearly what machine learning does in practice. It learns from many earlier emails and uses that experience to judge new emails.

Imagine: a Dutch online shop owner is self-employed and runs a small clothing webshop. She wonders why her email programme automatically puts some messages into Junk.

  1. The spam filter has seen millions of emails that people marked as spam.
  2. It learns to recognise signals such as senders, words and links.
  3. When she clicks This is spam, that can be a new example.
  4. The filter can then improve for her situation.
  5. Even so, it may miss a real customer email, so she checks the Junk folder once a week.

This example shows both the strength and the limitation at once. The filter takes work off her hands, but it is not foolproof. That is why a quick check remains sensible.

Do you want an AI chatbot such as ChatGPT or Claude to explain this to you? Use this practice prompt.

The expected result is a short explanation in simple language. If all goes well, the chatbot will also ask a question at the end to check whether you understand.

Common mistakes about what machine learning is

The biggest mistake is thinking that machine learning is the same as human understanding. It can be useful, but it does not understand the world like a person does.

  1. Thinking machine learning is the same as ChatGPT. ChatGPT is one application. There is also machine learning in spam filters and route planners.
  2. Thinking a computer understands or reasons like a person. A model recognises patterns in data and makes predictions. It does not have human understanding.
  3. Thinking the result is always correct because it comes from a computer. Treat the result as a prediction. Check important matters.
  4. Thinking the model always keeps learning from your use. Many models do not learn live. Check the provider’s settings for the use of your data for training.
  5. Thinking you need to be able to programme. You usually do not need that to use it. You do need it to build it, but ready-made tools often do the work.

A good attitude is practical. Use machine learning where it helps, but keep asking questions. Especially if the result affects people.

Frequently asked questions

Here are short answers to questions that often come up when you are just starting with machine learning.

Is machine learning the same as AI?

No. Machine learning is one way to build AI. AI is the broader term for systems that carry out tasks that normally require human judgement.

Is ChatGPT machine learning?

Yes. ChatGPT is a language model trained with machine learning. So it is an example of machine learning, but not the whole field.

Do I need machine learning in my company?

You are probably already using it in tools such as email filters. Building it yourself is usually not necessary, but you do need to know what questions to ask suppliers.

Does machine learning learn from my data?

That depends on the tool and the settings. Check with the provider whether your data are used for training and whether you can switch that off.

Do I need to be able to programme?

No, to use machine learning you usually do not need to programme. To build it yourself, technical knowledge is needed.

The next step is to understand how machine learning leads to chatbots such as ChatGPT. So read What is a language model and try the practice prompt above to have it explained back to you in simple words.

Was this article helpful?

New articles every day

Follow AIzine and never miss an article

We publish three new articles about AI every day. Follow us on Instagram or Facebook to see them in your feed.