What Is Machine Learning? How Computers Learn from Data

These systems learn by finding patterns in large amounts of data. The core idea: show the algorithm enough examples, and it will learn to generalise.

Training data

The system needs a training dataset — a large collection of examples from which it learns. To train a machine learning model to recognise cats, you feed it thousands of labelled images: 'this is a cat', 'this is not a cat'. It adjusts its internal parameters until it can reliably distinguish cats from non-cats.

Types of machine learning

Supervised learning trains on labelled data (each example has a known answer). Spam filters use this approach — trained on emails labelled 'spam' or 'not spam'.

Unsupervised learning finds patterns in unlabelled data. An unsupervised algorithm might cluster customers into groups based on buying behaviour, without being told what the groups should be.

Reinforcement learning learns by trial and error, receiving rewards for correct actions. AlphaGo — the system that beat the world's best Go players — was trained this way. What is artificial intelligence covers the broader field of which machine learning is a part.

Machine learning is already embedded in tools most people use every day, often invisibly.

Recommendation systems

Spotify, YouTube, Netflix, and Amazon all use machine learning to recommend content. The machine learning algorithm analyses what you have watched, liked, or bought, finds users with similar tastes, and predicts what you might enjoy next. These recommendation tools are so effective that they have reshaped entertainment consumption.

Language and speech

Translation tools, voice assistants (Siri, Alexa, Google Assistant), and autocomplete all rely on machine learning trained on vast amounts of text. Large language models — the systems behind AI chatbots — learn statistical patterns in language from billions of documents, allowing them to generate human-like text.

Healthcare and science

These tools can detect cancers in medical images with accuracy matching or exceeding specialist doctors. It is used to predict protein structures (a breakthrough in biology), identify new drug candidates, and forecast patient outcomes. Such technology is accelerating scientific discovery in ways that would have been impossible a decade ago.

Finance

Banks use these tools to detect fraudulent transactions in real time — flagging unusual patterns in milliseconds. Credit scoring, stock trading, and insurance pricing all use machine learning models to make better predictions from complex data.

Machine learning is powerful — but it raises serious challenges that society is still working out.

Bias in machine learning

These systems learn from data created by humans — and human data reflects human biases. A hiring tool trained on historical hiring data may discriminate against women or minorities if those groups were historically under-hired. Facial recognition tools have been shown to perform worse on darker-skinned faces. Addressing bias requires careful data collection, auditing, and diverse teams.

Explainability

Many such models are 'black boxes' — they produce answers without explaining their reasoning. A trained system might correctly identify a skin cancer from an image, but not be able to say why it decided that. For high-stakes decisions — medical diagnosis, loan applications, criminal sentencing — explainability matters enormously.

Jobs and society

This technology is automating tasks previously done by humans — from data entry to some forms of legal and medical work. This creates economic disruption alongside new opportunities. Understanding what these tools can and cannot do — and how to work alongside it — is becoming an essential skill. Coding and the internet are the foundations on which machine learning tools are built.