Machine Learning

Machine learning (ML) is the branch of artificial intelligence that gives software the ability to improve through experience rather than through explicit programming. Instead of a developer writing rules for every situation, a machine learning system is given a large set of examples and tasked with finding the patterns that connect inputs to outputs. Once trained, it can apply those patterns to new data it has never seen before. This is fundamentally different from traditional software, where the programmer must anticipate and encode every decision in advance.

The practical consequence is that machine learning can tackle problems that are far too complex to programme by hand. Recognising a face in a photograph, predicting whether a loan applicant will default, or detecting a fraudulent transaction all involve thousands of subtle variables. Writing rules for each variable would be impossible; learning them from millions of examples is tractable.

How it works#

Machine learning follows a broadly consistent process. First, you collect training data: a large set of examples with known outcomes. A fraud detection system, for instance, might be trained on millions of past transactions, each labelled as legitimate or fraudulent. The model processes these examples, adjusting its internal parameters until it can reliably distinguish between the two categories. Once the model performs well on examples it has not seen before, it is deployed to make predictions in the real world.

There are several main varieties. Supervised learning uses labelled data, where each example has a correct answer attached. Unsupervised learning looks for natural groupings in data that has not been labelled. Reinforcement learning trains a model by rewarding it for good decisions and penalising it for poor ones, a technique used in game-playing AI and robotics.

Examples#

Fraud detection. UK banks use machine learning to flag suspicious card transactions in real time. The model has been trained on historical transactions and has learned the patterns that distinguish genuine purchases from fraud. It can assess thousands of signals simultaneously, things like the time of day, the merchant category, and the cardholder's usual behaviour, in a fraction of a second.

Demand forecasting. Supermarkets and logistics companies use machine learning to predict how much stock they will need at each location. The model considers seasonal trends, local events, weather forecasts, and historical sales data. More accurate forecasts reduce waste and prevent out-of-stock situations, two significant cost savings for any retailer.

Why it matters#

Machine learning is the engine behind most of the AI capabilities that businesses and consumers interact with today. Recommendation engines, spam filters, translation services, voice assistants, and predictive maintenance systems all depend on it. As tools become more accessible, small and medium-sized organisations can now access machine learning capabilities that were previously available only to companies with large technology budgets.

For a clear comparison of AI, ML, and related ideas, read our tutorial AI vs ML vs Deep Learning, or visit the glossary for definitions of related terms like deep learning and neural networks.