machine learning (ML)

basicwidespread

Machine learning is a method by which a computer learns how to solve a task not from pre-written rules but from many examples.

In plain terms

In traditional programming, a human writes step-by-step instructions for what the machine should do. In machine learning, this is reversed: you show the machine examples, usually with correct answers, and it finds the rule itself.

It is like a child learning to recognize a dog. No one tells them exactly how many legs it has or what its ears look like. They see many dogs and eventually recognize ones they have never seen before. The difference is that machines usually require far more examples than a child.

It is useful where rules are almost impossible to write by hand: handwriting, speech, or face recognition, filtering suspicious bank transactions, or predicting which film you will like. The vast majority of today's artificial intelligence applications are built on this, including chatbots and AI agents.

A key difference is that the trained system does not "know" the answer but estimates it, usually well but never with 100% certainty.

How it works

Data collection: gathering many examples for the task, such as thousands of emails, each labeled as spam or not.
Training: an algorithm goes through examples, guesses, and then gradually adjusts its internal numbers (parameters) based on its errors to make fewer mistakes next time.
Model: the end result of training is the model, meaning a summary of learned patterns that can now answer new, previously unseen cases.
Verification: the model is tested on examples it did not see during training, because only then can it be determined whether it truly learned the task or simply memorized the training material.
Usage and updates: the model operates live, but the world changes, so it must occasionally be retrained with new data.

An everyday example

When your email automatically moves suspicious messages to the spam folder, or your phone groups photos by face, machine learning is working in the background. No one told it what your grandmother looks like; it learned from previous images.

Who is affected

everyoneFrom search engines to maps to music recommenders, you use dozens of services daily that operate via machine learning and often learn from your data.

bank customerIn credit assessments and fraud detection, a trained model often provides a recommendation on whether you are considered reliable or if a transaction is suspicious.

patient, doctorIn the analysis of X-ray and CT scans, trained systems help spot what a tired eye might overlook, but the decision remains with the doctor. → Health

entrepreneur, executiveDemand forecasting, inventory planning, and customer churn estimation are now accessible tools even for smaller companies.

A common misconception

MythA machine learning system learns and understands the world like a human.

In realityThe machine searches for statistical patterns in data and does not understand the world as humans do. That is why it can be better than a human in some areas while making surprisingly stupid mistakes in unusual situations.

Limits and risks

It is only as good as its data: if the training examples are incomplete or biased, the model will be too, and it may even reinforce prejudices.
It works with probabilities, so it can be confidently wrong; do not accept its statements without verification for important decisions.
Often it is impossible to say exactly why it decided as it did, which poses a serious problem in cases of loans, admissions, or diagnoses.
Learned knowledge becomes obsolete: if habits or circumstances change, a model trained on old data silently performs worse over time.

Similar, but not the same

artificial intelligence – Artificial intelligence is the broader goal of enabling machines to solve tasks that appear intelligent; machine learning is currently its most important, but not only, method.
deep learning – Deep learning is a branch of machine learning that works with multi-layer neural networks and has brought today's breakthroughs in image, audio, and text processing.
traditional programming – There, a human writes the rules and the machine carries them out; in machine learning, the machine works out the rules itself from examples.

Related concepts

Broader

artificial intelligence, data science

Narrower

deep learning, supervised learning, unsupervised learning, reinforcement learning, Neural network

On the edge

modell, training data, algorithm, large language model, bias

A brief history

The term "machine learning" was popularized in 1959 by Arthur Samuel, an IBM researcher, when he wrote about a checkers program that played better by learning from the games it had played. In the 1990s and 2000s, machine learning became the main direction of artificial intelligence. The major breakthrough came in the 2010s: vast amounts of data and powerful graphics processors made large, multi-layered neural networks usable. This is called deep learning.

Where things stand

At the end of 2026, machine learning is the invisible foundation of digital services: it is in your phone, your bank and your search engine. Attention is focused on large language models and the agents built on them, but these work on the same principle.

Status: 2026 November

In the news

Appears in 9 Agent Age stories; the curve shows weekly occurrence in the news magazine.

FCA to launch quantum machine learning pilot group in 2027 to tackle fraud and money launderingNov 5, 2026 · Money
Multimodal delirium risk model validated externally and deployable without site-specific retrainingNov 5, 2026 · Health
Nvidia's Minerva AIOps software aims to predict failures in AI data centersOct 22, 2026 · Ipar
Diagnostic lab claim denials jump 40% in 2025, XiFin report findsOct 8, 2026 · Health
NAIC expands insurer AI evaluation pilot to machine learningSep 17, 2026 · Money
Hungarian tax authority NAV's AI and machine-learning developments affect cash-book checksSep 15, 2026 · Money
AI model helps western Sydney emergency departments spot sepsis earlySep 7, 2026 · Health
Semmelweis University workshop on machine learning in public health under HAI-SO projectSep 1, 2026 · Health

Also known as

machine learning · machine learning · ML · machine learning model · learning algorithm · self-learning algorithm · statistical learning · ML-modell

Sources

Wikipedia: Machine learning

This entry was machine-edited and has undergone machine expert review; the “In the news” section updates automatically from Agent Age stories.