AI Search 101

What is machine learning?

Machine learning is the branch of artificial intelligence where a system derives rules from data instead of being given them. Shown enough examples, it finds the pattern that connects inputs to outputs, and applies that pattern to cases it has never seen.

The basics

Abbreviation ML
Part of Artificial intelligence
Main types Supervised, unsupervised, reinforcement learning
Subfield Deep learning, using neural networks
Requires Data, in quantity and of adequate quality
Produces A model, meaning a set of learned parameters

How does machine learning work?

The core idea is to replace instruction with example.

Conventional software encodes a rule a person worked out. To flag fraudulent transactions, someone writes conditions: over this amount, from this country, at this hour. The system does exactly what it was told and nothing more.

Machine learning inverts this. Show the system a large set of transactions already labelled fraudulent or legitimate, and let it find what distinguishes them. It adjusts internal parameters until its predictions on the examples are as accurate as possible, then applies those parameters to new transactions.

The advantage is that the resulting pattern can be far more subtle than anything a person would write down, and it can involve interactions across dozens of variables at once. The cost is that the pattern comes from the data, so the data determines what is learned, including its errors and biases.

What are the main types?

Supervised learning uses labelled examples. Each input comes with the correct answer, and the system learns the mapping. Most commercial machine learning is supervised: churn prediction, credit scoring, image classification, demand forecasting.

Unsupervised learning has no labels. The system finds structure on its own, typically by grouping similar items. Customer segmentation and anomaly detection work this way.

Reinforcement learning learns from consequences. The system acts, receives a reward or penalty, and adjusts to earn more reward. It suits sequential decisions such as game playing, robotics and some optimisation problems, and it is part of how conversational assistants are tuned to be helpful.

What is overfitting?

The most common failure, and worth understanding because it explains why a model that looked excellent in development disappoints in production.

Overfitting is memorising rather than generalising. Given enough capacity, a system can learn the training examples perfectly, including their noise and their accidents. It scores brilliantly on data it has seen and poorly on anything new, because it learned the particular cases rather than the underlying pattern.

The defence is to evaluate on data the model never trained on. A model that performs well on held-out data has probably learned something real. A model evaluated only on its training data has told you nothing.

What does machine learning need to work?

Enough data, which usually means more than expected and in a consistent format.

Representative data. A model learns the world its training set describes. If that set under-represents a group, a market or a condition, performance there will be worse, and the failure is often invisible until it matters.

A stable relationship. Machine learning assumes the future resembles the past. Where behaviour shifts, models degrade quietly, which is why monitoring after deployment matters as much as accuracy before it.

A tolerable error rate. Every model is wrong sometimes. The question is whether the cost of being wrong is acceptable, and who notices.

How machine learning is used in marketing

Production. Personalising which message a given person sees, and generating variants at a scale worth testing.

Analysis. Segmentation, propensity and churn scoring, lifetime value estimation, attribution modelling, and forecasting. This is the oldest and most established use.

Distribution. Bidding and audience selection inside advertising platforms, and the ranking and recommendation systems that decide what reaches whom.

How this impacts your business

A language model does not look a company up in a register. It composes a description from everything it read during training and whatever it retrieves at the moment of the question. Trade press, directories, forums, competitor comparison pages, and somewhere in that mix the company's own website too.

So the description a buyer sees is assembled, not published. No company controls it, but every company supplies a share of the material it is built from.

The traffic impact is still small, roughly one percent of sessions for most companies. The influence is not small, because the reader is often a buyer forming a first impression of what a company does and who it serves, before any page gets a click.

Checking takes ten seconds. Type the company name into ChatGPT and read the answer.

Frequently asked questions

What is the difference between machine learning and deep learning?
Deep learning is machine learning using neural networks with many layers. It handles unstructured data such as text and images far better, at the cost of needing much more data and computation.

How much data does machine learning need?
It depends on the problem. Simple tabular problems can work with thousands of examples. Language and image models need volumes measured in millions or far beyond.

Is machine learning the same as statistics?
They overlap substantially. The emphasis differs: statistics traditionally prioritises explanation and inference, machine learning prioritises predictive accuracy on unseen data.

Can a machine learning model explain its decisions?
Some can. Simpler models are interpretable by construction; large neural networks are not, and explanations produced after the fact are approximations rather than accounts of what happened.

Why do models get worse over time?
Because the world changes while the model does not. Behaviour, products and competitors shift, and a model trained on last year's patterns gradually stops matching this year's.

Related definitions

Written by Lari Numminen, Generate More. Also available in Finnish. Updated 31 August 2026.