Short answer: AI is software that learned how to do something from examples, instead of a programmer writing out every rule. Based on what you give it, it produces a prediction, a piece of text, a recommendation or a decision.
The official definition
The EU AI Act describes an AI system as a machine-based system that works with some degree of autonomy and infers, from the input it receives, how to generate outputs such as predictions, content, recommendations or decisions. Those outputs can influence the physical or digital world.
The key word is infers. A formula in Excel infers nothing: it does exactly what you typed. An AI system works out what the answer probably should be, based on what it has learned.
AI, machine learning and deep learning
These three terms get mixed up a lot. Here is how they fit together:
- AI is the umbrella term. It dates from 1955, when John McCarthy and colleagues used it in a proposal for a research project at Dartmouth College.
- Machine learning is the approach where a system gets better at a task by looking at examples and data. According to Stanford's AI100 study, it has been the main driver of AI's growth over the past decades.
- Deep learning is a kind of machine learning that uses large neural networks: computing models loosely inspired by brain cells. Instead of writing a new algorithm for each problem, you let the network shape itself around the data you feed it.
What is generative AI?
Generative AI is AI that makes something new: text, images, sound or code. The Dutch Data Protection Authority (Autoriteit Persoonsgegevens) describes it as AI that can generate new data, where the most popular tools produce text and images you can no longer tell apart from human work.
ChatGPT, Claude, Gemini and Grok are examples of generative AI. They are built on large language models. To see how those work, read How does a language model work?
What does this mean for you?
AI is not magic, and it is not a person. It is a tool that recognises patterns and builds on them. That makes it strong at tasks with lots of text or repetition, and weak where you need the one correct answer and nobody checks it.
Sources
- AI Act Service Desk (European Commission): Article 3, definitions
- A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence (1955)
- Stanford AI100: A Short History of AI
- OpenAI: Introducing OpenAI (2015), on deep learning
- Autoriteit Persoonsgegevens: Hulpmiddel inzet generatieve AI & de AVG (PDF, Dutch)
Accessed 21 October 2026.
Quiz
Test yourself: what is AI?
Five short questions about this article. You see straight away whether your answer is right.
0 of 5 answered
Your answers are not stored anywhere.