In plain terms
An umbrella term, and no single technology. It covers the spam filter in your inbox, the route your navigation app picks, the system that flags a suspicious card payment and the assistant that drafts your email. What they share: each does something that used to need a person's judgement.
Why it matters
The word is used so widely that it hides more than it shows. When a supplier says “AI-powered”, the useful question is which kind: a set of rules, a prediction model trained on data, or a language model? They differ in cost, in risk, in the data they need and in the mistakes they make. Most of what is called AI today is generative AI, which is only the newest layer of a seventy-year-old field.
Example
A bank uses three kinds of AI on one loan application. A rule engine checks that the documents are complete. A machine-learning model trained on past loans estimates the probability of default. A language model drafts the letter to the customer. All three are sold as AI; they are built, tested and supervised in three different ways.
Most often confused with
AI vs. Machine learning
Machine learning is the dominant way of achieving AI today, which is why the two words are used as synonyms. They are not. AI also includes systems that follow hand-written rules or search through possibilities, with no learning at all. All machine learning is AI; not all AI is machine learning.
Origin: The term was coined by John McCarthy in 1955, in the proposal for the Dartmouth workshop held the following summer.
Under the hood
Two historical traditions: symbolic AI (hand-written rules, logic, search, expert systems) and statistical AI (machine learning and, since the 2010s, deep learning). The field has moved through cycles of optimism and “AI winters”; the present wave rests on three things arriving together: very large datasets, GPU computing and the transformer architecture. Useful working distinctions: narrow systems built for one task versus general-purpose models; predictive AI (classify, forecast) versus generative AI (produce content); and assistive tools versus agents that act. Legal definitions matter as well: the EU AI Act defines an AI system by its ability to infer, from the input it receives, outputs such as predictions, content, recommendations or decisions, with some degree of autonomy.