When people say "AI" in 2026, they're almost always talking about one specific kind of software: a program that finds patterns in large amounts of data, and then uses those patterns to make predictions or generate new content it was never explicitly told to produce.
That's different from older "AI" in movies or games, which was usually just a fixed set of if-this-then-that rules written by a programmer. Modern AI learns its own rules from examples, rather than having every rule hand-written.
The three ideas worth separating
- Artificial Intelligence (AI) — the broad goal: software that performs tasks we associate with human intelligence (recognizing images, understanding language, making decisions).
- Machine Learning (ML) — a specific technique for reaching that goal: instead of programming rules directly, you show the system many examples and let it work out the patterns statistically.
- Generative AI — a category of ML models built specifically to produce new content (text, images, audio, code) rather than just classify or predict a number.
Every tool you've probably heard of — chatbots, image generators, code assistants — is generative AI, which is itself a kind of machine learning, which is itself a kind of AI.
A worked example
Say a bank wants software that flags unusual card transactions.
The older approach was to write the rules by hand: flag anything above a set amount, flag anything from a country this customer has never used before. A person decided each rule and the software followed it exactly. That is AI in the broad sense — software doing a job that used to need human attention — but nothing in it learned anything.
The machine learning approach is different. You take a large set of past transactions, each already marked as ordinary or fraudulent, and let the system work out for itself which combinations of features tend to accompany fraud. Nobody writes the rule down. The system arrives at something closer to a weighting, and it can pick up combinations a person would not have thought to look for.
Generative AI is a third thing again. Ask a chat tool to explain the flagged transaction to the customer in plain language and it produces a paragraph nobody wrote in advance. It is not sorting things into categories; it is composing.
Same problem, three very different pieces of software. Being able to say which one a product is using tells you most of what to expect from it.
Where the labels break down
Two honest warnings before you go further.
First, "AI" is a marketing word at least as often as a technical one. Plenty of products described as AI are the hand-written rules of the first example with a newer label on the box. That is not necessarily bad software, but it is worth knowing which one you are buying.
Second, the boundaries are not clean. A real product usually combines all three: fixed rules for the parts that must never vary, a learned model for the pattern-matching, and a generative model for the text you end up reading. Asking "is this AI?" is usually less useful than asking which part of the system is doing the work you actually care about.
Why this matters before you go further
A lot of confusion about AI comes from treating it as one single, all-knowing thing. In reality, each system is trained on a specific dataset, for a specific purpose, with real limitations. Keeping "AI → ML → Generative AI" straight will make every later lesson in this course click faster, because we'll keep referring back to which layer we're talking about.
Go deeper: AI in the glossary · How does AI actually work? · LLM vs AI vs machine learning
Key takeaway: AI is the destination, machine learning is the road, and generative AI is one specific vehicle on that road.