What AI is (and what it is not)

You probably use AI already (today)

You type "I'll be late for the bo…" on your phone and it suggests "boat". You open the translator to understand an email from a German tourist. You take a photo of your grandson and the camera draws a square around his face. The phone gallery groups the "beach" photos by itself.

All of this is artificial intelligence. You have been using it for years without calling it that.

So what is AI?

AI is a wide name. It covers programs that do things that used to seem to need a person: recognising a face, understanding what you say to your phone, translating, writing.

In this course we care about one type in particular: chat assistants, such as ChatGPT, Claude, Gemini or Copilot. You type a question, like a message, and they answer with text. Inside, every one of them has a language model.

A machine that guesses the next word

The main idea fits in one sentence: a language model is a machine that guesses the next word, trained on large collections of text and other data.

Imagine you had read millions of books, newspapers, websites, recipes and manuals. After that much reading, if someone said "Once upon a time…", you would know very well what usually comes next. The model does the same, but with more text than a person could read in a thousand lifetimes.

When you ask it "write an invitation to the Espírito Santo soup dinner (a traditional Azorean parish festival) in our parish", it does not fetch an invitation stored in a drawer. It writes one word. Then it picks the most likely word to follow. Then another, and another. Because it learned from so many invitations, the result looks as if it was written by someone who knows the subject.

What AI is NOT

This is where most people get it wrong. It is worth reading slowly.

• It is not a search engine. A search engine such as Google looks for pages that already exist and shows them to you. A language model writes a new answer. Some assistants can also go onto the internet, but that is an extra part, not the main engine.
• It is not a database. It has no file card with the pharmacy's opening hours or the parish council's phone number. If you ask, it writes what seems most likely. It may be right. It may not.
• It does not "think" or "know" the way a person does. It has no opinions, memories or will of its own. It writes like someone who knows, because it learned from texts written by people who did. Sometimes it gets things very right. Sometimes it is wrong, with the same confidence.
• It is not a robot. It has no body and does not walk around the house. It is a program that runs on computers, usually far away from you, and answers when you type or speak to it.

What has changed in recent years

AI has been around for many years. Some of it was made of rules written by hand by programmers, line by line. "If the email has the word 'prize' and a strange link, send it to junk." "If the caller presses 1, put them through to accounts." It worked well in the cases someone had thought of. It failed at everything nobody had thought of.

But for decades there have also been programs that learn from examples instead of following rules (this is called machine learning). Many spam filters and phone keyboards learned that way.

What is new is the scale and the free text. Today's models learned from enormous amounts of text and other data. They write new answers, in everyday language, about almost any subject. This is called generative AI: AI that creates new things, such as a letter, a summary or the image of a boat at sunset that never existed.

Why this matters to you

With this idea in your head, a machine that is very good at guessing the next word, you already know where it helps and where you should not trust it blindly.

It helps a lot with writing and rewriting, summarising, explaining difficult things in simple words, giving ideas and translating.

It does not replace the official source, the doctor, the lawyer, or your own head.