What AI actually is (and what it is not)

You have heard that AI will transform your job, replace your job, or both. Neither claim helps you on a Tuesday morning with a stack of work in front of you. What helps is knowing what these tools actually do, because once you know that, it becomes obvious which of your tasks they can take and which they cannot.

This module takes about five minutes and gives you the mental model everything else in this course builds on. No maths, no jargon you have to memorise.

AI is not one thing

"AI" is a label stretched across dozens of unrelated technologies. The spam filter in your inbox is AI. So is the system that reads a chest X-ray. So is ChatGPT. They share almost nothing beyond the label. When someone tells you AI can or cannot do something, the first useful question is: which AI?

This course is about one specific kind: the one that arrived in 2022 and is the reason you are reading this. Large language models: ChatGPT, Claude, Gemini, Copilot. Tools you talk to in ordinary language and that answer in ordinary language.

What a language model actually does

A language model is a text-prediction engine trained on an enormous amount of writing. Give it some text and it produces the text that most plausibly follows. That is the entire mechanism. There is no database it looks things up in, no reasoning engine, no understanding in the way you understand this sentence.

This sounds reductive, and people often stop there and conclude the tools are worthless. That is the wrong conclusion. Predicting plausible text turns out to be startlingly powerful, because most professional work is text with a predictable shape: a lesson plan looks like a lesson plan, a discharge summary looks like a discharge summary, a polite decline looks like a polite decline. A machine that has read a million of each is genuinely useful at producing the next one.

The one distinction that matters

Almost every AI success and every AI disaster comes down to a single distinction: are you asking it to generate something, or to retrieve something?

• Generation: drafting, rephrasing, summarising, translating, brainstorming, restructuring, explaining a concept at a different level. This is what the tool is built for and it is genuinely strong.
• Retrieval: specific facts, figures, dates, citations, quotes, anything about your organisation or about recent events. This is where it will confidently make things up.

The trap is that both come back in the same fluent, confident voice. A model does not sound less certain when it is inventing. There is no tell. Building the habit of asking yourself "am I generating or retrieving?" before you trust an answer is probably the single highest-value thing in this whole course.

So what should you actually use it for?

Start with tasks where you can instantly judge the output, and where a blank page is the hard part. Rewriting something you already wrote. Turning rough notes into prose. Producing five variations of a paragraph so you can pick one. Explaining a concept three different ways so you can teach it. In all of these, you are the expert and the AI is the fast, tireless, mediocre first-drafter, which is exactly the right division of labour.