What AI actually is, in plain terms

What you'll be able to do afterwards: Explain to a colleague, in one sentence, what a chatbot is doing — and what it cannot do.

Everything in this course rests on one idea, and it fits in a sentence: a language model is a very good guesser of what word comes next.

That is the whole mechanism. You give it some text. It calculates which word is statistically most likely to follow, writes that word, then repeats the process with everything so far. A paragraph is a few hundred of those guesses in a row. There is no lookup of facts, no database of truths, and no internal model of the world being consulted. There is a guess, then another guess.

Once you hold that idea, the rest of how these systems behave stops being mysterious.

Why it seems to know things

Because language is full of facts. If the text so far is "The capital of France is", the next word is almost always "Paris" — that is simply what tends to follow those words in written English. The model produces it for the same reason it produces any other word: it is the likely continuation.

This works astonishingly well, which is why the technology is worth your attention at all. But notice the shape of it. The model is not retrieving an answer. It is continuing a pattern. Where the pattern is strong — capitals, standard definitions, common code, ordinary business writing — the continuation is usually right. Where the pattern is weak or the specifics are rare, the model still produces something, because producing nothing is not one of the options.

That is the source of both the power and the danger, and it deserves its own lesson. Lesson 5 covers it properly.

What it has no memory of

By default, a conversation is the only memory. Close the window and it is gone. This trips people up constantly in the first week. You explain your business on Monday, come back Tuesday, and it has no idea who you are.

The workarounds are worth knowing because they shape what you can build:

• Repeat the context. Clumsy but honest, and it works. This is most of what "prompting" is.
• A saved instruction set. Most tools let you store standing instructions — who you are, how you write, what to avoid. This is the single highest-value thing you can set up.
• Attached documents. Paste or upload the relevant file each time. The model reads it as part of the text it is continuing, which is why it can answer questions about a document you gave it and not about a document you did not.

There is no fourth option where it just remembers, unless the tool explicitly offers memory — and when it does, you should read what it stores.

It has no idea what is true

This is the part people resist. The model has no mechanism for checking whether what it wrote is true. It has a mechanism for making the next word likely. Those are different things, and the gap between them is where every AI embarrassment lives.

A model will produce a confident, well-formatted, plausible paragraph citing a report that does not exist, because that paragraph is a likely-looking continuation of your question. It is not lying. Lying requires knowing the truth.

The one habit that protects you

Treat a model like a bright, extremely fast, extremely well-read junior colleague who has never once said "I don't know" and who is unfamiliar with your business.

You would not take that colleague's numbers to a client without checking. You would not hand them something you couldn't verify. But you would absolutely let them draft, summarise, reformat, brainstorm and organise — because they are fast, and you are going to read it anyway.

That is the working posture: let it produce, and keep the judgement. Every lesson that follows is a refinement of that sentence.

Do this now

Open your usual assistant and ask it one question you already know the answer to — something specific to your own work. Watch it answer confidently and, very likely, get some detail wrong. Not because it is bad, but because it has no way to know your specifics.

That small experiment is the whole foundation. You have just seen the mechanism.

Next: lesson 2, where you learn to ask in a way that gets you something usable.