What artificial intelligence actually is

From the history you know which idea won: don't program in knowledge, teach from examples. Let's unpack what that means in practice — it's the foundation of the whole course.

A regular program: a world of hard rules

A classic program is a set of instructions written by a person: "if the button is pressed, open the window", "if the total is over 1000, ask for confirmation". A program will never do anything that isn't in its rules. That's its strength (predictability) and its ceiling: try writing rules that tell a cat from a dog in a photo. Ear shape? Varies by breed. Whiskers? Not always visible. People tried for 30 years and got nowhere.

AI: a world of examples

Machine learning flipped the approach: instead of a thousand rules for "what a cat looks like", you show the model a million photos with the answers — cat / not cat — and it finds the patterns itself: combinations of shapes, textures and proportions a person couldn't even put into words. That's what machine learning is — the foundation of all modern AI.

You've been using AI for years — you just didn't notice

• 📱 Face unlock — a neural network recognises you even though you look slightly different every day.
• 🎵 Music and video recommendations — a model that learned your taste from your listening history.
• 📧 The spam folder — AI trained on billions of emails: rules can't catch spam, it mutates.
• 🗺️ Traffic in your maps app — a prediction built from thousands of cars' data.
• 📷 Your phone camera — night mode and portrait blur are neural networks too.

Notice what they share: every one is a task where you can't write the rules, but collecting examples is easy.

Chess versus Go: the two approaches, side by side

In 1997 the Deep Blue computer beat the world chess champion Garry Kasparov — and that was a triumph of the old approach: an army of programmers and grandmasters loaded chess knowledge into the machine by hand, and from there it won by churning through millions of positions a second. The trick didn't work for Go: it has more positions than the universe has atoms, brute force is helpless, and not one master could put "the rules of a good move" into words.

So Go stood unbeaten for twenty years — until AlphaGo learned it from millions of games and from playing against itself. One frontier fell to rules, the other only to learning. Keep this pair in mind: it tells you instantly which approach fits where.

Three kinds of AI you'll run into

• Recognising: looks at data and says "what this is" — a face in a photo, spam in an email, a defect on a production line.
• Predicting: estimates "what will happen" — traffic an hour from now, a customer about to leave, the next word your keyboard suggests.
• Generative: makes something new — text, images, music, code. This is the one that exploded in 2022, and most of this course is about it.

Common myths (let's clear them now)

• "AI means robots." A robot is a body, AI is the "brain". Almost all modern AI lives with no body at all: in your phone, your browser, a server.
• "AI thinks like a human." It doesn't: it finds statistical patterns. Sometimes the result is indistinguishable from thinking, but the mechanism is different — and we'll get into the differences (hallucinations, for one).
• "AI has goals and desires." A model has a mathematical objective that people gave it. Worries about AI are worth taking seriously, but start from facts, not movie plots.

The core formula of this course

Learn one contrast — it explains most of how AI behaves:

A regular program: a person writes the rules → the machine follows them → predictable, but limited to what was written.
AI: a person gives examples → the machine works out the rules itself → it handles tasks that have no rules, but its conclusions are statistics, not a guarantee.

Both sides of the coin come from that. The superpower: AI solves problems nobody could program (speech, vision, translation, conversation). The weakness: it can be wrong confidently and without warning — its "rules" were pulled out of examples, and on rare or new cases the statistics let it down.

Do this now: your first real result in a minute

Enough theory — let's put hands on AI. You don't have to go anywhere: there's a Run button under the prompt and a real AI answers right here. Change one word on the first line — your own job — and run it:

You are an experienced assistant. I'm a marketer.
(swap in your own: accountant, teacher, chef, driver — anything)

Do three things:
1. Name 5 typical work tasks of mine where you could help.
2. For the most common one, show a finished example: a short text or plan you'd produce.
3. Ask me 2 clarifying questions to make the result more accurate.

In under a minute you'll have an answer tailored to your job — that's the first real result this whole course is for. Don't like the answer? Add "make it shorter" or "give me a different example" and send again — AI will happily redo it. You just applied everything we talked about: you gave the model examples to aim at (a role and a task), and it worked out the rules of the answer itself.

Can't see the button? You're not signed in yet — signing up is free and takes one click. And in module three we'll set you up with your own AI access outside the course, step by step.