What AI actually is.
No maths and no jargon. Here's what's really going on when you type a question into a chatbot, and why it sometimes gets things confidently wrong.
- Today's AI chatbots work by predicting likely words, based on patterns learned from huge amounts of text. They don't look facts up or “know” things the way you do.
- That makes them fast and genuinely useful for drafting, summarising and explaining, and also able to be wrong in a confident, convincing way.
- You don't need to understand the technology to use it well. You need to know what it's good at, and to check what it gives you.
Prefer to watch?
The video covers the same ground as the page below, in a few minutes.
AI is a broad word for a simple idea
“Artificial intelligence”, or AI, is a broad label for computer systems that do things we normally think need human intelligence: understanding language, recognising images, spotting patterns or making predictions.
You've been using it for years without calling it AI. Your email's spam filter is AI. So is the text from your bank asking whether a card payment was really you, the sat nav working out the quickest route, and the streaming service suggesting what to watch next.
What changed recently is a new kind of AI that can write. Tools like ChatGPT, Microsoft Copilot, Google Gemini and Claude can draft an email, summarise a report or explain a spreadsheet formula in seconds, in ordinary language. When people at work say “AI” now, this is usually what they mean.
How a chatbot actually works
Chatbots run on something called a large language model, or LLM. “Model” just means a computer program that has learned patterns from examples. “Large language” means it learned from an enormous amount of text: books, websites, articles and more.
During that learning, called training, the model played one game billions of times: read some text, guess the next word, check the guess, adjust. Over time it got extremely good at predicting which words are likely to come next, in almost any style and on almost any subject.
So when you ask it a question, it isn't looking up an answer in a database. It's writing a reply one small piece at a time, each time choosing a likely next word based on everything that came before. It's a bit like the predictive text on your phone, scaled up enormously.
That sounds too simple to be useful, but predicting well means picking up a lot about grammar, facts, tone and how ideas connect. That's why the results can feel so human.
Why it can be confidently wrong
Because a chatbot predicts what a good answer sounds like, it can produce something fluent and believable that simply isn't true. It might invent a statistic, a quote, a policy or a reference. This is called a hallucination, and it happens with every tool, even the best ones.
The tricky part is that a wrong answer looks exactly like a right one. There's no change of tone and no warning light. The model isn't lying, because it has no idea what's true. It's just producing likely-sounding text. A few other things are worth knowing:
- It can be out of date. A model learns from information up to a certain date. Some tools can search the web to fill the gap, but not all do, and not every time.
- It doesn't know your organisation. Unless it has been connected to your company's documents, it knows nothing about your customers, prices or procedures. It will happily guess.
- It can be good at hard things and bad at easy ones. It might write a decent report, then miscount the items in a list. That's normal, and another reason to check.
What AI isn't
It helps to clear up a few common myths and worries.
- It isn't thinking or feeling. It can sound warm, confident or apologetic, but there's no understanding or intention behind the words. It's a very capable pattern machine.
- It isn't always right just because it's a computer. We're used to calculators being correct. Chatbots are different. Treat what they give you like a first draft from a quick, keen new colleague.
- It isn't one single thing. “AI” covers chatbots, image tools, forecasting systems, fraud detection and much more. Different tools do very different jobs.
- It isn't only for technical people. You talk to these tools in plain English. The skills that matter most are knowing your work, explaining what you need and spotting when something's off.
A few words you'll hear
You'll meet these terms again and again, so here they are in a line each.
- Generative AI. AI that creates new content, such as text, images or audio, rather than just sorting or predicting.
- Prompt. Whatever you type or say to an AI tool: your question or instruction.
- Agent. An AI tool that can take actions as well as answer, such as searching, filling in forms or working through a task in several steps.
- Agentic AI. The general name for AI that works through a task step by step, using other tools, rather than just answering: for example, going through a large set of data, spotting a pattern such as stock running low, and flagging or drafting what to do about it. A person should still check the important steps.
- Data lake. One big store where an organisation keeps lots of its data, in whatever shape it comes, so AI and reporting tools can draw on it.
- RAG. Short for retrieval-augmented generation: the AI looks up the right bits of your organisation's documents first, then answers from them.
- Copilot. A general name for an AI assistant built into software you already use, like email or spreadsheets. It's also the brand name of Microsoft's AI tools.
Spot the confident guess
- Open whichever AI tool your organisation has approved. If there isn't one, a free public chatbot is fine for this, because you won't be using any work information.
- Ask it about something you know really well that isn't confidential: a common process in your line of work, the rules of a sport you follow, or the history of your town.
- Read the answer slowly. Note what's right, what's vague and what's wrong or made up.
- Ask a follow-up: “How sure are you about that? Which parts might be wrong?” Notice whether it backs down, digs in or invents more detail.
- Keep the result in mind. That gap between “sounds right” and “is right” is the most useful thing to understand about AI.
Share it with your manager.
If your whole team is being asked to get up to speed on AI, a short practical session together saves weeks of guesswork.
About team training →Pass it on.
It's free, there's no sign-up, and it's written for people who aren't in IT.