What is artificial intelligence and how does a chatbot work? Explained simply
The difference between AI, machine learning and deep learning, how ChatGPT or Gemini write word by word, and the key moments in AI history. No jargon.
We use the term artificial intelligence for almost everything: your phone’s assistant, spam filters, self-parking cars and ChatGPT. But what exactly is it, and how does a program manage to write like a person? Here’s the explanation, no formulas required.
Three layers that often get mixed up
Picture three circles, one inside the other:
| Layer | What it is | Example |
|---|---|---|
| Artificial intelligence | The whole field: any system that does tasks we associate with human intelligence, like reasoning, recognizing, translating or planning | A chess program |
| Machine learning | Instead of writing rules by hand, you give the system thousands of examples and it adjusts its numbers until it gets things right | Your email spam filter |
| Deep learning | Neural networks with many layers that learn complex patterns in text, images or sound | A chatbot like ChatGPT, Gemini or Claude |
All deep learning is machine learning, and all machine learning is AI. But not the other way around.
How a chatbot writes: word by word
An assistant like ChatGPT runs on a language model. Its core idea is surprisingly simple: guess which piece of text comes next, over and over. Here’s how it handles your message:
- Split the text. Your sentence is broken into pieces called tokens. Sometimes a token is a whole word, sometimes part of one, like “intelli” and “gence”.
- Give it meaning. Each token becomes a list of hundreds of numbers. Words with similar meanings end up with similar numbers.
- Connect the context. The attention mechanism decides which words matter for understanding each other word. In “the cat that chased the dog was tired”, it helps work out that the cat is the one who’s tired.
- Pick the next one. Every possible token gets a probability. One is chosen, added to the text, and the process starts again.
That’s why answers appear on screen bit by bit: the model is writing them token by token. Try it yourself with the simulator on our home page.
So does AI “think”? Today’s models perform very useful reasoning, but their foundation is still predicting text from learned patterns. That’s why they’re sometimes wrong with total confidence. We explain it in why AI makes things up.
How it learns: training a model
Before it can chat, the model is trained on enormous amounts of text. During training it tries to guess the next word millions of times, and every time it’s wrong it slightly adjusts its parameters: the billions of numbers that make up the model. In the end, those numbers capture language patterns, facts about the world and ways of reasoning.
Then usually comes fine-tuning with example conversations and feedback from people, so it answers in a helpful and safe way.
Seventy years in seven moments
| Year | What happened |
|---|---|
| 1950 | Alan Turing asks whether a machine can think and proposes an “imitation game” to test it |
| 1956 | The term “artificial intelligence” is coined at the Dartmouth conference in the US |
| 1997 | IBM’s Deep Blue beats world chess champion Garry Kasparov |
| 2012 | AlexNet, a deep neural network, dominates image recognition and kicks off the deep learning era |
| 2016 | DeepMind’s AlphaGo beats Lee Sedol at Go, a game thought to be beyond machines |
| 2017 | The paper Attention Is All You Need introduces the transformer, the basis of today’s language models |
| 2022 | AI chatbots go mainstream and generative AI enters work and everyday life |
What it’s good at, and where it fails
Where it helps most:
- Drafting, summarizing and translating text.
- Writing and reviewing code.
- Explaining difficult topics and analyzing data.
- Creating and editing images, audio and video.
- Automating repetitive tasks.
What to watch out for:
- It can make up facts with total confidence.
- It can reflect biases from the text it learned from.
- Its knowledge has a cutoff date, unless it searches the web.
- Don’t share personal or confidential data unless you need to.
Next step
Now that you know how it works, learn to ask for things properly in how to write good prompts, and keep the AI glossary handy when you hit an unfamiliar term.
