🧠 What is a neural network?
A pattern-spotting machine built from simple parts, inspired (loosely) by your brain.
3 min read
A neural network is a machine that spots patterns. It's made of tiny units called 'neurons' arranged in layers, connected by adjustable weights. The name is borrowed from the brain, but the resemblance is loose — think of it more like a network of dials that get tuned until the machine gets the answer right.
Here's the core idea. You feed data in one end — a photo, a sentence, a number. It flows through the layers, each layer transforming it a little, and a result comes out the other end: 'this is a cat', 'this is spam', 'the next word is mat'.
At first the weights are random, so the machine is wrong. But you show it the right answer, and a rule called 'backpropagation' nudges every weight slightly to make the machine *less* wrong next time. Do this millions of times and the machine gets very good — not by being told the rules, but by learning them from examples.
That's the whole magic and the whole mystery: nobody wrote the rules for 'what makes a cat a cat'. The network discovered them by adjusting millions of dials. We can see that it works, but we can't easily explain *how* it decided — which is why AI is powerful and also hard to fully trust.
Watch a neural net learn (live)
Blue and orange dots are two categories. The background is the network's live "opinion" of which is which — watch it sharpen from a blur into clean regions as it learns. No rules were given; it worked the pattern out itself.
Train a tiny neural network live: watch it learn to separate blue dots from orange dots, one mistake at a time.
🕒 Try it in 60 seconds
- 1Watch the demo on this page train — notice it's terrible at first, then improves.
- 2Ask an AI chat: 'What's the difference between a neural network and normal programming?'
- 3Then ask: 'Why is it hard to explain exactly how a neural network decides?'
Want the full picture?
Get the free 5-day "AI in your job" email course — jargon-free, five minutes a day.