What is a neural network?
A neural network is a computer system loosely modeled on the brain: layers of simple math units called neurons, joined by adjustable numbers called weights. It learns by processing many examples and nudging those weights to shrink its errors. Modern AI, including ChatGPT, runs on neural networks holding billions of weights.
Why — the first-principles explanation
At its core a neural network is just a big pile of multiplication and addition arranged in layers. Each artificial neuron takes several numbers in, multiplies each by a weight, adds them up, and passes the result through a simple on/off-style function. The output of one layer becomes the input to the next. Stack enough layers and the network can represent extremely complicated relationships, like turning the pixels of a photo into the label "cat."
The magic is not the structure, it is the learning. A fresh network starts with random weights and gives useless answers. You show it an example with a known correct answer, measure how wrong it was, and use calculus (an algorithm called backpropagation) to compute which direction to nudge every weight so the error gets a little smaller. Repeat this millions of times over huge datasets and the weights slowly settle into values that produce good answers. Nobody hand-programs the rules; the rules emerge from the data.
A network with many hidden layers is called a deep neural network, which is where the term "deep learning" comes from. The same basic idea powers image recognition, speech-to-text, and the large language models behind chatbots, which are neural networks with hundreds of billions of weights trained to predict the next word in a sentence.
An example that makes it click
Imagine teaching a kid to sort fruit by feel, blindfolded. At first they guess randomly. Each time they guess "apple" and you say "no, orange," they slightly adjust the little rules in their head: bumpy skin leans orange, smooth-and-hard leans apple. After a few hundred fruits, those tiny adjustments add up to a reliable sorter, even though nobody wrote down a rulebook.
A neural network learns the same way, except the "little rules" are numbers (weights) and the adjustments are automatic. With enough practice examples, the numbers settle into a setting that gets the answer right.
Key facts
- A neural network has three kinds of layers: an input layer, one or more hidden layers, and an output layer.
- Connections between neurons carry adjustable numbers called weights; learning means tuning these weights.
- The tuning algorithm is called backpropagation, first popularized for neural nets in 1986.
- A network with many hidden layers is a deep neural network, the basis of 'deep learning.'
- Large language models like the ones behind ChatGPT are neural networks with hundreds of billions of weights (parameters).
▶ The 60-second explainer (script)
What is a neural network? It's a computer system loosely modeled on the brain. Picture layers of simple math units called neurons. Each one takes numbers in, multiplies them by adjustable values called weights, adds them up, and passes the result to the next layer. On its own that's just arithmetic. The clever part is how it learns. You start with random weights, show the network an example with a known answer, measure how wrong it was, and automatically nudge every weight to make the error smaller. Do that millions of times over huge datasets and the weights settle into values that give good answers, without anyone writing the rules by hand. Stack many layers and you get deep learning. Stack enough and train on text, and you get the large language models behind chatbots like ChatGPT, which hold hundreds of billions of these weights. So a neural network is really just math plus practice.
What authoritative sources say
People also ask
Is a neural network the same as the brain?
No. It borrows the loose idea of connected neurons, but it is math running on computer chips and is far simpler and less flexible than a real brain.
What is a 'weight' in a neural network?
A weight is an adjustable number on each connection that controls how much one neuron influences the next. Training tunes these weights to reduce errors.
What's the difference between a neural network and deep learning?
Deep learning just means a neural network with many hidden layers. All deep learning uses neural networks, but a shallow network isn't 'deep.'
Do I need one to use AI?
No. Neural networks run behind the scenes. You can use tools like ChatGPT without ever touching the underlying network.