Neural network means a computer program made of many small, simple calculating units joined in layers, which learns from examples by adjusting the strength of the connections between them. It is the structure inside most of today’s AI, including chatbots and image makers.
Imagine a huge mixing desk in a recording studio, with thousands of sliders. Sound goes in one side and music comes out the other. Training a neural network is like nudging every slider, over and over, until what comes out is right. The final slider positions are what the network has “learned”.
In a sentence: “The photo app uses a neural network that learned what cats look like from millions of labelled pictures.”
What the word means
The plain point: the name borrows from the brain, but it is maths running on a computer.
IBM, the computer company, defines it as layers of simple “neurons” that learn from data.1 IBM adds that the idea was inspired by the brain’s nerve cells.1
In a living brain, a neuron is a nerve cell. In AI, a “neuron” is just a small calculation.
A simple example: show a network a photo, pixel by pixel. The numbers pass through the layers, and at the end it says “cat” or “dog”. If it is wrong, its settings are nudged so it does a little better next time. After a great many nudges, it gets good at it.
How it works, in three parts
The plain point: information goes in, passes through middle layers, and an answer comes out.
| Part | What it does |
|---|---|
| Input layer | Job: takes in the starting information, such as the pixels of a photo. |
| Hidden layers | Job: the middle layers where the “neurons” do their calculations. |
| Output layer | Job: gives the final answer, such as “cat”. |
Google’s course for developers also calls the middle layers “hidden layers”, and their nodes “neurons”.2 Each neuron multiplies what comes in by a number called a weight, then adds it up.2 Training changes those weights bit by bit.1
The weights are the sliders. When people talk about a model’s “weights”, they mean these learned numbers, and they are the model itself. See parameters meaning.
Why it matters to you
The plain point: it explains both how AI learns and why it is hard to see inside.
- There is no list of facts inside. What it learned is spread across a huge number of weights, so it can be fluent and still wrong. See hallucination meaning.
- “Open weights” means those numbers are published, so anyone can download and run the model. See open weights meaning.
- Why it gave an answer is hard to read off. Researchers are working on it. Does AI understand anything? covers what they have found.
Often confused with
- Deep learning: training neural networks that have many layers. See deep learning meaning.
- A model: a trained neural network is one kind of AI model. See AI model meaning.
- A digital brain: the brain was the inspiration, not the blueprint. Whether AI “thinks” is a separate debate.
Where the word comes from
The name comes from the brain: IBM says “the inspiration for neural networks comes from the biological neurons in the human brain”. The idea dates to 1943, when “Warren McCulloch and Walter Pitts proposed the first mathematical model of a neuron”. In 1958 Frank Rosenblatt introduced the perceptron, which IBM calls “the historical ancestor of today’s networks”.3
Questions people ask
Is a neural network like a human brain?
Only loosely. The brain was the inspiration: IBM says “the inspiration for neural networks comes from the biological neurons in the human brain, which communicate through electrical signals”. In practice each “neuron” is a small sum, not a living cell.1
Neural network vs deep learning: what is the difference?
Deep learning is training neural networks that have many layers. IBM says “‘deep learning’ refers to the training of models with at least 4 layers”.4
What is a hidden layer?
Any layer between the input and the output. Google: “additional layers between the input layer and the output layer are called hidden layers, and the nodes in these layers are called neurons.”2
Learn more on this site
- Start here: AI explained for complete beginners.
- The bigger picture: machine learning meaning and artificial intelligence meaning.
- Chips that run it: what AI actually runs on.
- The words: the AI glossary.
- Related: Explainable AI meaning.
- Related: What Is an AI Winter? AI Winter Meaning in Plain English.
QUICK CHECK · 3 QUESTIONS
Pick an answer to see if you got it. Nothing is recorded.
1. What is a neural network?
- A living brain grown in a lab
- Many small calculating units in layers that learn by adjusting their connections
- The cables that connect computers to the internet
Show the answer
Answer: Many small calculating units in layers that learn by adjusting their connections. It is maths running on a computer, built of simple units joined in layers, and it learns from examples.
2. What are a model’s weights?
- The learned numbers inside the network, like slider positions
- How heavy the computer is
- The list of facts the model has stored
Show the answer
Answer: The learned numbers inside the network, like slider positions. Training nudges the weights bit by bit; they are what the network has learned, and they are the model itself.
3. Why can a neural network be fluent and still wrong?
- It reads from a single list of checked facts
- It only works with pictures
- What it learned is spread across a huge number of weights, with no list of facts inside
Show the answer
Answer: What it learned is spread across a huge number of weights, with no list of facts inside. There is no list of facts inside, so a fluent answer is not a guarantee of a correct one.
Related terms: Agent · Machine learning · Narrow AI (weak AI) · Natural language processing (NLP) · Prompt · Token
- 1IBM, What Is a Neural Network?, 6 Oct 2021, read at source 1 Oct 2026“A neural network is a machine learning model that stacks simple ‘neurons’ in layers and learns pattern-recognizing weights and biases from data to map inputs to outputs.” ↑a“the inspiration for neural networks comes from the biological neurons in the human brain, which communicate through electrical signals.” ↑b“By adjusting these values during training, the network gradually learns to make accurate predictions.” ↑c“the inspiration for neural networks comes from the biological neurons in the human brain, which communicate through electrical signals.” ↑d
- 2Google, Neural networks: Nodes and hidden layers, updated 3 Dec 2025, read at source 1 Oct 2026“additional layers between the input layer and the output layer are called hidden layers, and the nodes in these layers are called neurons.” ↑a“take the sum of the product of each of its inputs … and a unique weight parameter, plus the bias.” ↑b“additional layers between the input layer and the output layer are called hidden layers, and the nodes in these layers are called neurons.” ↑c
- 3IBM, What Is a Neural Network?, published 6 Oct 2021, read at source 12 Oct 2026“In 1943, Warren McCulloch and Walter Pitts proposed the first mathematical model of a neuron”; “Later, in 1958, Frank Rosenblatt introduced the perceptron, an algorithm designed to perform pattern recognition.” “The perceptron is the historical ancestor of today’s networks” ↑
- 4IBM, What Is Deep Learning?, 15 Sep 2025, read at source 1 Oct 2026 (as cited on deep learning meaning): “‘deep learning’ refers to the training of models with at least 4 layers (though modern neural network architectures are often much ‘deeper’ than that).”↑
The comparison on this page (a studio mixing desk) and the cat-or-dog example are this site’s own ways of explaining, not claims from these sources.
Free printable AI cheat sheets, and AI news every week.
Sixteen one-page AI guides and checklists in one free PDF, no sign-up. And if you want to keep up, the newsletter covers what changed in AI each week, in plain English, with the sources.
SUBSCRIBE FREE ↗ GET THE CHEAT SHEET PACK