layman/terms
AI words

Neural network

nyoor · ul net · wurk

In shortA kind of software made of many simple connected parts that learns patterns from examples, loosely inspired by the brain.

Almost every modern AI, from face unlock to ChatGPT, runs on a neural network. The idea is simpler than it sounds.

Picture a huge team passing notes along. Each person reads the notes from the people behind them, decides how much to trust each one and passes their own note forward. At the end, someone shouts the answer. When the answer is wrong, everyone slightly adjusts who they trust.

That’s roughly a neural network. It’s software made of many simple connected parts that, together, learn to spot patterns from examples.

How it learns

Repeat that millions of times and it gets very good at the task.

Why “neural”?

It was loosely inspired by brain cells (neurons) connecting to each other. But it’s a very rough likeness: it isn’t a digital brain.

What it isn’t

It isn’t understandable line by line. Its knowledge is spread across billions of tiny adjustments, which is why even experts can’t fully explain why it gives a particular answer.

Where the comparison breaks down

A team of people knows what they’re talking about. Each part of a neural network is just a number being multiplied and added. Only the whole thing together does anything useful.

How the words fit together

Machine learning is the approach. A neural network is the most common engine for it. LLMs are very large neural networks trained on text.

Good to know

“Deep learning” just means using neural networks with many layers. It's the same idea, made very large.

When you'll hear it

“It's a deep neural network.” A neural network with many layers, which can learn very complex patterns.

“The weights are adjusted during training.” The strength of each connection is tuned as it learns.

“It has billions of parameters.” It has billions of adjustable connections, which is roughly how big it is.

Say it in a meeting

“Can we explain how this AI reaches its decisions well enough for our customers or regulators?”