What it means

An artificial neural network is a computing structure loosely inspired by biological brains, though built out of straightforward linear algebra and calculus. Numbers enter an input layer, get multiplied by adjustable weights across hidden layers, pass through non-linear activation functions and produce a prediction at the end.

During training, the system compares its guess against the target answer, measures the error and nudges millions (or billions) of weights slightly to do better next time.

Why it matters if you're a student

Even if you never write matrix maths yourself, grasping what weights are strips the mysticism out of AI conversations:

→ a trained model is essentially a giant file of tuned numbers plus the code to run inputs through them → there is no conscious homunculus inside deciding to be helpful or dishonest → changing how a network behaves requires either changing its inputs or updating those weights

An everyday example

Think of marking coursework with a rubric where you start with random weightings for structure, sources, clarity and originality. After comparing your scores against an experienced external examiner across hundreds of essays, you keep adjusting each weight until your predicted grade matches theirs closely.

Common misconception

Calling these systems "neural" tricks people into thinking they learn the way human students do. A human child recognises a tractor after seeing two pictures and hearing what it does on a farm; a standard neural network usually needs thousands of varied examples and can still get tripped up by an unusual camera angle.

  • Deep learning is a form of machine learning using multi-layered neural networks to find complex patterns in raw text, audio, images and sensor inputs.

  • Machine learning is a branch of computing where algorithms learn patterns from historical data to make predictions or decisions without manual rules.

  • A transformer is a neural network architecture using attention mechanisms to weigh relationships across sequences, powering modern language AI models.

  • An open-weight model is an AI model whose trained numerical parameters are publicly downloadable, letting teams inspect, fine-tune and self-host it.

Where it matters

Explore how this concept shows up inside real graduate workflows and sector briefings:

Articles that mention it

Get the fortnightly briefing

theVANTAGEd examines AI across real industries with critical thinking, every two weeks. Written by students.