What Is a Neural Network? A Plain-English Explanation for Non-Technical Executives

A neural network is a type of machine learning system loosely inspired by the structure of the human brain. The brain is made of billions of neurons connected by synapses. The strength of the connections between neurons changes as we learn: frequently-used connections strengthen, rarely-used ones weaken.

A neural network mimics this at a computational level. It consists of layers of artificial 'neurons' (mathematical functions) connected by parameters (numbers that represent connection strength). During training, these parameters are adjusted repeatedly until the network produces accurate outputs on the training data. The trained network can then be applied to new inputs it has not seen before.

Imagine a large organisation where decisions pass through many layers of management. An input (a customer order) is received at the bottom and passes through layer after layer of personnel, each of whom applies their own rules of thumb to assess it, until it reaches the top where a final decision (approve or reject) is made. Each person's judgment has been shaped by years of experience seeing similar cases.

A neural network is similar: inputs pass through layers of mathematical functions, each of which transforms the input in a way learned from training data, until a final output is produced. The 'experience' is not human memory but numerical parameters adjusted during training to produce correct outputs.

This analogy is imperfect but captures the key insight: the system learns from examples rather than being explicitly programmed with rules, and its decisions emerge from the combined effect of many simple transformations rather than from any single decision rule.

Understanding that AI systems are neural networks with learned parameters (rather than explicitly programmed rules) has two practical implications for business leaders.

First, neural networks are not transparent in the way that traditional software is. You cannot look inside a neural network and understand why it made a specific decision the way you can read a decision tree or a rules list. This is the 'black box' problem that has significant implications for explainability requirements in regulated contexts.

Second, neural networks learn from data. The quality of the data they are trained on determines the quality of what they learn. A neural network trained on biased data learns biased patterns. A neural network trained on incomplete data has gaps in what it knows. This is why data quality is so foundational to AI performance.