Chapter 09 · Sustainability & AISustainability & AI
Neural network
Definition
A computational model composed of layers of simple interconnected units whose connection strengths are adjusted during training, loosely inspired by biological neurons.
References
This source provides part of the technical or institutional basis for the definition.
This source supports the explanation of how the term is applied, measured or governed in practice.
Overview
What it means
Networks learn by propagating errors backward through their layers to update weights — the backpropagation algorithm. Depth (many layers) enables the hierarchical feature learning reviewed in LeCun, Bengio and Hinton's 2015 Nature paper; scale is what links capability to energy use.
How it is used
Neural networks underlie essentially all modern AI applications in sustainability, from image interpretation to language analysis to control systems. Their behaviour is statistical and data-dependent, which shapes both their power and their failure modes.
Why it matters
As the substrate of contemporary AI, neural networks define what the technology can and cannot do: they interpolate brilliantly within their training distribution and extrapolate unreliably beyond it — a crucial caveat for environmental decisions under novel conditions.