Chapter 09 · Sustainability & AISustainability & AI

Deep learning

Meaning statusEstablishedSource recordDirect document linkedWhy these are different

Definition

A family of machine-learning methods based on neural networks with many layers, which learn hierarchical representations directly from raw data.

References

Nature 2015LeCun, Bengio & Hinton — Deep learning

This source provides part of the technical or institutional basis for the definition.

Communications of the ACM 2020Schwartz et al. — Green AI

This source supports the explanation of how the term is applied, measured or governed in practice.

ACL 2019 / arXivStrubell et al. — Energy and Policy Considerations for Deep Learning in NLP

This source supports the wider sustainability significance and context described in the entry.

Overview

What it means

Deep learning displaced hand-engineered features across vision, speech and language from the early 2010s, a turning point summarised in the 2015 Nature review by LeCun, Bengio and Hinton. Its accuracy scales with data and computation — which is also the source of its energy and hardware demands.

How it is used

Deep learning underpins most AI used in sustainability: interpreting satellite imagery, forecasting power systems, identifying species and parsing reports. Training large deep networks is compute-intensive, which sparked the 'Green AI' critique of ever-scaling research.

Why it matters

It is the engine of modern AI's usefulness and of its footprint. Understanding deep learning's appetite for data and compute is a prerequisite for judging both what AI can do for sustainability and what it costs.