Skip to main content

Understanding sustainability in the context of ML/AI

Instruction and application
Complete

For years, many teams optimised accuracy first and treated energy and ethics as secondary. That mindset no longer scales: regulators, investors and users expect AI that isefficient, fair and explainable.

What is sustainability in ML/AI?​

Sustainability means developing and operating AI to minimise harm (environmental and social) while maximising durable value. That includes energy-aware training and inference, responsible data practices, hardware lifecycle thinking and governance that prevents discriminatory outcomes.

How sustainability shows up in your role​

  • ML engineers / data scientists: model efficiency, experiment discipline, choice of hardware and region, documentation of data cuts.
  • Product and strategy: align roadmaps with ESG commitments; justify compute spend with business and climate risk.
  • Compliance / governance: map AI systems to privacy minimisation, auditability and emerging AI rules (for example EU AI Act risk tiers).

Environmental impact (high level)​

Training and serving large models increases electricity demand for GPUs/TPUs, cooling and networking. Major operators publish emissions trends and clean-energy procurement strategies—see for example reporting on data centre electricity growth.

Social and ethical dimensions​

Sustainability is not only CO₂. Bias, transparency and data governance determine whether AI reinforces inequality. Examples in public discourse include scrutiny of automated benefits risk scoring and historical issues with large-scale biometric collection without consent.

ESG framing​

Environmental: lower-carbon compute, efficient architectures, renewable-aligned scheduling. Social: fairness, accessibility, stakeholder impact. Governance: policies, audits, ownership and documentation so systems can be reviewed.

Case study: optimising data centre cooling​

Google partnered with DeepMind to model data centre plant behaviour and recommend control actions. Google reported a large reduction in cooling energy—see DeepMind’s write-up.

Action item: Case study quiz​

Question 1 of 2
In the data centre cooling example, what did the ML system primarily optimise?
  • A. Marketing spend on AI products
  • B. Plant controls such as fans and chillers to reduce cooling energy while meeting safety constraints
  • C. Employee headcount in facilities teams
  • D. Network bandwidth between regions
Correct Answer: B

Feedback: The case is about operational efficiency of cooling, not unrelated business metrics.

Which pillar of ESG most directly covers “fairness and non-discrimination in model outcomes”?
  • A. Environmental
  • B. Financial leverage
  • C. Social
  • D. Liquidity
Correct Answer: C

Feedback: Equity, labour and user impacts typically sit under the Social dimension alongside governance processes.