Skip to main content

Advocating for sustainable ML/AI

Instruction and application
Complete

Technical fixes are not enough if culture and incentives ignore sustainability. Advocacy means making efficiency, fairness and governance visible in roadmaps, budgets and reviews.

Driving organisational change​

Anchor sustainability next to accuracy, latency and revenue. Executives sponsor targets; engineers propose architectures; compliance maps controls; CSR links narratives for customers and investors.

Practical moves by role​

  • Engineers: publish energy and latency baselines per model tier; propose distillation or batching for hot paths; instrument cost dashboards.
  • Leaders: fund renewable-aligned regions, cap “open-ended” GPU sandboxes and reward teams that retire zombie experiments.
  • Governance: add sustainability checkpoints to model risk reviews alongside fairness and privacy.

Training and education​

Run internal sessions on green software, cloud carbon tools and responsible data practices. Pair hands-on labs (measuring a training job) with policy context (what regulators expect).

Staying current​

Follow updates on the EU AI Act, ISO AI work, Green Software Foundation, Partnership on AI and Responsible AI Institute.

Key point​

Advocacy succeeds when sustainability has owners, metrics and budget—not only slide decks.

Action item: Reflection​

Reflection
Which stakeholder groups must be involved for a sustainability programme to stick in your organisation?
Your reflection here...
What is one policy or process change you could propose within 30 days?
Your reflection here...
How will you measure whether advocacy translated into behaviour change?
Your reflection here...

Quick check​

Question 1 of 1
Which approach best sustains green AI practices beyond a one-off hackathon?
  • A. A single email announcement
  • B. Integrating sustainability metrics into model review gates and team KPIs
  • C. Banning all deep learning
  • D. Outsourcing responsibility to a vendor without internal ownership
Correct Answer: B

Feedback: Processes and incentives embed practice; events alone rarely change defaults.