Read the argument critically

The assigned article asks readers to consider generative AI’s environmental cost without treating AI as the entire data-centre problem. It also links responsible AI to human participation: communities that are historically underrepresented need agency in shaping how AI systems represent them.

This is a perspective article, not a lifecycle assessment. Treat its numerical comparisons and examples as prompts for further verification—not settled project evidence.

Ethical lenses for a social-good project

• Need: does AI solve a real user problem, or is it novelty?
• Proportionality: is the benefit worth energy, compute, cost and operational complexity?
• Data: whose data is used, with what consent, privacy and retention?
• Representation: which groups are missing, stereotyped or disproportionately harmed?
• Human agency: can users understand, contest, correct or decline the AI output?
• Reliability: what errors, hallucinations and unsafe outputs are foreseeable?
• Accountability: who owns monitoring, incident response and remediation?
• Equity: who gains the benefit and who carries environmental or social cost?

Environmental decision record

For any proposed GenAI feature, document:

1. user outcome and non-AI alternative;
2. expected request volume and model size/service;
3. avoidable calls, caching/batching and output limits;
4. infrastructure/provider evidence available;
5. data-centre geography or energy information if material;
6. quality and harm-monitoring plan;
7. decision: adopt, narrow, prototype, defer or reject.

Practical reduction strategies

• Use the smallest system that meets the need.
• Avoid repeated generation when a deterministic or cached result works.
• Shorten inputs/outputs and remove redundant calls.
• Prototype with representative users before scaling usage.
• Track usefulness and failure, not only usage volume.
• Make environmental uncertainty explicit instead of claiming false precision.

Source

• Generative AI, Ethics & Its Impact on the Environment