Page 106 - AI Standards for Global Impact: From Governance to Action

AI Standards for Global Impact: From Governance to Action

Figure 48: UNESCO Report on Smarter, Smaller, Stranger: Resource Efficient
Generative AI and the future of digital transformation

14�3 AI-driven environmental sustainability across industries

The intersection of AI and environmental sustainability across diverse sectors was highlighted
through examples of practical implementations and theoretical frameworks aimed at aligning
AI development with climate and ecological goals.

Examples included:

• CodeCarbon is an open-source tool that tracks CO₂ emissions from code execution and
provides actionable insights for researchers, developers, and data scientists. AI’s footprint
extends beyond CO₂ emissions to water usage and hardware manufacturing. Scope 1–3
impacts include:

o Scope 1: direct emissions from cooling data centres
o Scope 2: emissions from electricity used
o Scope 3: chip production and infrastructure

Concrete mitigation techniques such as model pruning, quantization, distillation, and optimized
training locations, while encouraging users to "use the smallest model for your need" and extend
the lifespan of hardware, could help address these challenges.

• Discussions on the sustainability risks of generative AI noted that hardware growth cannot
match model scale, leading to the need to consider:

o Cost-effective, generalizable models
o Repurposing pretrained models for new use cases
o Open-source ecosystems to ensure transparency and democratized development
o Greater efficiency through edge computing and modular designs

94