Page 102 - AI Standards for Global Impact: From Governance to Action
P. 102
AI Standards for Global Impact: From Governance to Action
The ongoing Swiss-based lifecycle assessments of Mistral AI models, evaluating emissions
from both training and inference over a two-year period, will also help inform standardized
environmental impact assessments across countries. Recommendation ITU-T L.1450 provides
"Methodologies for the assessment of the environmental impact of the information and
communication technology sector" and work is underway on a new ITU standard to assess the
environmental impact of artificial intelligence systems.
France's national strategy for sustainable AI, focusing on multistakeholder collaboration among
government, academia and industry, is developing actionable initiatives with international
partners, such as the recently launched Coalition for Sustainable AI.
Figure 44: France strategy on AI and environment
HSBC focused on challenges in current AI footprint assessments, noting issues such as:
• The lack of standardized tools and lifecycle boundaries, which hinders comparability
• Underreporting of Scope 3 emissions (e.g., hardware, e-waste)
• Omission of water usage and cooling overheads
• Over-reliance on estimates rather than observed telemetry data
The ITU report on Measuring What Matters: How to Assess AI's Environmental Impact was also
launched at the event. This report offers a comprehensive overview of current approaches to
evaluating the environmental impacts of AI systems.
90
