Page 43 - AI Standards for Global Impact: From Governance to Action

AI Standards for Global Impact: From Governance to Action

o There is a pressing need for new metrics and independent scientific assessments to
evaluate AI's environmental footprint.
o The industry faces significant challenges in addressing AI's environmental impact,
including the need for cost-effective energy-saving technologies, metrics to evaluate
environmental effects beyond energy consumption (e.g. water use and material AI Part 1: International
impact), and the integration of technical standards into practical applications.

b) Innovations for the energy efficiency of AI
o Scientists are exploring ways to reduce AI's environmental impact by improving
software efficiency, such as compressing large foundation models, repurposing them
for multiple tasks, and employing techniques like quantization, while also addressing
energy-intensive digital hardware through innovations like neuromorphic computing.
o To reduce AI's environmental impact, the industry can focus on three key areas:
adopting green energy supply, implementing energy-saving technologies and efficient
products during AI operations, and promoting circularity in energy and materials to
minimize emissions throughout the AI system lifecycle.

c) Areas where collaboration and standards are needed:
o A collaborative effort to rethink the integration of hardware and software is essential
for creating sustainable AI systems, with energy efficiency being prioritized as a core
value to drive environmentally friendly advancements in both scientific and industrial
AI development.
o A global, interdisciplinary effort is essential to maximize AI's environmental benefits,
requiring human-centered metrics, reliable methodologies, and risk assessment to
ensure effective and sustainable solutions.
o Accurate GHG emission management, supported by relevant standards and technology
solutions, is essential to identify and address critical areas for emission reduction,
ensuring a systematic approach to decreasing AI's environmental footprint. Existing
standards provide a foundation for sustainable AI systems, including green data centres,
requirements for telecom sites, monitoring network intensity, and circular economy
practices. However, as technologies advance, there is a need to update requirements
continuously and further improve the design and efficiency of ICT infrastructure.
d) The following reports on energy efficiency and AI were announced during the session:

i. Measuring what matters: How to assess AI’s environmental impact published by ITU
ii. Methodology to assess Net Zero progress in cities – published by United for Smart
Sustainable Cities (U4SSC) initiative
iii. Guidelines for cities to achieve carbon Net Zero through digital transformation –
published by U4SSC initiative
iv. Artificial Intelligence for Climate Action: Advancing Mitigation and Adaptation in
Developing Countries – published by United Nations Climate Change – Technology
Executive Committee (UNFCCC-TEC)
v. Smarter, Smaller, Stronger: Resource-Efficient Generative AI & the Future of Digital
Transformation – published by United Nations Educational, Scientific and Cultural
Organization (UNESCO)

31