Page 107 - AI Standards for Global Impact: From Governance to Action

AI Standards for Global Impact: From Governance to Action

• ZTE emphasized that global collaboration and human-centred principles are necessary to
balance innovation with energy constraints. ZTE’s vision of a three-tier strategy for green
AI includes:

o Efficient infrastructure: Enhancing computing and cooling through distributed Part 2: Thematic AI
architecture and energy-aware networking, in particular when developing high-speed
networks.
o Intelligent empowerment: Optimizing AI through multimodal learning and model
compression techniques.
o Efficient implementation: Deploying AI across sectors such as smart grid, disaster
warning, green finance, and personalized energy tracking.
• PrevisIA uses AI to forecast deforestation in the Brazilian Amazon by analysing satellite
imagery, historical deforestation patterns, topography, water bodies, and socioeconomic
data, with a focus on detecting unofficial roads – a key predictor of forest loss. The
initiative integrates three pillars: AI-driven road detection via Sentinel-2 imagery,
predictive modelling with geospatial dashboards, and collaborative enforcement with
state prosecutors. Partnering with Public Prosecutor’s Offices in Pará, Amazonas, Acre,
and Amapá, PrevisIA achieved 73% forecast accuracy (2021–2024) within a 4km radius.
Its reports enable embargoes, fines, and legal action against illegal deforestation,
demonstrating AI’s potential for proactive environmental protection. The system integrates
remote sensing and geospatial AI to assess risk factors like proximity to roads, showing
that 95% of deforestation occurs within 5.5 km of roads. PrevisIA supports enforcement
actions (e.g. fines and licensing suspensions) and empowers authorities with real-time
alerts.
• UNFCCC launched the TEC’s technical paper on AI for Climate Action, which focuses on
applications in developing countries, including:
o Mitigation with energy optimization, emissions tracking, and renewable energy.
o Adaptation with early warning systems and localized services for smallholder farmers.
o Policy with practices to accelerate and transfer climate actions to developing countries.
o Governance with responsible AI frameworks to mitigate risks in fragile contexts.
Some of the key takeaways are highlighted below:

a) AI’s environmental impact is multidimensional, encompassing emissions, water, material
extraction, and infrastructure. Inference and deployment now dominate energy use, not
just model training.
b) Tools (including open-source tools) are being developed to offer practical pathways
for emissions estimation and reduction and can help align developers' daily work with
sustainability goals.
c) Hardware limits are a real constraint on AI scalability, calling for innovations for greater
model efficiency in the interest of sustainability.
d) Sectoral applications (e.g. deforestation prevention, smart energy, and public health)
showcase AI’s potential for impact — if implemented responsibly.
e) Shared global progress is essential to environmental sustainability, highlighting the
importance of capacity-building in developing countries.
f) Opportunities for ITU-T Study Group 5;

o Support the development of standardized metrics and tools (e.g. model-level energy
benchmarks, and lifecycle impact metrics).
o Collaborate on guidelines for responsible AI deployment in climate-sensitive and
resource-constrained environments.
o Facilitate cross-sector knowledge-sharing and open innovation platforms to promote
scalable, low-impact AI solutions.

95