Page 80 - AI Standards for Global Impact: From Governance to Action

AI Standards for Global Impact: From Governance to Action

• Telenor’s deployment of AI-driven traffic forecasting was presented as an example of how
to reduce RAN energy consumption. The system powers down components during low-
demand periods, achieving a 4% energy saving and longer energy-saving windows. The
solution supports sustainability goals and is scalable to more network areas.

11�4 Outcomes

During this session, several announcements were made on the release of challenges, datasets,
and open-source tools to support a wide range of applications of AI in Networks.
• Large Wireless Models Challenge: The session started with a presentation of the Large
Wireless Models Challenge hosted by ITU. The challenge will provide participants with
large-scale unlabelled data to pre-train custom models that can extract universal features
and then be used in fine-tuning for different downstream tasks.
• OPEA challenge: Similarly, the OPEA challenge invites competitors to to design,
implement, and deliver a practical generative AI application using the OPEA platform.
The application should address a real enterprise use case by leveraging OPEA’s modular
architecture and evaluation methodology.
• Arabic Telecom LLM: The first-of-its-kind Arabic Telecom LLM was developed by Khalifa
University researchers in partnership with Du, Microsoft, and Nokia
• Echo: Echo is - an open-source development platform from Shanghai University, built on
the Venus RISC-V simulator, designed for communication-AI fusion. See theinitial release
of the open-source protocol stack and supporting infrastructure The current version
includes a simulator and compilation support for AI computing units and mobile network
protocol stacks, with foundational physical layer processing enabled via RISC-V-based
extensions.

This session also explored how AI transforms network architecture across all layers, from core
to edge. Talks covered standardized intelligence stacks, AI infrastructure design, and intent-
based automation. Presenters also highlighted the value of digital twins in cybersecurity and
radio optimization, showing how AI-native strategies can drive real-time adaptability, security,
and efficiency in 5G and 6G networks.

Key Takeaways:

• Standardized AI pipelines and governance frameworks are needed to build truly AI-
native 6G networks. Current gaps in model lifecycle management were highlighted and
a standards-driven roadmap for explainable, interoperable AI system was proposed.
• A vision for embedding AI directly into the design of future networks was shared and the
concept of "3C" networks (co-design, collaboration, and composability) was considered
as a foundation for building AI infrastructure that integrates physical and cyber systems.
• A multi-agent LLM framework for intent-driven network automation showed how
specialized agents, interacting with external tools, can improve explainability and
scalability in managing complex telecom tasks.
• The shift from cloud to edge AI for real-time network applications was explored with use
cases from European and UK projects, outlining opportunities for standardization in edge
computing and tiny ML for IoT networks.
• Distributed digital twins were proposed as a means to simulate cyberattacks on large-scale
networks, with a demo showing how these models can help train AI systems to detect and
respond to threats without risking live infrastructure.
• Combining sensing and AI across devices and network layers can enable advanced robotic
applications. Some early prototypes demonstrating new 6G-enabled services and business
models were presented and related challenges were explored.

68