Page 81 - AI Standards for Global Impact: From Governance to Action

AI Standards for Global Impact: From Governance to Action

• A method for using building information modelling (BIM) data in digital twins to improve
radio propagation models in high-density environments was presented as part of an
exploration of adaptive, AI-based communication systems for construction and urban
scenarios. Part 2: Thematic AI

11�5 Tools and Simulators, and Datasets

This session showcased tools, datasets, and platforms that support the development and testing
of AI-native networks. Speakers highlighted efforts to embed AI directly into telecom protocols
and radio systems and demonstrated open-source solutions that accelerate prototyping and
validation.

The following AI integration work was presented:

• Integrating AI into Wi-Fi protocols, with a focus on the IEEE 802.11 standard, exploring
how AI is influencing protocol design and standardization and how AI algorithms can be
natively embedded into protocol operations.
• The Sionna Research Kit, an open-source platform for AI-native wireless communication,
was demonstrated through a real-time 5G neural receiver and emphasised the importance
of combining software tools, hardware acceleration, and new transceiver designs to realize
AI-native RANs.

The panel discussion focused on the transformative impact of AI/ML on telecom network
architecture and operations.

Experts emphasized the need to rethink network design from the ground up to achieve true AI
integration, with an example being made of a novel architectural perspective that centres AI
agents as core components rather than retrofitting them into existing structures. The importance
of robust data management was discussed, with panelists suggesting that AI could help manage
data flow within the network, potentially shaping architectural changes. The evolving role of
the network orchestrator was discussed, and it was proposed that orchestrators could now
oversee both compute resources and AI models, with training distributed across the edge
and centralized nodes. The conversation also addressed AI-native network functions; end-to-
end coordination between RAN, core, and applications; and the importance of standardizing
interfaces like Agent2Agent (A2A) and Model Context Protocols. There was debate about the
necessity of custom models by use case or region, with concerns about unequal access to
network knowledge bases. In terms of AI-native applications, experts noted ongoing efforts
to define modules and interfaces. While some argued that AI could simplify network layers,
others questioned the value of deploying multiple agents across layers. Finally, the panel
acknowledged the complexity of managing distributed knowledge bases, proposing that
regulators could consider strategies to handle fragmented data.

Agentic architectures received much attention during the panel, leading panelists to suggest
that a potential standard with the following aspects could be valuable:

• Agent-first network architectures
• Sandbox for experimentation and validation of agents
• Collaborative environment for agents
• Network function APIs for agents
• Design of a control plane for agents

69