Page 79 - AI Standards for Global Impact: From Governance to Action

AI Standards for Global Impact: From Governance to Action

network compression was shared to demonstrate effects on reducing communication
overhead. The result was that transmission costs can drop below 1% with little accuracy
loss. This enables efficient model updates across distributed mobile networks.
• The use of LLMs in telecom was explained, showing that current models are not optimized
for telecom-specific tasks. Examples of custom benchmarks and datasets combining Part 2: Thematic AI
synthetic and real network data were shared. These tools help evaluate and adapt LLMs
for real-world telecom use cases.
• IBM’s approach to making LLMs more suitable for enterprise use was shared. The use of
“generative computing,” which replaces prompts with structured programming to improve
control, was explained. This helps reduce hallucinations and boosts performance, even in
smaller models.
• The role of edge devices and real-time intelligence in enabling autonomous operations
and how disaggregated and cloud-native access networks create new opportunities for
AI were highlighted, focusing on Open RAN as a foundation for modular AI pipelines.
• An example of LLM-powered agents managing radio frequency (RF) and IoT systems
was provided. These agents can perform tasks like beamforming and scheduling by
reasoning over spatio-temporal data. Agentic AI can thus improve decision-making and
system adaptability. In IoT systems, LLMs enhance RF sensing by incorporating natural
language processing, enabling sensors to interpret and generate human language for
smarter device communication. By integrating the natural language modality, LLMs
facilitate sophisticated multimodal data analysis, combining RF data with textual and audio
inputs for comprehensive insights. This capability improves anomaly detection, contextual
understanding, and decision-making, helping IoT systems become more intelligent and
adaptive.

11�3 Standards and open source

This session focused on the role of standards, datasets, and open-source tools in building
practical, sustainable, and scalable AI-driven networks. Speakers highlighted the value of open-
source platforms in accelerating innovation, the critical need for trusted data in AI validation,
and how AI can directly improve network efficiency and sustainability. Together, the talks
demonstrated how collaboration and transparency are key to shaping AI-native communication
systems.

Some of the main issues highlighted during the session were:

• How open-source LLMs and AI agents are driving innovation in networking, exploring
use cases such as automated troubleshooting and intelligent network management and
collaboration through open-source tools.
• The need for high-quality, reliable datasets to validate AI/ML in 6G systems and
challenges in collecting multimodal and frequency-diverse data highlight the importance
of measurement campaigns to support trustworthy AI models in complex network
environments.
• An open-source RISC-V-based library that integrates AI computing with wireless baseband
processing was presented to show how it allows for software-defined protocol stacks
and decoupling of hardware and software. This enables scalable and sustainable mobile
network upgrades without overhauling infrastructure. RISC-V is an open standard
instruction set architecture based on reduced instruction set computer (RISC) principles.
• The Sionna Research Kit, an open-source platform for prototyping AI-native RAN systems,
was showcased using a real-time neural receiver compliant with 5G. The implementation
addressed challenges in latency, hardware acceleration, and developing new signal
processing algorithms for future networks.

67