AI and networks

Our research focuses on four key strands:

• AI-native network intelligence - using data-driven models to make real-time decisions about how networks are configured, managed and optimised.
• Autonomous, self-managing infrastructures - designing systems that can monitor themselves, predict issues and reconfigure without human intervention.
• Networks for AI - engineering high-performance, programmable connectivity that can efficiently support distributed training, inference at the edge and large-scale AI workloads.
• Human-centred, trustworthy connectivity - ensuring AI-enhanced networks remain transparent, efficient and aligned with user and societal needs.

Building on these core themes, our projects explore how AI can be embedded throughout the entire communications stack:

Enabling experimentation

Our national experimentation platform, JOINER’s AI Acceleration Facility, provides the compute, tools and open interfaces needed to design, train and test advanced AI models directly within live network environments. It also serves as a testbed for networks for AI, enabling experimentation with new architectures and protocols to support AI workloads end-to-end, from cloud to edge, under realistic performance, latency and energy constraints.

The REASON Open Networks Project is developing the foundations of AI-driven open networks; from multi-access technology intelligent control to end-to-end AI orchestration. Here we focus on new methods for data-driven decision making, dynamic resource allocation and predictions, enabling networks that can anticipate demand, detect anomalies and respond in real time. Working with industry, REASON translates our core research into practical mechanisms that make open networks intelligent, resilient, efficient and flexible.

With Project ARANA, we push further into the realm of agentic AI systems – collaborative, autonomous agents that learn from one another to optimise performance across the network. These agents coordinate to reduce latency, improve energy efficiency and adapt to changing conditions across the entire network stack, moving us closer to truly self-organising infrastructure.

Across every layer of communications

Our AI-native approach also underpins collaborations with the HASC and TITAN Telecoms Hubs, where we are applying intelligent orchestration, digital twins and context-aware analytics. These capabilities support adaptive, trustworthy and human-centred connectivity solutions, allowing us to explore how AI can support future use cases while meeting real-world requirements around reliability and accountability.

Taken together, these initiatives turn AI from an add-on into a core design principle. At Smart Internet Lab, we are embedding intelligence not only at the network edge or in the core, but across every layer of communications – and designing the networks that future AI systems themselves will rely on.