Myths vs reality: why everything you know about AI in networks is wrong
The UK is on its way to becoming an AI superpower, particularly with the recent £31 billion agreement between the government and several US tech giants. It has built a…
The UK is on its way to becoming an AI superpower, particularly with the recent £31 billion agreement between the government and several US tech giants. It has built a…
The UK is on its way to becoming an AI superpower, particularly with the recent £31 billion agreement between the government and several US tech giants. It has built a solid foundation to ensure success by drafting strategies for compute, spectrum, Open RAN, and safety. However, there are still questions about how it might enhance connectivity.
Recent studies have shown that network failures and outages cost UK businesses millions every single year. In fact, without serious mobile network upgrades, the UK risks losing £230 billion in economic potential.
In this blog, we demystify the myths around AI as it currently exists in networks and advanced connectivity technologies and show the way forward.
Networks today are brittle, reactive, and painfully manual. Every new function requires humans to handcraft policies, debug conflicts, and coordinate apps across the infrastructure.
The University of Bristol’s Agentic AI changes the rules: networks write their own policies, resolve conflicts autonomously, and validate performance in cyberspace through digital twins before interacting with the real world. It will build networks that reason and govern themselves.
Network traffic is chaotic, spatially and temporally distributed, and constantly shifting.
Imperial College London has developed Spatial-Temporal Reinforcement Learning (STRL), which combines gated recurrent units and graph attention networks to capture temporal traffic patterns and spatial dependence across practical networks. This enhances the management and configuration of network resources and antennas without using channel status information.
For years, the telecoms industry has been stuck in an arms race of hardware: building bigger arrays, stacking more antennas, and throwing brute-force deep learning at the problem. But this leads to sky-high costs, energy waste, and inefficient training that ignores the actual physics of wireless signals.
By embedding domain-specific knowledge of wireless into AI, Queen Mary University of London has transformed antenna systems into intelligent, efficient learners. Using graph-based modelling for Continuous Aperture Array (CAPA) and Pinching Antenna System (PASS), our deep learning architectures leverage the natural structure of wireless channels, slashing training costs while boosting performance beyond conventional MIMO.
AI is also about erasing uncertainty inside the hospital walls. University of Cambridge researchers are demonstrating that with AI-powered optical wireless positioning, hospitals can track patients, inventory, staff, and life-saving equipment, with centimetre-level accuracy, even in places where GPS and WiFi fail.
No more delays in finding a patient when every second matters. No more guesswork in emergency care.
By combining AI models with new transceivers, this system delivers precise, real-time indoor localisation down to the centimetre, which is resilient, interference-resistant, sustainable, and aligned with the NHS’s Net Zero Healthcare mission.
Today’s satellites are dumb workhorses (essentially bent pipes): they waste spectrum, burn through power, and struggle to maintain performance in high-mobility scenarios because the waveforms they rely on were never designed for such dynamic conditions. They can’t sense and communicate at the same time, and their performance deteriorates in complex, fast-changing environments such as low Earth orbit — where rapid Doppler shifts, time-varying channels, and constant handovers between moving nodes expose the limits of current signal designs.
Neuromorphic AI and turbo-estimation, being developed by Loughborough University, can change the game. By mimicking the brain with spiking neural networks and utilising adaptive signal processing techniques, satellites can now think and react in real-time. This means there will be no more dropped links in motion and wasted energy or spectrum. Satellites will see, sense, and communicate simultaneously — powering secure defence, global connectivity, and UK leadership in space technology.
The future has to go beyond patching old systems with smarter tools. It will be built on networks that learn, sense, adapt, and govern themselves. This will be built on truly intelligent infrastructure: satellites that see while they connect, antennas that optimise themselves, resilient timing and positioning systems that eliminate blind spots, and AI tools that orchestrate entire networks without human micromanagement.
TITAN is part of the Federated Telecoms Hubs (FTH) alongside CHEDDAR, HASC, and JOINER. Together, we conduct world-leading R&D and create a new knowledge-sharing framework, roadmap technology, develop skills, and engage with standards while instilling a commercialisation-focused mindset across academia.
If you have any questions about the research discussed in this article, please contact titan-enquiries@eng.cam.ac.uk.
