The path to ‘Mobile AI’ laid out in GSMA report

The global digital economy is increasingly driven by the convergence of advanced mobile communications and artificial intelligence, according to a paper published by the GSMA and GTI Telecom. The report states that as 5G scales globally, mobile networks will expand coverage and quality of service, while AI will move from cloud to on-device and the edge. Pervasive mobile connectivity enables widespread access to AI, while AI simultaneously reshapes network architecture.

The end-game is termed “Mobile AI”, which relies on a collaborative device–edge–network–cloud system that combines network reliability and low latency with AI algorithms capable of perception and decision-making. Structure is described as a “three-layer, four-dimension[al]”: vertically linking foundation, execution, and application layers, and horizontally integrating four domains: AI for Network, Network for AI, Mobile AI agents/terminals, and Mobile AI applications.

The report frames Mobile AI as requiring global cooperation and shared standards, and anticipates that as 5G-Advanced and 6G mature, mobile networks and AI will become a foundation for large-scale intelligent services.

The report’s central claim is that Mobile AI will be built around the interaction between devices, networks, edge computing, and cloud platforms, with mobile infrastructure carrying traffic and supporting AI workloads. Mobile traffic associated with AI services is expected to grow, and AI-related network traffic is forecast to increase at a CAGR of over 70% over the next decade. Around 2031, AI traffic may exceed traditional application traffic in global networks, the authors claim.

The growth in demand for edge AI processing will accelerate, since edge inference systems rely on device-to-network communication, thus placing new requirements for uplink capacity and close-to-zero latency.

The Mobile AI era

The report’s proposed architecture for Mobile AI is summarised as a “device–edge–network–cloud” system. Devices perform local sensing, edge infrastructure handles low latency computation, and clouds provide initial training and on-going reasoning. The telecoms network links these layers, managing traffic and service quality. The telecoms network itself may have to be at least partially-optimised by its own AI instances.

A large part of the report focuses on AI acting on the networks themselves. AI can assist network planning and operational optimisation, using real time data to adapt capacity and configuration. Similarly, in operations and maintenance, AI systems can identify anomalies, predict faults, and to some extent, coordinate responses. The report argues that these abilities will support the end-game of fully-autonomous network management.

Networks supporting AI applications

Intelligent devices and agents will generate new patterns and high quantities of traffic, and different service requirements. Some applications require low latency for control (robotics or remote operation), while others are based on the use of data-rich sources such as video and sensor streams. Networks will need more flexible service models than traditional best-effort, one-size-fits-all workloads.