Finnish initiative 6G Flagship has published a new white paper introducing an artificial intelligence framework that can benefit the next generation of mobile communications. The report also analyses the impact of AI language models (LLM) in 6G.
“By incorporating AI directly into the network, we anticipate marked improvements in areas such as radio and network optimisation, privacy and security […], and meeting stringent latency and other network-specific criteria,” the document reads.
“Such advancements are poised to be invaluable for domains like the Internet of Things (IoT), robotics, smart cities, and autonomous systems, to name a few.”
One of the main crucial steps for AI integration with 6G is establishing a new framework that the authors called “AI interconnect.”
It should facilitate tasks such as:
- Seamless AI orchestration within the 6G network.
- Dynamic interaction between AI models and network components.
- Integration with open APIs across the edge-cloud continuum.
- Optimized AI task execution, including selection, placement, and coordination.
The framework is based on the MAPE-K Model: monitor, analyse, plan, execute with knowledge. The goal is to enable self-adaptation and automation.
AI and LLM Use Cases
The paper also highlights use cases where LLM promises to play a great impact in 6G networks:
| Scenario | Description | LLM/GPT Support | Potential Impact |
| Near real-time Analytics | Immediate processing and analysis of data at its source | Rapid, real-time adaptation to incoming data streams for more accurate and immediate pattern recognition than traditional models | Enhanced, more accurate decision-making processes and a user experience tailored to real-time data trends |
| Interoperability | Connecting systems across formats, protocols and APls | Natural language capabilities to interpret and generate diverse API calls, facilitating better protocol intercommunication | Seamless, frictionless communication between diverse systems, reducing system integration challenges and improving efficiency |
| Distributed Al | Al operations spread across multiple nodes | Swift multi-node communication and distributed reasoning and intent, leveraging vast training to infer across varied data sources | Superior resource allocation due to dynamic node coordination, resulting in faster, more efficient Al operations |
| loT Device Management | Managing vast arrays of interconnected devices | Real-time data processing to handle device configurations and adapt to device behavior on-the-fly | Improved device performance and health, leading to prolonged device life and better user experience |
| Network Security | Protecting network from threats and anomalies | Advanced patter recognition from a vast range of data, making anomaly detection more accurate and early | Proactive threat neutralization, minimizing vulnerabilities and ensuring more robust network integrity |
| Content Caching | Storing content closer to the user to reduce latency | Predictive modeling based on vast internet content understanding, forecasting user content needs more accurately | Personalized user experience with instant content delivery, reducing wait times and enhancing user satisfaction |
| Dynamic Spectrum Allocation | Allocating bandwidth dynamically based on the need | Deep traffic pattern understanding, predicting bandwidth needs based on diverse internet use-cases | Optimized bandwidth distribution, minimizing wastage and ensuring high quality communication experiences |
| Augmented Reality (AR) | Enhancing AR experiences with real-time data processing | Rich content understanding allows for generating and adapting AR data in diverse real-world scenarios | More realistic and responsive AR environments, offering users immersive and dynamic experiences tailored to their context |
| Autonomous Vehicle Coordination | Coordinating self-driving vehicles in real-time | Advanced predictive capabilities based on understanding a wide range of driving scenarios and conditions | Safer, more efficient driving routes and strategies, potentially reducing accidents and improving traffic flow |
| Smart Grid Management | Managing electricity distribution in real-time | Advanced predictive capabilities based on understanding a wide range of driving scenarios and conditions | Enhanced grid reliability, fewer outages, and optimized energy distribution catered to real-time needs |
“In particular, the co-evolution of both LLMs and 6G appears fascinating,” the document reads. “Therefore, this co-evolution presents a unique situation, and there are challenges to be considered.”
According to the authors, these challenges include the need to make sure LLM/GPT capabilities keep pace with the evolving needs of 6G and clear guidelines to integrate different approaches to LLM and networks.
“In light of our exploration, we believe that the integration of GPTs and LLMs with 6G networks has immense promise, and a collective push towards establishing shared practices and frameworks is crucial,” the paper concludes.
