Explainable AI for autonomous networks

5G network complexity is increasing dramatically with new technologies, a multitude of new services, new operator roles, new radio technologies, and new customer categories. In 6G, we expect even more services with high requirements to fulfill, and more tunable network parameters. This calls for intelligent Radio Access Network (RAN) automation. By combining AI techniques with a flexible architecture we can reach a higher degree of autonomous operation within RAN. However, their inclusion adds to the complexity of training them, managing their interactions, and understanding their behavior. In this presentation, we analyze how explainable AI (XAI) can help us to cope with increasing complexity, we present some initial ideas on how XAI can assist in improving our understanding of an AI-driven radio resource optimization to improve energy efficiency according to the traffic demand variations. In this context, XAI techniques could not only provide explanations of the agent’s behavior based on the knowledge of how actions influence the environment, but also in identifying relevant features and managing the complexity of training AI models.