Toward Native Explainable and Robust AI in 6G Networks: Current State, Challenges and Road Ahead

Abstract

6G networks are expected to face the daunting task of providing support to a set of extremely diverse services, each more demanding than those of previous generation networks (e.g., holographic communications, unmanned mobility, etc.), while at the same time integrating non-terrestrial networks, incorporating new technologies, and supporting joint communication and sensing. The resulting network architecture, component interactions, and system dynamics are unprecedentedly complex, making human-only operation impossible, and thus calling for AI-based automation and configuration support. For this to happen, AI solutions need to be robust and interpretable, i.e., network engineers should trust the way AI operates and understand the logic behind its decisions. In this paper, we revise the current state of tools and methods that can make AI robust and explainable, shed light on challenges and open problems, and indicate potential future research directions.