XAI-on-RAN: Explainable, AI-native, and GPU-Accelerated RAN Towards 6G

AI in critical infrastructure requires transparency: The paper argues that as AI becomes central to mission-critical systems like autonomous vehicles, remote healthcare, and industrial automation, the "opacity of AI decisions poses risks" that demand explainable AI (XAI) solutions rather than black-box models.

Explainability involves real trade-offs: The research identifies a mathematical framework showing that transparency, latency, and computational efficiency are competing priorities—meaning organizations cannot maximize all three simultaneously when deploying AI in 6G networks, requiring deliberate design choices.

Hybrid approaches can improve performance: The paper's proposed "xAI-Native" hybrid model "consistently surpasses conventional baseline models in performance," suggesting that combining explainability with AI optimization doesn't necessarily sacrifice accuracy or speed.

Non-public networks demand higher standards: The paper emphasizes that dedicated network slices for specific industries (like healthcare or industrial automation) have heightened reliability and safety requirements, making explainability particularly critical in these controlled, high-stakes environments rather than just public networks.

6G design must prioritize trustworthiness from the start: Rather than treating explainability as an afterthought, the paper advocates for "AI-native" architecture that builds transparency into next-generation networks from the ground up, aligned with 3GPP standards for responsible network design.