Generative Explainability for Next-Generation Networks: LLM-Augmented XAI with Mutual Feature Interactions

LLMs can make AI decisions more understandable to non-experts: The paper demonstrates that using a large language model to translate technical AI explanations into plain language improves how useful and comprehensive those explanations are—increasing usefulness by 12.2% compared to standard technical outputs—making it easier for network operators without AI expertise to understand why systems make specific decisions.

Current explanation methods fall short for real-world operators: Existing explainability techniques like SHAP produce outputs that are too technical for non-specialists to act on, creating a trust barrier even when AI systems are working correctly. This framework addresses that gap by focusing on what operators actually need to know.

Considering how features interact matters: The framework goes beyond simply ranking which factors influence a decision; it also examines how different factors work together. This more nuanced approach to explanation appears to provide operators with more complete and actionable insights into AI system behavior.

Human evaluation confirmed the approach works: The researchers tested their framework with actual specialists and achieved 97.5% correctness while maintaining high agreement among evaluators, suggesting the natural language explanations accurately represent what the AI system is doing—an important validation for real-world deployment.

This approach could improve operator trust in AI-driven networks: By making AI decision-making transparent and understandable to non-technical staff, this framework addresses a key barrier to adopting AI in critical infrastructure like telecommunications networks.