Bridging the Trust Gap in AI-Driven Optical Networks with Structured Explainability

Paper in proceeding, 2026

AI/ML models can automate optical-network decisions, yet operators distrust their opaque outputs. Existing explainability methods help but remain hard to interpret and not directly actionable. We propose the Capture–Characterize–Communicate (3C) framework, which formalizes explainability as an end-to-end pipeline, i.e., from capturing model behavior,  through local explanations, to human-readable decision guidance. The framework is demonstrated on two optical-network problems: explainable RL for RMSA and LLM-augmented explainability for QoT estimation, where it produces auditable, operator-facing explanations.