EXplainable Neural-Symbolic Learning (X-NeSyL) methodology to fuse deep learning representations with expert knowledge graphs: the MonuMAI cultural heritage use case

Deep learning models achieve high accuracy but lack transparency: The paper demonstrates that while modern deep learning excels at detection and classification tasks, these "black-box" systems are difficult to debug, interpret, and explain to non-technical stakeholders—a critical limitation for responsible AI deployment.

Combining AI with human expertise improves trustworthiness: X-NeSyL fuses deep learning with expert knowledge graphs (symbolic AI), allowing the system to provide explanations that domain experts can validate, rather than relying solely on opaque algorithmic outputs.

A new metric measures alignment between machine and human explanations: The methodology introduces an explainability metric that quantifies how well the AI system's reasoning matches expert human explanations, providing a concrete way to assess whether an AI system is truly interpretable to its intended audience.

Symbolic knowledge prevents the performance-explainability trade-off: Rather than forcing a choice between accuracy and interpretability, the approach leverages both deep learning's generalization capabilities and symbolic AI's rule-based clarity, demonstrating that these strengths need not be mutually exclusive.

Real-world application in cultural heritage shows practical value: The MonuMAI use case demonstrates the methodology's applicability to concrete problems, suggesting that neural-symbolic approaches can work beyond theoretical frameworks in domains where expert knowledge and accurate detection both matter.