Interpretable Uncertainty Quantification in AI for HEP