Human-Friendly Explanations Checklist for Reinforcement Learning: XRL H-F-E Checklist
Journal
Artificial Intelligence – COMIA 2025 : 17th Mexican Congress, Mexico City, Mexico, May 12–16, 2025, Proceedings, Part II
ISSN
1865-0929
1865-0937
Publisher
Springer Nature Switzerland
Date Issued
2025
Author(s)
Contreras Olivas, Daniel Adrian
Type
text::book::book part
Abstract
Explainable Reinforcement Learning (XRL) has emerged as a critical subfield at the intersection of reinforcement learning (RL) and Explainable Artificial Intelligence (XAI), aiming to render the decision-making processes of learning agents interpretable, transparent, and accessible to human users. This paper introduces a comprehensive evaluation framework, the XRL H-F-E Metrics, to assess the human-friendliness of explanations generated by XRL systems. Drawing from interdisciplinary literature in computer science, cognitive psychology, philosophy of science, and human-computer interaction, the framework is structured across four dimensions: foundational principles (e.g., correctness, robustness, bias mitigation), cognitively aligned explanation types (e.g., “why”, “why not”, counterfactuals), characteristics of “good” explanations (e.g., contrastiveness, selectivity, causality), and human-friendly presentation attributes (e.g., comprehensibility, interactivity, personalization). This checklist provides both a theoretical model and a practical tool for fostering transparency and trust in RL applications, while also identifying key directions for future research, including quantitative metrics, adaptive explanations, and emotionally responsive interfaces. ©The authors ©Springer.
License
Acceso Restringido
How to cite
Contreras Olivas, D.A., Martinez-Villaseñor, L. (2025). Human-Friendly Explanations Checklist for Reinforcement Learning: XRL H-F-E Checklist. In: Martínez-Villaseñor, L., Martínez-Seis, B., Pichardo, O. (eds) Artificial Intelligence – COMIA 2025. COMIA 2025. Communications in Computer and Information Science, vol 2553. Springer, Cham. https://doi.org/10.1007/978-3-031-97910-1_21