Explainable & Trustworthy AI
XAI taxonomies, counterfactual explanations, and the path from AI principles to responsible systems and regulation.
How can people understand, trust and safely govern machine learning models? This research line builds the conceptual foundations of Explainable Artificial Intelligence (XAI) and Responsible AI, and develops new methods for explaining models in ways that are actionable and reliable.
My work on explainability started from a practical question in one of my application domains: what lies beneath the black-box models used for road traffic forecasting (Barredo Arrieta et al., 2019). It grew into the 2020 taxonomy of XAI concepts, methods and challenges toward responsible AI (Arrieta et al., 2020), one of the most cited papers in the field and the core of Alejandro Barredo Arrieta’s doctoral thesis. Since then the line has moved in three directions:
- from explainability to the broader notion of trustworthy AI, connecting principles, ethics, technical requirements, auditing and regulation such as the EU AI Act;
- new explanation methods (counterfactual, neuro-symbolic and perturbation-based explanations for object detectors and 3D data) and ways to measure the quality of explanations;
- explainability as a requirement in high-stakes domains such as medicine, industry, transportation and federated learning.
Much of this work is done with Andreas Holzinger (BOKU University), Francisco Herrera and Natalia Díaz-Rodríguez (University of Granada) and Luca Longo (Technological University Dublin).
Main contributions
- Foundations and reviews: the XAI taxonomy toward responsible AI (Arrieta et al., 2020), a 2023 review of what is known and what remains to be done (Ali et al., 2023), the community manifesto of open challenges for XAI 2.0 (Information Fusion Best Paper Award 2025) (Longo et al., 2024), a survey of XAI for time series (Rojat et al., 2021) and research frontiers in machine learning and knowledge extraction (Holzinger et al., 2025).
- From principles to responsible systems and regulation: a framework that connects AI principles, ethics and the seven requirements for trustworthy AI with auditable responsible-AI systems and regulation (Díaz-Rodríguez et al., 2023), an analysis of the nature, requirements and regulation of trustworthy AI (Herrera et al., 2025), the right to be forgotten in AI (Lobo et al., 2023), and a design framework for operationalizing trustworthy AI in healthcare (Moreno-Sánchez et al., 2026).
- Counterfactual explanations that are plausible, minimal and adversarially robust, produced with generative models (Del Ser et al., 2024) or from data morphology (Pascual-Triana et al., 2025), including counterfactuals for 3D point clouds and semantic segmentation, some of them with LLMs in the loop (Holzinger et al., 2025; Rozajac et al., 2025; Kočić et al., 2025).
- Explaining perception models: black-box and segment-based explainers for object detectors in industrial settings (Andrés et al., 2024; Andres & Del Ser, 2025) and the neural-symbolic Greybox XAI framework for image classification (Bennetot et al., 2022).
- Evaluating explanations and explaining temporal models: a quantitative measure of the strength of post-hoc explanations (Bello et al., 2025), post-hoc explainability of deep echo state networks and reservoir computing models (Barredo Arrieta et al., 2022; Armentia et al., 2022), and the use of explanations to detect out-of-distribution samples (Martinez et al., 2022).
- Explainability in applications: information fusion for robust and explainable medical AI (Holzinger et al., 2022), explainable COVID-19 lesion delineation, Alzheimer’s progression and angiographic reconstruction (Li et al., 2023; Rahim et al., 2023; Wu et al., 2022), explainable defences against Byzantine attacks in federated learning (Rodríguez-Barroso et al., 2024) and explainable QoE estimation (Bilbao et al., 2025).
Publications in this area (28)
All my publications related to this research line, in reverse chronological order, including those mentioned above.
2026
2025
2024
- Information FusionExplainable Artificial Intelligence (XAI) 2.0: A Manifesto of Open Challenges and Interdisciplinary Research DirectionsInformation Fusion, 2024