rTAIM \;(Rebuilding Trust in AI Medicine) \;Monthly Seminars
\n\;
\nSeminar #26
\nDialogical Models and Counterfactual Questions for Epistemic Robustness in XAI
\nRocí\;o Martí\;n Istilart (Universidad Nacional del Sur\, Argentina)
\nWe are happy to announce the forthcoming 26th rTAIM \;Online Seminar\, \;with the participation of Rocí\;o Martí\;n Istilart on the 27th March 2026\, \;17h00-18h00 Lisbon Time Zone\, via Microsoft teams.
\nhttps://teams.live.com/meet/9378537563070?p=6c27AWdBD7eGSN48wB
\nID Teams: \; \;9378537563070 | Password: \;uT7jS3]
\n# Seminar 26: Explainable Artificial Intelligence (XAI) faces the challenge of providing explanations that are not only technically accurate but also epistemically robust and meaningful for users. This paper proposes a normative-epistemological framework for the formalization of counterfactual explanations within a dialogical setting. Building on the Explanation&ndash\;Question&ndash\;Response (EQR) protocol and Walton&rsquo\;s dialog theory\, the work introduces a structured set of critical counterfactual questions designed to test whether automated inferences remain stable across alternative contexts or depend on spurious correlations. The proposal integrates counterfactual reasoning into human&ndash\;machine dialogues as new locutions\, enabling explanations to be dynamically scrutinized in a way analogous to human argumentative practices. This approach aims to strengthen the causal\, contextual\, and selective dimensions of XAI explanations while improving their transferability and user alignment. An illustrative example in the domain of automated criminal decision-making shows how critical counterfactual questioning can help uncover hidden biases and enhance transparency. Although the framework is currently conceptual\, it outlines a promising path toward improving the epistemic quality and defensibility of automated decisions. The paper concludes that embedding counterfactual explanations within dialogical protocols such as EQR can contribute to more trustworthy\, context-sensitive\, and critically testable XAI systems\, while also identifying key technical and empirical challenges for future research.
\nShort bio: \;Rocí\;o Martí\;n Istilart is a PhD student in Philosophy at Universidad Nacional del Sur (Argentina) and a research fellow of CONICET (since 2026). Her work focuses on the epistemology of artificial intelligence\, particularly on bias detection\, explainable AI (XAI)\, and the philosophical analysis of automated decision-making. She holds a teaching degree in Philosophy from Universidad Nacional del Sur.
\nrTAIM \;Seminars: https://ifilosofia.up.pt/activities/rtaim-seminars
\nhttps://trustaimedicine.weebly.com/rtaim-seminars.html
\nOrganisation:
Steven S. Gouveia (MLAG/IF)
Mind\, Language and Action Group (MLAG)
Instituto de Filosofia da Universidade do Porto &ndash\; UID/00502/2025
Fundaç\;ã\;o para a Ciê\;ncia e a Tecnologia (FCT)
____________________________________________
\nInstituto de Filosofia (UI&\;D 502)
Faculdade de Letras da Universidade do Porto
Via Panorâ\;mica s/n
4150-564 Porto
Tel. 22 607 71 80
E-mail: ifilosofia@letras.up.pt
http://ifilosofia.up.pt/