Researchers propose computational argumentation as foundation for explainable, contestable Evaluative AI

A new position paper on arXiv proposes using computational argumentation as a formal foundation for Evaluative AI (EAI), a class of AI systems designed to support human decision-making by presenting multiple competing hypotheses alongside evidence for and against each. The authors argue that EAI systems should avoid producing a single recommendation, instead enabling users to weigh alternatives with full context.

The paper, titled 'Towards an Argumentative Foundation for Evaluative AI' and authored by Xiang Yin, Tim Miller, Nico Potyka, Antonio Rago, and Francesca Toni, frames computational argumentation as a mathematically grounded paradigm that can make EAI systems explainable and contestable. This approach emphasizes transparency and user engagement, allowing stakeholders to challenge or refine the reasoning presented by the system.

The authors position their work as setting a long-term research agenda toward distributed and human-centered EAI systems. They suggest that argumentation-based EAI could better align with human decision processes, which often involve weighing pros and cons rather than accepting a single output.

The paper was submitted to arXiv on April 25, 2026, and is categorized under Artificial Intelligence (cs.AI) and Multiagent Systems (cs.MA). While it is a position paper and not an empirical study, it contributes a conceptual framework intended to guide future research in explainable, interactive AI systems.