In a recently published position paper on ArXiv, researchers advocate for the development of Evaluative AI (EAI). This approach shifts away from systems that produce a single recommendation, focusing instead on supporting human decision-making by presenting competing hypotheses alongside evidence for and against each one.
Computational Argumentation as a Foundation
The paper identifies computational argumentation as a particularly suitable paradigm to provide a formal, computable foundation for EAI. This framework ensures that AI systems are not only explainable but also contestable, allowing human users to engage critically with the machine's reasoning.
The proposal establishes a long-term research agenda aimed at distributed and human-centered EAI systems. By moving beyond "black-box" outputs, the goal is to create a transparent environment where AI serves as a tool for dialectical reasoning rather than an automated oracle.
