A Taxonomy of Questions for Critical Reflection in Machine-Assisted Decision-Making

Overreliance on AI recommendations reduces critical thinking: The paper identifies that decision-makers risk depending too heavily on machine recommendations, which leads to lower cognitive engagement—a key concern for responsible AI deployment in high-stakes fields like healthcare.

Asking the right questions can combat AI overreliance: The researchers developed a taxonomy of questions based on the Socratic method to help decision-makers think more critically about AI recommendations. This structured approach to questioning is presented as a practical tool to increase deliberate decision-making rather than passive acceptance of AI outputs.

The taxonomy addresses a gap in responsible AI practice: While explainable AI research exists, there was no systematic framework for helping humans engage critically with machine recommendations. This taxonomy fills that gap by providing specific questions decision-makers should ask themselves when using AI tools.

Clinical decision-making was the initial focus: The paper concentrates on healthcare as a starting point, where the stakes of over-relying on AI are particularly high. Feedback from medical researchers confirmed the taxonomy's relevance for this domain, suggesting it could help maintain physician judgment alongside AI tools.

The work supports compliance with emerging AI regulations: The authors explicitly note their taxonomy aims to support "effective human oversight as required by the European AI Act," connecting responsible AI practices to regulatory requirements for human control in AI-assisted decision systems.