Questioning AI: Promoting Decision-Making Autonomy Through Reflection

Over-reliance on AI recommendations undermines human decision-making: The paper identifies a critical problem where people depend too heavily on machine recommendations (such as clinical decision support systems in healthcare), which weakens their own reasoning abilities—a concern explicitly addressed in the European AI Act's Article 14 on human oversight.

Asking questions can restore human autonomy in AI-assisted decisions: The proposed solution involves having AI systems generate questions about decisions rather than just providing recommendations, which encourages people to think critically and maintain awareness of alternatives beyond what the automated system suggests.

Current AI systems risk reinforcing the status quo: The paper highlights that machine recommendations tend to repeat existing patterns and possibilities, potentially limiting human consideration of novel or different options that fall outside the system's training data.

Balancing reliance requires interdisciplinary approaches: Addressing over-reliance on AI demands insights from multiple fields—machine learning, user experience design, psychology, and philosophy—rather than technical solutions alone, suggesting that responsible AI requires collaboration across expertise areas.

Human oversight is foundational to trustworthy AI: The paper aligns with the EU's identification of human agency and oversight as the first requirement for trustworthy AI systems, positioning meaningful human control as essential rather than optional in AI-supported decision-making.