Communicating Uncertainty in AI-Supported Nurse Scheduling: Expert Co-Design and Evaluation of an Explainable AI Prototype

Uncertainty communication in AI systems can backfire: The study found that when the AI system communicated uncertainty in its nurse scheduling forecasts, healthcare professionals frequently misinterpreted this information and became less confident in their decisions, rather than making more informed choices. This suggests that simply showing uncertainty isn't enough—it must be communicated in ways that users actually understand.

Explanations alone don't guarantee trust or adoption: While nurses generally viewed AI explanations positively, these explanations "did not significantly affect decision confidence or advice acceptance." This indicates that making AI systems more transparent through explanations is necessary but insufficient for effective human-AI collaboration in healthcare settings.

Healthcare professionals need human-centered AI design: The research used expert co-design with actual nurses throughout the development process, highlighting that responsible AI in healthcare requires involving end-users from the start. Generic AI solutions won't work—systems must be tailored to how healthcare professionals actually think and work.

AI decision support requires careful interface design: The study used multiple evaluation methods including eye tracking and thematic analysis to understand how users interact with AI systems, not just whether they use them. This detailed attention to user behavior is essential for identifying where AI communication fails in practice.

Responsible AI in healthcare demands iterative evaluation: Rather than deploying AI systems and hoping they work, this research demonstrates the importance of mixed-methods evaluation with real users to identify problems before implementation—critical for high-stakes domains like nurse scheduling that directly affect patient care.