Nurse scheduling is a high-stakes optimization problem: shifts must be filled while respecting working-time limits, qualifications, and other rules. When AI suggests a schedule, nurses and managers need to know not just the recommendation, but how confident the system is in it. A new preprint from arXiv reports on the co-design and evaluation of an explainable AI prototype aimed at exactly this challenge.

The researchers worked with domain experts to shape how the AI communicates uncertainty, rather than simply adding a generic confidence score. The prototype was then evaluated to see whether the chosen explanations helped users interpret AI suggestions appropriately. The work highlights that uncertainty communication is not a technical afterthought but a design problem that benefits from direct input from the people who will rely on the system.

Because this is a single source, there are no conflicting findings to compare. The paper's contribution is in its process: expert co-design paired with evaluation of an explainable AI interface for a real-world operational task. The authors argue that making uncertainty visible in this way is key to trustworthy AI-assisted scheduling.