Explainability and AI Confidence in Clinical Decision Support Systems: Effects on Trust, Diagnostic Performance, and Cognitive Load in Breast Cancer Care

High AI confidence scores can backfire: The study found that when AI systems displayed high confidence in their recommendations, clinicians trusted them more but paradoxically made worse diagnostic decisions due to overreliance on the system, reducing diagnostic accuracy in breast cancer detection.

Low confidence scores promote caution at a cost: While low confidence scores prevented overreliance and encouraged clinicians to think more critically, they also decreased trust in the system and caused clinicians to spend significantly more time on diagnoses, potentially creating workflow inefficiencies.

Explainability features can increase cognitive burden: Some of the transparency features designed to help clinicians understand AI recommendations actually increased stress levels and cognitive load rather than reducing it, suggesting that more explanation isn't always better.

One-size-fits-all AI design won't work: The research demonstrated that age, gender, and professional role significantly shaped how different clinicians perceived and interacted with the AI system, indicating that effective clinical AI tools need to account for diverse user needs and backgrounds.

Balancing transparency with usability is critical: The study emphasizes that successful AI in healthcare requires careful design that doesn't just maximize explainability, but thoughtfully balances transparency, ease of use, and cognitive demands to build appropriate trust without encouraging dangerous over-reliance.