Toward Patient-Centered AI Fact Labels: Leveraging Extrinsic Trust Cues

Abstract

AI technologies in healthcare hold great promise for addressing numerous challenges, but ensuring that patients understand, trust, and adopt these technologies remains a significant hurdle. While the HCI community has proposed AI documentation frameworks (e.g., model cards) to enhance understanding, patient perspectives in the healthcare AI documentation remain underexplored. To address this gap, we designed prototypes based on existing frameworks and gathered feedback from 18 participants to explore their perspectives on AI documentation in cardiology, a domain where high-stakes AI tools are increasingly used and understanding users' trust in AI is essential. Our findings revealed patient needs for more detailed information about healthcare AI technologies, the importance of extrinsic trust cues (e.g., regulatory status), and the integration of AI documentation into existing care processes. Based on these findings, we discuss two design implications: enhancing patient-centeredness in AI documentation and leveraging extrinsic trust cues to improve its design. This study contributes to the HCI community by amplifying the patient voice in designing AI documentation and offering actionable insights into leveraging extrinsic trust cues effectively.