论文库

原文摘要

Laws and regulations deeply shape physicians' workflows and impact how they incorporate AI-based decision support tools to care for patients. Despite calls to comply with legal standards, many AI-based decision support tools have not transitioned well to clinical settings. The medical community remains concerned about their legal liability, risks to patients, and limited clinical value of such tools. To better understand how to address these concerns in the design of AI tools for clinical decision-making, we conducted 19 interviews with physicians in the U.S. We used scenario-based design to investigate physicians' legal concerns in the context of an AI-based drug recommendation tool. Drawing on the ethics of care, we surface connections between physicians' care and the healthcare laws and regulations that deeply shape their information needs. We describe how physicians wanted to use the tool in individual and collective practices to care for themselves, their patients, their colleagues, work institution, and even the broader medical profession. We discuss how the design of AI-based decision support tools can present legal ambiguities that would prevent their caring actions and lead to non-use of the AI tool. Through our work, we contribute to responsible AI development in healthcare. Our analysis illuminates why and how certain kinds of information reduce legal ambiguity and support physicians' care when using AI-based decision support tools. We contribute an understanding of how to help physicians navigate legal liability and draw implications for grounding AI tools' designs in their care.