Establishing and implementing a responsible artificial intelligence framework: a 1-year review.

Abstract

This work highlights successes and challenges of implementing a novel responsible artificial intelligence (RAI) framework, emphasizing healthcare disciplines needed to operationalize it. UNC Health developed an RAI framework to assess artificial intelligence (AI) solutions, featuring a 21-question intake survey aligned with institutional goals to promote fairness, transparency, accountability, and trustworthiness, and evaluated by clinical, analytical, and operational experts. Twelve survey evaluations revealed low fairness scores and resulted in 83% conditional approvals. Learnings included the importance of representative training datasets, systematic evaluation of vendor-provided models, and robust post-implementation monitoring. Challenges included the infrequency of analyses stratified by demographics, limited vendor transparency, and reliance on volunteer engagement for survey evaluations. Our framework provides a roadmap to assess AI tools in healthcare but requires overcoming implementation barriers like resource constraints and vendor cooperation. Future iterations should consider tiered evaluations based on risk likelihood and member engagement for scalability.