A Zoo of AI Transparency Indicators: What Do Users Want (and Need) in Hospitals?

Artificial intelligence (AI), whether generative, specialized, or agent-based, is increasingly integrated into the daily lives of many citizens. In healthcare, several applications are already in clinical use, influencing patient care. Yet, our understanding of these learning systems remains partial. This article shows results from a study which is part of the XpliCAI project (Geneva University Hospitals, HUG). This paper describes the second study which examines the perception of available indicators for transparency in AI gathered from different sources. Workshops were conducted at HUG, involving 25 participants from diverse stakeholder groups, including healthcare professionals, medical students, patients, and caregiver partners. Participants simultaneously classified 18 transparency indicators for AI in healthcare into three levels of importance, followed by discussions to explain their choices. Participants prioritized interactional and algorithmic transparency, emphasizing explainability, data monitoring, and the right to human review, while social transparency was perceived as less immediate but ethically important. Overall, results indicate a strong consensus around practical, human-centered forms of transparency, with greater divergence on systemic aspects such as environmental impact and governance.