This study investigates psychiatrists' perceptions of AI-based clinical support tools (CSTs) for treating major depressive disorder (MDD). Eighty-three psychiatrists reviewed clinical notes and two CSTs (a summary and a treatment recommendation) presented as either AI- or psychiatrist-generated, with varying accuracy. Results showed a preference for human-derived CSTs, particularly for summaries, regardless of accuracy. This preference was less pronounced for accuracy-related ratings. For treatment recommendations, the preference for human-derived information only appeared when the recommendations were correct. Clinical expertise and AI familiarity showed little impact on results. The findings suggest a bias against AI-derived information, highlighting the importance of considering user perceptions in AI integration into psychiatric care.

Marta M. Maslej, Stefan Kloiber, Marzyeh Ghassemi, Joanna Yu, Sean L. Hill