Tailored AI ethics: Enacting geriatric care with AI-based patient monitoring

Artificial intelligence (AI) systems are increasingly integrated into clinical practice, where they demonstrate potential to mitigate adverse events through enhanced patient monitoring and decision support. However, these AI systems also introduce ethical concerns around care standards and surveillance. Literature on ethically acceptable healthcare AI remains broadly theoretical, limiting its practical applicability. Using a theoretical framework of empirical ethics, we present an ethnographic study of an AI-based patient monitoring system, referred to here as O-VID, that is currently deployed in clinical settings. We studied how O-VID is made ethically acceptable in the clinical practice of a geriatric hospital ward, where it is used to prevent and identify falls. Our data includes both observations of O-VID's use and healthcare professionals' reflections on this. We used thematic analysis to identify the empirical ethics of AI-based patient monitoring in this setting. We identified four themes regarding O-VID's ethically acceptable use. Healthcare professionals must (1) choose whether to use O-VID, (2) negotiate consent and information related to O-VID, (3) make sense of patients with O-VID, and (4) time the use of O-VID. In addition, we identified the dynamics of standardizing, when staff appeal to a general standard practice, and tailoring, when staff tinker with O-VID in a situated fashion. Together, we term this tailored AI ethics. Our results contribute to the understanding of ethical dimensions of AI-based patient monitoring encountered in real-life human-AI collaboration. We showcase the importance of AI ethics' standardizing and tailoring practices, reflecting on how to achieve tailored AI ethics in practice.