From Model Cards to Subtle Care: Competing Logics of Ethics in Healthcare AI

Abstract

Ethical considerations are increasingly integrated into healthcare artificial intelligence (AI) systems, yet they are typically operationalized in a technical logic disconnected from concrete healthcare practices. This contrasts the emerging logic of situated empirical ethics enacted by practitioners during care, leaving correspondence between these two logics largely unexplored. We present STR.AI, an interactive prototype of an AI-based patient monitoring system. We designed STR.AI grounded in empirical research on the ethical considerations of AI-based patient monitoring, materialized within interactive, provocative model cards. Through STR.AI, we reflect on possible tensions between the logics of operationalization and empirical ethics, with ethics’ 1) temporality as predefined and context-dependent, 2) function as constraining and revealing choices, and 3) scope as standardized and situated. We contribute knowledge on tensions between formal operationalization and ethical realities in clinical practice. We discuss design considerations for bridging such gaps, highlighting mismatches in designers’ and practitioners’ engagements with healthcare AI ethics.