The Making of Ethical AI: Developing Machine Learning Solutions for Healthcare

Abstract

Artificial intelligence (AI) is often framed as a powerful solution to society’s most pressing challenges. In critical domains such as healthcare, AI promises to enhance diagnostic accuracy, optimize clinical workflows, and expand access to care. Yet high-profile failures and persistent risks raise the question of how AI practitioners maintain momentum for uncertain and unproven technologies as solutions to healthcare problems. Drawing on a comparative ethnography of three research institutions, participant observation at other workshops, conferences, and professional events, interviews with clinical and technical experts, and analysis of media, policy, and scientific texts, I demonstrate how practitioners position AI as an ethical and legitimate healthcare response despite persistent limitations and uncertain impacts. I argue that AI endures not because it is seen as the most effective healthcare intervention or because it has consistently produced clinical gains, but because practitioners moralize their work as serving a common good. I theorize moralization processes as a mechanism through which actors sustain momentum for technoscientific projects by presenting them as ethically grounded. Moralization refers to the social practices and institutional processes that align technological projects with ideals such as equity, care, and wellbeing. Across settings, practitioners manage uncertainty and doubts about technical efficacy by building relationships, adapting institutional narratives, and appealing to moral commitments. In this way, moralization stabilizes healthcare AI by generating consensus and shared purpose in a field marked by uncertainty and rapid change. At the same time, moralization can obscure technological limitations, diffuse accountability, and divert attention from alternative solutions. By situating healthcare AI within broader transformations in the cultures and organization of technoscience, the dissertation advances a new model of technoscientific authority grounded not in detachment or claims of neutrality, but in active alignment with other institutions, actors, and values. More broadly, it explains how moral claims, professional authority, and organizational dynamics sustain commitment to ethically fraught and uncertain technological solutions to collective problems.