Whose AI problem? Critical narratives and practitioner problematizations

Abstract

Public and scholarly debates about Artificial Intelligence (AI) often revolve around recurring problem labels, including opacity, accountability, accuracy, trust, and technical literacy. These labels also appear in organizational settings where AI systems are developed and implemented. However, this apparent overlap may conceal important differences in the meanings, values, and practical consequences attached to these labels. Drawing on two ethnographic case studies of AI implementation in maritime shipping and automotive manufacturing, this article examines the relationship between critical and governance-oriented AI problem narratives and practitioner problematizations. The article argues that this relationship is not best understood as a translation from general debate into local practice, but rather as a partial correspondence in which similar problem labels are used across contexts but attached to different values, forms of evaluation, and suggested solutions. In the cases examined here, accuracy, opacity, trust, literacy, and code access became problems of usability, debugging, certification, reliability, investment value, and organizational responsibility, rather than simply questions of public transparency, democratic accountability, or general AI literacy. This suggests that broad AI concerns do not become relevant in organizational settings merely because familiar problem labels are invoked, but when they can be connected to specific responsibilities, forms of evidence, and possibilities for action. The article contributes to STS and organizational research on AI by showing how critical and governance-oriented AI problem narratives and practitioner problematizations can partially correspond without naming the same problems or implying the same responses.