Constructing the Human-AI Boundary with Metaphors - A Demystification-Legitimation Paradox

Paper i proceeding, 2026

This paper examines a central challenge in human-centric AI (HCAI), which is, how practitioners in safety-critical organizations define and maintain the boundary between human expertise and AI capability during AI adoption, and what this means for meaningful human oversight. We conducted a qualitative cross-case study based on 28 semi-structured interviews in two Nordic transport organizations. Our analysis followed a two-stage approach combining inductive identification of metaphors with boundary work theory. We identify seven metaphor clusters shaped by a demystificationlegitimation paradox, where practitioners both frame AI as a simple tool to preserve accountability (e.g., wrench, moose gap) and as a driver of organizational change (e.g., cow paths, digital mirror/swamp, playground). Two metaphors (energy-saving animal, moving sideways) combine both functions. We introduce metaphorical boundary work to describe how figurative language constructs and manages the line between human and AI roles. These metaphors act as informal governance mechanisms that shape responsibility, professional roles, and AI legitimacy in practice. We also present the Metaphor Matrix as a reference tool for reflecting on such discourse in design and governance. The findings suggest that human-centeredness is not only a system feature, but an ongoing organizational process shaped through everyday language.