As awareness of ethical concerns with artificial intelligence (AI) grows, organizations that design these technologies have increasingly recognized the need for “responsible AI” (RAI). Yet despite growing recognition of the importance of RAI, its implementation remains uneven and contested, as many RAI principles clash with the dominant discourses of technical rationality which permeate the cultures of technology companies. This dissertation examines how these challenges influence the ways that AI technologists advocate for RAI in development work. Based on qualitative fieldwork and interviews with 28 AI technologists, I present findings in two empirical studies. In study 1, I examine how AI technologists communicate about ethical issues in their day-to-day work, and propose that technologists navigate a diffuse epistemic stigma associated with the perspectives that underlie RAI work. Rather than marks associated with physical, or social taints, I argue that this stigma has roots in conflicting epistemic perspectives and moralized work practices. Technologists deploy a range of communicative tactics to manage this stigma and accomplish RAI work. In study 2, I analyze how aspects of RAI work are rendered visible and invisible in organizational structures, and the consequences of this visibility for the well-being and careers of technologists who take on RAI work. I find that organizational structures obscure the nature and effort of much RAI work, and that the benefits of increasing recognition may be attenuated by increased efforts to articulate RAI work through rationalized regimes. Throughout, this project constructs an empirical window into the hidden work of advocating for and accomplishing RAI objectives within large technology companies; showing that what often seems to be technical is organizational.
