Conscience and its place in the governance of AI agents: the RDA-Microsoft Data Director

Abstract

This position paper, in the Science, Policy, and Ethics in Conversation (SPEC) series, builds on a set of comments contributed to the community review of the RDA-Microsoft Data Director Agentic AI Tool Blueprint and takes that Blueprint as a worked case. It argues that the specification provides fully for the researcher who assents and makes no provision for the researcher who refuses; that this asymmetry is a structural product of how agentic systems are specified, since assent can be written into a requirement while refusal, in its most consequential form, cannot; and that the capacity at stake is conscience, which a specification cannot contain but whose place it can preserve or foreclose. It follows the recurrence of this asymmetry across sprint-style standard-making, the disclosure and provenance of AI-assisted scholarship, research assessment, and AI regulation, and concludes that the governance task for the agentic era is to keep building and moving the places where conscience can be exercised and a person can refuse. Position paper in the Science, Policy, and Ethics in Conversation (SPEC) series.