Original Human Agency: Why Human-in-the-Loop Is Not Enough

Keeping a human in the loop is commonly treated as sufficient protection against the loss of human control. This paper argues that final approval can remain with a person even when AI has already framed the problem, formulated the purpose, selected the criteria, and supplied the reasons for action. I introduce Original Human Agency as the capacity to remain the responsible author of one’s purposes, judgements, actions, and self-formation. On this basis, I distinguish derivative agency, in which human judgement becomes progressively dependent while formal authority remains intact, from augmented agency, in which AI expands human understanding and action without replacing their direction. Drawing on philosophical accounts of agency and self-formation together with emerging evidence from AI-assisted knowledge work, I argue that systems should be evaluated not only by output quality or human oversight, but by their effects on purpose formation, reason articulation, contestability, and the continued development of human judgement.