Agentic AI and the Attribution of Extended Cognitive Abilities

Abstract

This paper brings the panel’s concern with relational agency into the epistemic domain. Inquiry, understood as the goal-directed activity of settling questions, is never simply an individual cognitive achievement: it depends on others who gather, filter, frame, and sometimes answer on our behalf. Drawing on joint work with Johan Largo on LLMs, the paper argues that many technologically mediated inquiries are best understood through the category of the epistemic proxy: a distinct agent or system that performs epistemic labour for a principal, under the principal’s aims and oversight, without thereby becoming part of the principal’s cognition or acquiring the standing of an epistemic authority. This framework clarifies why LLMs are neither mere tools nor authorities, but it also illuminates ordinary interpersonal reliance. To rely on another as a proxy is to expose oneself to their framing, omissions, sycophancy, and failures of challenge while remaining responsible for what one comes to believe. The paper therefore treats proxy relations as a site of relational vulnerability: SDTs reshape agency not only by adding artificial intermediaries, but by reorganising the human chains of reliance through which inquiry, trust, and responsibility are distributed.

Part of panel Disruptive Technology and Relational Agency: Therapy, Intimacy, Memory, Trust, and Individualization - part 1