Persons Predict, Predictors Don't Person: A Response to "A Pragmatic View of AI Personhood"

Abstract

Leibo et al. (2025) propose treating AI personhood as a flexible bundle of obligations to be pragmatically assigned without inquiry into whether AI systems possess the capacities that make personhood meaningful. This paper argues that their pragmatist framework, while correctly identifying genuine governance challenges, commits a category error by conflating prediction with reasoning. Personhood presupposes capacities that prediction machines categorically lack: the ability to grasp norms, recognize violations, and be answerable for conduct. Assigning "responsibility" to systems incapable of responsibility does not close accountability gaps but obscures where responsibility actually lies. The paper concludes that pragmatism without ontology produces governance frameworks that pattern-match to accountability without instantiating it.