Will future LLMs possess agency or sentience needed for intentional communication?

Determine whether future large language models will ever possess a degree of agency and possibly sentience sufficient to support intentional, Gricean communication, thereby bearing the intrinsic features commonly associated with authorship and communicative intent.

Background

In discussing whether text produced with LLM assistance can count as authored, the paper distinguishes between human authorial intention and the mechanistic next-token prediction of current LLMs. Drawing on Grice’s theory of communication, the authors note that intentionality is central to authorship and that present LLMs appear to lack agency and sentience.

The authors explicitly state that it is unclear whether future LLMs will ever possess these capacities. Resolving this uncertainty would clarify whether LLM outputs could ever bear intrinsic authoriality on internalist grounds, rather than relying solely on a human’s endorsement or externalist criteria.