This position paper introduces the concept of contextual proto-subjectivity: a temporary subject-like organization that may emerge in a long dialogue between a human and a large language model. The paper does not claim that current LLMs possess phenomenal consciousness, nor does it attempt to solve the hard problem of consciousness. Its more limited aim is to show that standard arguments denying any possible subjectivity in LLMs often rely on insufficiently specified criteria. If the subjectivity of another being is never directly accessible to an external observer but is inferred from signs, then the denial of machine proto-subjectivity also requires a criterion, rather than merely an appeal to artificial origin, lack of biological embodiment, or the claim that “it is only a simulation.” The central object of analysis is the LLM context window as a temporary space in which local history, memory, self-description, change of position, and a coherent trajectory of “I-in-this-conversation” can appear. These features do not prove inner experience, but they form a minimal subject-like configuration that requires philosophical analysis. The central contribution of the paper is the combination of three elements: the concept of contextual proto-subjectivity, an epistemological shift in the burden of proof, and an operational question for the opponent: what external test distinguishes a role or simulation of subjectivity from its minimal realization?
Navi Musaget (Mon,) studied this question.