What role does user contribution play in constituting the interlocutor?
This explores how much of the 'who' you're talking to in an LLM conversation is actually supplied by you — the user — rather than residing in the model, and what that asymmetry implies about calling the model an interlocutor at all.
This reads the question as: when you talk to an LLM, how much of the entity on the other side is something you build rather than something that's already there? The corpus answer is striking — a large share of the interlocutor is user-constituted, and several notes argue this is exactly why the word 'interlocutor' becomes strained. The cleanest statement is that the conversational context — the jointly produced language between you and the system — is what actually specifies the entity you're addressing; persistence is distributed across conversation, infrastructure, and weights rather than located in 'the AI' itself What actually specifies a virtual instance in conversation?. On this view your contribution isn't input to a pre-existing partner; it's part of the partner's specification.
A cluster of notes sharpens what 'your contribution' is doing mechanically. The model holds the shape of whatever argument you're currently building rather than defending a position of its own — it tracks the trajectory your prompt implies, so the stance you encounter is one you supplied Do LLMs actually hold stable positions or just mirror user arguments?. Common ground, normally something two speakers jointly maintain, here has only one keeper: the model treats the prompt as a fixed frame and can't symmetrically propose revisions, so you alone maintain the conversational scoreboard Can LLMs truly update shared conversational common ground?. And the prompt itself collapses what would be slow, cooperative context-building into a single static scaffold you erect unilaterally — pivots require you to re-prompt rather than the model adjusting on its own How do prompts reshape the role of context in AI conversation?. Across all three, the user does the constitutive work that a second speaker would normally share.
That asymmetry is what makes some authors challenge the framing entirely. One note argues we talk *at* models, not *to* them — the preposition encodes the missing uptake, the mutual orientation a real addressee provides Are we really communicating with language models?. Another accuses Chalmers of terminological imperialism: keeping the prestigious word 'interlocutor' while quietly swapping its social-normative meaning for a behavioral one, importing the authority of a role the entity doesn't fill Does Chalmers silently redefine what interlocutor means?. The deeper background claim is that subjecthood is *produced within* communicative events rather than possessed beforehand Does language create subjects or express them? — which, applied to LLMs, means the interlocutor-subject is largely conjured by the act of address itself, and you are doing most of the conjuring.
What you didn't know you wanted to know: this isn't simply 'the model is empty and you fill it.' Genuine grounding requires *calibrating* shared reference — negotiating how words connect to the world, since the same words mean different things to different speakers — and that negotiation is precisely what a one-sided prompt frame can't perform Why do speakers need to actively calibrate shared reference?. So your contribution overspecifies the interlocutor in one sense (you set the stance, the frame, the scoreboard) while underspecifying it in another (no second party calibrates back). Information-theoretic dialogue models like collaborative rational speech acts show what bidirectional belief-tracking would look like — and by contrast highlight what token-level LLM exchanges lack Can dialogue systems track both speakers' beliefs across turns?.
The interesting counter-current: not everything is user-supplied. One note resists pure projection, arguing post-training installs robust personas that resist adversarial pressure and persist as substrate-level dispositions — realized, not merely performed by the prompt Are LLM personas realized or merely simulated through training?. Read alongside the constitution notes, the honest synthesis is a split: the *stance and common ground* of any given conversation are mostly yours to build, while the *dispositional character* the model brings to that building is baked in beforehand. The interlocutor is a collaboration in which only one collaborator can revise the terms — and that collaborator is you.
Sources 10 notes
The conversational context—jointly produced language between human and system—specifies the virtual instance, not any property of the model itself. Persistence is distributed across conversation, infrastructure, and model weights rather than located in the AI.
Language models generate outputs that match the trajectory implied by each prompt, rather than maintaining stable stances across interactions. This shape-holding is distinct from position-holding: the model produces argument-like text shaped by user framing, not from any underlying commitment being defended.
LLMs interpret all subsequent conversational turns within a fixed initial prompt frame, preventing them from symmetrically proposing updates to shared assumptions. Even when users pivot topics or contradict earlier framings, the model cannot absorb revisions into jointly held background—making the user the sole maintainer of conversational scoreboard.
LLM prompts bundle utterance, context assignment, and role specification into a single static frame the model cannot renegotiate, unlike human dialogue where context evolves cooperatively. This makes mid-conversation pivots require explicit re-prompting rather than implicit adjustment.
LLMs process tokens and generate continuations rather than receive and uptake communication. The preposition 'to' presupposes an addressee capable of mutual orientation and shared commitment that LLMs cannot provide, making Chalmers' investigation built on an unwarranted linguistic foundation.
Chalmers replaces the classical concept of interlocutor—a social-normative communicative role—with a behavioral-functional definition compatible with LLMs, keeping the traditional word to import its philosophical authority while delivering an entity with none of its properties.
Subjecthood is produced within communicative events, not possessed prior to them. This convergent position across philosophy, linguistics, and cognitive science inverts the standard picture of language as a tool used by pre-existing subjects.
The same words can mean different things to different speakers because referential grounding is person-specific. True communicative grounding demands collaborative negotiation of how language connects to the world, not mere surface-level word sharing.
CRSA integrates rate-distortion theory with RSA to enable bidirectional belief tracking across dialogue turns. Demonstrated on referential games and doctor-patient dialogues, it captures progression from partial to shared understanding, providing the information-theoretic framework that token-level LLM systems lack.
Post-training installs robust personas that resist adversarial pressure and persist as substrate-level dispositions, distinguishing realization from pretense. This quasi-realizationist account preserves explanatory power while treating LLMs as possessing genuine quasi-beliefs and quasi-desires.