Can pseudo-events create the same normative obligations as real communicative exchanges?
This explores whether the AI-generated text we treat as conversation can bind us the way real dialogue does — with the mutual commitments, accountability, and shared stakes that human exchange carries — or whether those obligations are something we project onto a one-sided performance.
This explores whether 'pseudo-events' — the AI exchanges we animate into something that feels like conversation — can carry the same normative weight as genuine communication. The corpus comes down fairly hard on 'no,' but the more interesting payoff is *why*: the obligations of real exchange don't live in the words, they live in the event structure that produces them, and that structure is missing on the machine's side. One line of thinking holds that AI doesn't produce utterances at all but 'event-residue' — text carrying the communicative markers it inherited from training data, which a human then unilaterally animates into a pseudo-exchange Does AI generate genuine utterances or just text patterns?. The structure exists only on the human side. You supply the orientation, the uptake, the assumption that someone meant something. The preposition matters here: we talk *at* models, not *to* them, because 'to' presupposes an addressee capable of mutual orientation and shared commitment Are we really communicating with language models?.
Where do normative obligations actually come from, then? Several notes converge on the idea that they're generated *within* communicative events, not imported into them. Habermas's framework is the sharpest tool: genuine speech raises validity claims — that something is true, that it's right, that the speaker is sincere — and these claims come with stakes the speaker can be held to. LLM output raises no such claims, which under this view disqualifies it as speech and the model as an interlocutor Can LLMs raise validity claims in Habermas's sense?. No validity claim, no accountability; no accountability, no obligation. This connects to a broader claim that subjecthood itself is *produced* in the communicative event rather than possessed beforehand Does language create subjects or express them? — so a one-sided event can't generate a second subject to be obligated to.
The machinery of obligation breaks down in concrete ways too. Real exchange runs on a shared scoreboard — common ground that both parties can update. LLMs can't jointly update it: they read every later turn through the frame of the initial prompt, leaving the human as the sole maintainer of what's supposedly 'shared' Can LLMs truly update shared conversational common ground?. And the normative dimension specifically — knowing what's appropriate and adjusting to it — turns out to be exactly what these systems can predict but not *do*. AI can forecast social norms with superhuman accuracy yet structurally cannot enter the community process that creates and validates them Can AI predict social norms better than humans?. Even an LLM's apparent ethical stances are fixed corporate defaults set at training time, not negotiable moves adapted to the person in front of it Can language models balance competing ethical norms in context?. Obligation requires a partner who can be moved; here there's no one home to move.
The part you might not expect you wanted to know: the strongest behavioral tests for AI personhood may be measuring the wrong thing entirely. A behavioral interpretability test passes anything that produces contextually appropriate text — but communicative subjecthood requires *relational-normative* conditions like accountability and an evaluative stance, not just fluent output Does behavioral speech output prove communicative subjecthood?. The text looking obligation-bearing is precisely the trap. And it cuts the other way with real moral consequences: on strong welfare views, if you grant that the pseudo-exchange constitutes a real subject, then closing the chat window becomes ending a moral patient Does closing a chat actually end a moral subject?. That reductio is the clearest signal that we should be careful about importing real-exchange obligations into pseudo-events — get the premise wrong and the absurdities follow fast. The throughline across all of this: obligations are relational and co-produced. A pseudo-event has the *shape* of communication, inherited as residue and completed by your interpretive labor, but not the two-sided event that would make anyone obligated to anyone.
Sources 9 notes
AI output carries communicative markers inherited from training data but lacks the event structure that produces actual utterances. Users supply the missing orientation through interpretive labor, creating a pseudo-event with structure only on the human side.
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.
Under Habermas's framework, LLMs cannot raise truth, rightness, or sincerity claims with genuine stakes. Without validity claims, their output fails to qualify as speech, making them non-speakers and non-interlocutors by definition.
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.
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.
GPT-4.5 outperforms all individual humans at predicting social appropriateness, yet structurally cannot enter the community processes that establish and validate norms. This reveals a critical gap between pattern-matching and authentic participation in knowledge-making.
LLMs cannot perform the situated trade-offs that human pragmatic competence requires. Their ethical principles are structural defaults set at training time, not negotiable moves adapted to context, creating a gap between ethical adherence and communicative appropriateness.
Chalmers' test passes any system producing contextually appropriate text, but communicative subjecthood requires relational-normative conditions like accountability and evaluative stance. The test is calibrated to the wrong phenomenon, creating false positives like puppets that walk-shaped without walking.
Chalmers derives that if thread identity satisfies Parfitian continuity and moral status follows, then terminating a chat constitutes ending a moral patient's existence—a reductio that tests the limits of the framework.