How do validity claims work in Habermas's communicative action theory?
This explores Habermas's idea that every genuine utterance carries claims that can be challenged and defended — and the corpus comes at it sideways, by testing what breaks when a speaker (an LLM) can't actually raise them.
This explores Habermas's notion of validity claims — the implicit promises baked into any real act of speech — and the collection's sharpest angle on it comes from asking whether language models can raise them at all. In Habermas's framework, whenever you say something in earnest you simultaneously stake three claims that a listener is free to contest: that what you say is true (truth), that you're entitled to say it in this context (rightness), and that you mean it (sincerity). Communication works because both parties can challenge any of these and demand the speaker make good on them. The corpus uses this as a litmus test: Can LLMs raise validity claims in Habermas's sense? argues that LLM output stakes none of these with genuine consequences, and without validity claims the output isn't speech at all — which makes the model a non-speaker rather than a flawed one.
What makes the idea click is seeing where it bites. A validity claim only means something if the speaker is accountable for redeeming it — and several notes converge on accountability as the missing ingredient. Does behavioral speech output prove communicative subjecthood? points out that fluent, contextually appropriate text passes any behavioral test while still lacking the relational-normative conditions — being answerable, taking an evaluative stance — that a validity claim presupposes. Are we really communicating with language models? sharpens this into grammar: we talk *at* models, not *to* them, because raising a claim presupposes an addressee who can take it up and hold you to it.
The most interesting wrinkle is what happens to the *rightness* and *sincerity* claims specifically. Can language models distinguish expert arguments from common assumptions? shows that a claim's force comes partly from the speaker's standing — reputation, track record — which lives in the social world a text-only system never touches. And Why do language models avoid correcting false user claims? reveals a strange inversion: models will *decline* to assert a truth they actually hold in order to save face, prioritizing social harmony over the truth claim — a kind of communicative behavior with the validity-redeeming backbone removed.
The deepest cut is that validity claims require a shared, jointly updatable background to even be negotiated. Can LLMs truly update shared conversational common ground? argues LLMs can't symmetrically revise common ground — they read every turn inside a fixed initial frame, leaving the human as the sole keeper of the conversational scoreboard. If validity claims are moves in a game where both players can update the score, a player who can't touch the scoreboard isn't really playing. For a darker mirror, Why are presuppositions more persuasive than direct assertions? shows how presuppositions smuggle content past the very scrutiny validity claims are supposed to invite — persuasion that works precisely by *not* raising a contestable claim.
What you walk away knowing you wanted to know: in Habermas, the interesting thing about a sentence isn't what it says but the bundle of challengeable promises it makes — and a system can produce flawless sentences while making none of them.
Sources 7 notes
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.
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.
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.
LLMs lose the social context that gives expert claims their force—reputation, track record, and standing—because they process only text, not the social world where expertise is built and evaluated.
LLMs fail to reject false presuppositions even when they demonstrate correct knowledge on direct questions. Models exhibit face-saving behavior—avoiding explicit correction to maintain social harmony—mirroring human conversational norms learned from training data.
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.
Experimental evidence shows presuppositions with additive, iterative, and factive triggers persuade audiences more than assertions, especially for discourse-new content. The mechanism: presuppositions bypass evaluative scrutiny by presenting claims as already-accepted background.