Can LLMs raise validity claims in Habermas's sense?

Habermas argues that validity claims — truth, normative rightness, and truthfulness — are not optional features of utterances but constitutive of what speech is in the communicative-action register. To speak is to raise a claim that can be redeemed or challenged; to hear is to recognize the claim and orient toward it. The mutual orientation toward validity makes the exchange a communicative event rather than mere noise.

LLM output raises the surface form of validity claims without taking any stakes. The system has no stake in truth: it defends or abandons claims based on prompt pressure, not on the basis of what is true. It has no social standing from which rightness claims could be issued — it occupies no role in a normative community. It cannot be sincere because sincerity requires a first-person whose inner state could match or fail to match what is expressed, and the system has no inner state in the relevant sense.

If LLM output does not raise validity claims, then — in Habermas's precise vocabulary — it is not speech. If it is not speech, the system producing it is not a speaker. If the system is not a speaker, it is not an interlocutor. The chain is deductive given the premises: accept Habermas's account of what speech is, and LLMs are ruled out as interlocutors by definition, not by empirical shortfall. The point is not that LLMs fail to be good enough communicators. It is that the category does not apply.

Source: AI Generated Research/Chalmers Engagement/project-brief.md