What makes communication relational in ways belief is not?
This explores why you can attribute a belief to a single mind in isolation, but communication only exists in the space between two oriented parties — and what that asymmetry means for whether LLMs can be said to communicate at all.
This explores why belief and communication behave so differently when you try to strip them down to a solo, mechanical core. The sharpest version of the distinction in the corpus is the 'quasi-prefix' test Why does the quasi-prefix fail for communication?: you can describe a belief functionally — a 'quasi-belief' is still recognizably belief-shaped, a state inside one head that can be specified without reference to anyone else. Try the same move on communication and it doesn't weaken, it vanishes. Remove the other person's uptake and you're not left with 'quasi-communication'; you're left with text generation that a human has to interpret unilaterally. Belief survives isolation. Communication is constituted by the relationship, so isolation destroys it.
Why the difference? Belief is a property a system can have on its own; communication is something two parties *do* to and for each other. The corpus frames it as social action between persons — it carries speaker responsibility and requires mutual uptake, doing actual work inside a relationship rather than just moving information from one buffer to another Does AI really communicate or just distribute information?. This is why the preposition matters: we talk *at* language models, not *to* them, because 'to' presupposes an addressee capable of mutual orientation and shared commitment Are we really communicating with language models?. You can have a private belief; you cannot have a private communication.
That asymmetry is exactly why LLMs are easier to credit with beliefs than with genuine communication. A modest, deflationary case can be made for ascribing metaphysically undemanding states like beliefs and desires to LLMs — the same graded courtesy we extend to non-human animals — without claiming consciousness Can we defend modest mental attributions to large language models?. Beliefs are cheap to attribute because they're functional and individual. Communicative subjecthood is not, and this is where behavioral tests mislead us: a system can produce contextually perfect speech while lacking the relational-normative conditions — accountability, an evaluative stance, answerability — that real communication demands Does behavioral speech output prove communicative subjecthood?. The test detects speech-shaped output, not communication, the way a marionette is walk-shaped without walking.
The deeper, stranger payoff is that communication doesn't just *require* a subject on each end — it may *produce* them. Several strands converge on the view that subjecthood is a role generated within communicative events, not a possession you bring to them; language is the event through which a speaker becomes a subject Does language create subjects or express them?. Belief presupposes a holder; communication can be where holders come into being. And once you see communication as relational rather than as rational information exchange, the idealized cooperative picture cracks too — real communication runs on ethos, credibility, and influence, with affect and standing as constitutive features rather than noise Does rational cooperation actually describe how AI communication works?.
The thing you may not have known you wanted to know: this is why two LLM agents can 'collaborate' beautifully by passing latent states directly between each other Can agents share thoughts without converting them to text? and still not be communicating in the human sense. Lossless transfer of internal representations is the *purest* form of information distribution — and precisely because it removes the gap that interpretation, uptake, and mutual orientation live in, it removes the relationship. Belief can be copied. Communication has to be risked.
Sources 8 notes
Unlike belief, which can be characterized functionally as quasi-belief, communication is constitutively relational. Removing the intersubjective element doesn't weaken communication but eliminates it entirely, leaving only text generation—which humans must interpret unilaterally.
Communication is a relational act between persons that does work in a relationship; AI generates content without this relational structure, speaker responsibility, or mutual uptake. The conversational interface obscures this structural difference.
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.
Both robustness and etiological deflationist arguments beg the question against inflationism. A graded approach ascribing metaphysically undemanding states like beliefs and desires—while withholding consciousness claims—mirrors how we treat non-human animals.
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.
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.
Gricean cooperative pragmatics presume rational interlocutors coordinating shared understanding. But real communication runs on ethos, pathos, and strategic influence. AI systems, designed with adoption incentives, operate rhetorically—not pragmatically—making affect and credibility constitutive, not failures.
LatentMAS enables agents to share internal representations directly via KV caches, reaching 14.6% accuracy gains and 70.8-83.7% token reduction with no additional training. Hidden embeddings preserve reasoning fidelity that text-based systems cannot.