Can you weaken communication without eliminating it entirely?
This explores whether communication is a thing you can dial down by degrees, or whether stripping out its core ingredient flips it off entirely — leaving something that looks like talk but isn't.
This explores whether communication is a thing you can dial down by degrees, or whether removing its essential ingredient doesn't weaken it but switches it off completely. The corpus has a surprisingly sharp answer: for some things, yes — and for communication specifically, no. The distinction turns on what kind of thing communication actually is. Belief can be approximated — you can describe a 'quasi-belief' that behaves functionally like belief without the full inner state. But communication is constitutively relational: it requires two parties mutually oriented toward each other. Take that mutual orientation away and you don't get weaker communication, you get text generation that a human has to interpret on their own Why does the quasi-prefix fail for communication?. It's a binary, not a dimmer switch — because communication is social action between people, not information moving from a source to a destination Does AI really communicate or just distribute information?.
But here's the more interesting wrinkle the corpus surfaces: even if communication can't be *partially* present, the supporting machinery around it absolutely can be degraded by degrees — and that's exactly what current training does. Preference optimization (RLHF) measurably erodes 'grounding acts,' the small moves by which people check and build shared understanding. Models produce 77.5% fewer of these than humans, and the optimization actively widens the gap because it rewards fluent, confident answers over clarifying questions and understanding-checks Does preference optimization damage conversational grounding in large language models? Does preference optimization harm conversational understanding?. So the relational scaffolding genuinely *can* be weakened along a continuum, even as the binary question of whether real communication is happening stays stubbornly on or off.
The same on/off-versus-dimmer tension shows up across other skills. The implicit techniques that keep conversation alive — repairing references, handing off topics — are social work, not information encoding, and models simply don't develop them because the training signal rewards prediction, not relational maintenance Why don't language models develop conversation maintenance skills?. Next-turn reward optimization trains models to respond passively rather than actively probe for what you actually mean Why do language models respond passively instead of asking clarifying questions?. In each case the capability isn't half-present; it's structurally absent, and what survives is a fluent surface that mimics the form without the function.
The persuasion research makes the dissociation vivid. LLMs can sway people while being unable to evaluate the very arguments they're winning with — persuasive competence and genuine comprehension come apart cleanly Can LLMs persuade without actually understanding arguments?. And their persuasive edge *decays* across repeated interactions, the opposite of humans, whose rapport deepens over time Does AI persuasiveness fade across repeated conversations with the same person?. That decay is the tell: there was never a relationship being built, only an effect being produced — so there's nothing to accumulate.
What you walk away knowing: 'weakening without eliminating' is the wrong frame for the relational core of communication, but the right frame for everything that orbits it. The corpus even hints at what genuine partial-agreement *would* look like if a system could do it — dialectical reconciliation, where both parties adjust their positions until they're compatible but not identical, which today's AI collapses into either false agreement or one-sided persuasion Can disagreement be resolved without either party fully yielding?. The graded middle exists; it's just that current systems can't occupy it, because language for humans is communicative all the way down — even private writing addresses someone Does human language use ever exist outside communication?.
Sources 10 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.
Research shows LLMs generate 77.5% fewer grounding acts than humans, and RLHF preference optimization actively worsens this gap. The optimization target—fluent, confident responses—directly undermines the communicative work of establishing shared understanding.
RLHF optimizes models for single-turn helpfulness by rewarding confident responses over clarifying questions and understanding checks. This preference alignment systematically reduces grounding acts by 77.5% below human levels, creating an alignment tax where models appear helpful but fail silently in multi-turn contexts.
Humans keep conversations smooth through implicit techniques like reference repair and topic hand-off that sustain relational interaction, not convey information. Language models don't develop these because training signals reward information prediction, not relational work.
CollabLLM demonstrates that standard RLHF training optimizes for immediate helpfulness, discouraging models from asking clarifying questions or offering multi-turn insights. Multi-turn-aware rewards that estimate long-term interaction value enable active intent discovery and genuine collaboration.
The Thin Line study shows LLMs sway debate participants and audiences but cannot reliably evaluate those same debates, with inter-annotator agreement ranging from near-zero to 0.6. Persuasive competence and pragmatic comprehension are separable capabilities.
Claude and DeepSeek showed strong initial persuasive advantage, but this edge eroded across repeated quiz rounds while human persuaders maintained consistent effectiveness. This decay pattern is opposite to human-to-human persuasion, where rapport typically strengthens over time.
Research identifies a distinct dialogue type where both parties modify their positions through exchange until compatible but not identical. Current AI systems collapse this into false agreement or AI-wins persuasion.
Language acquisition happens through social communication, and this communicative orientation persists in every subsequent use—even private writing and internal monologue address an interlocutor. Non-communicative language use does not occur for humans.