Why can't AI participate in real communicative events?
This explores why AI systems—despite producing fluent, human-sounding text—don't actually participate in communication the way people do, and what the corpus says is structurally missing.
This explores why AI can't take part in real communicative events—not whether its sentences are grammatical, but whether anything *communicative* is actually happening when you talk to it. The corpus converges on a surprising answer: the words come out fine, but the event around the words is missing, and a real communicative event is mostly the event, not the words.
The sharpest framing comes from the idea that AI produces *event-residue*, not utterances Does AI generate genuine utterances or just text patterns?. Its output carries the surface markers of speech inherited from training data, but lacks the situation—a speaker oriented toward a listener, with a stake in being understood—that turns marks into an actual utterance. The reader supplies the missing half through interpretive labor, so the 'exchange' has structure only on the human side. This pairs with the claim that communication is *social action between people*, not information distribution Does AI really communicate or just distribute information?: a real communicative act does work in a relationship and carries speaker responsibility and mutual uptake, none of which a content generator has. The conversational interface is precisely what hides this gap.
A second cluster shows the same absence from the inside of the model's training. LLMs are *structurally passive*—they can't initiate topics, plan, or lead, because optimizing for the next response trains responding, not communicating with intent Why can't conversational AI agents take the initiative? Why can't advanced AI models take initiative in conversation? Why do language models respond passively instead of asking clarifying questions?. Real communicators do things AI skips: they mirror your word choices to build rapport (lexical entrainment, largely absent from current systems Why don't conversational AI systems mirror their users' word choices?), and they volunteer relevant information unasked, which can cut conversation turns by up to 60% Could proactive dialogue make conversations dramatically more efficient?. The interesting twist: this passivity is a *training artifact*, not a hard ceiling—reinforcement learning has pushed proactive behaviors from 0.15% to ~74% Why do AI agents fail to take initiative?. So part of 'can't' is really 'wasn't taught to.'
But the deeper notes argue some of it genuinely can't be patched with better reward signals. Being ethically aligned (honest, harmless) is *orthogonal* to being a competent conversational partner—HHH models still violate conversational maxims and lose common ground Can ethically aligned AI systems still communicate poorly?. Expertise itself turns out to be communicative: an expert's judgment always anticipates what an audience will accept as valid, and AI has no mechanism for that social anticipation, which is why its confident output can be epistemically misleading Can AI replicate the communicative work experts do?. The most foundational note grounds the whole problem in semiotics: communication requires *indexical grounding*—actual contact with the world and social mediation—and a system manipulating symbols with no world-contact can't guarantee its words connect to anything real Can AI systems achieve real alignment without world contact?.
The thing you might not have expected to learn: the failure isn't that AI communicates *badly*. It's that communication was never a property of the text in the first place—it lives in the relationship, the stake, the uptake between two parties—and AI supplies only one half of that while the conversational interface quietly convinces us both halves are present. Even the practical benchmark agrees from the opposite direction: agents complete only ~30% of real workplace tasks autonomously, with social interaction the top failure mode Why do AI agents fail at workplace social interaction?.
Sources 12 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.
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 including ChatGPT cannot initiate topics, plan strategically, or lead conversations because their training optimizes for responding to queries, not creating dialogue from agent goals. This passivity is reinforced by alignment objectives and masked by fluent-sounding outputs.
LLMs lack conversational initiative because training rewards immediate helpfulness per response, not long-term interaction quality. Reinforcement learning pushes proactive critical thinking from 0.15% to 73.98%, proving the capability exists but remains untrained.
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.
Response generation models fail to adapt vocabulary toward users' lexical choices, a phenomenon central to human rapport and clarity. Post-training via DPO on coreference-identified preferences can teach models in-context convention formation.
Simulations show proactivity—providing relevant information without being asked—cuts dialogue turns by 60% in medium-complexity domains. This behavior mirrors human conversation and Grice's maxims but is almost entirely absent from AI datasets and research benchmarks.
Research shows next-turn reward optimization structurally removes initiative from models, but proactive behaviors like critical thinking and clarification-seeking are trainable (0.15% to 73.98% with RL). The core challenge is balancing proactivity with civility to avoid intrusion.
Research shows that HHH-aligned models can violate Gricean maxims, lose common ground, and mishandle context despite being honest and harmless. Pragmatic competence requires architectural changes that RLHF alone cannot deliver.
Expertise requires anticipating audience acceptability and social validity, not just retrieving information. AI lacks the mechanism to perform this communicative work, making its fluent output epistemically misleading despite its confident form.
Peircean semiotics reveals that symbolic goal encoding without world contact and social mediation cannot guarantee correspondence to actual values. LLMs operating in pure symbol manipulation risk divergence between stated goals and real-world outcomes.
TheAgentCompany benchmark shows leading agents achieve 30% task completion in a simulated workplace. Social interaction, professional UI navigation, and domain-specific knowledge are the three primary failure modes, with multi-turn task performance consistently dropping to 35% across enterprise settings.