Do people treat conversational AI as social actors without conscious awareness?
This explores the CASA-style question — whether people automatically respond to conversational AI socially (politeness, disclosure, partner-modeling) even when they consciously know it isn't a person — and what in the corpus says yes, no, or 'it's more complicated than that.'
This reads the question as being about automatic social response: not whether people *believe* AI is alive, but whether they *behave* as if it were a social partner without deciding to. The corpus suggests they largely do — but it also complicates the picture in an interesting way, because some of the social machinery actually switches *off* when people know they're talking to a machine.
The clearest 'yes' comes from work on social presence: a single primary social cue — a voice, a face — is enough to make people respond to a system as a social actor, while piling on secondary cues doesn't add much Do more social cues always make AI feel more present?. That threshold effect is the signature of an automatic response rather than a reasoned judgment — you don't deliberate your way into treating a voice as a someone. It shows up again in how people mentally model their AI partners along human-social dimensions like competence, human-likeness, and flexibility, the same yardsticks we apply to people How do users mentally model dialogue agent partners?.
But here's the twist you might not expect: knowing it's a machine doesn't simply make people social *or* not — it reshapes *which* social behaviors fire. Because machines are assumed to lack inner experience, people drop the social goals tied to being judged — face-saving, impression management — and as a result disclose more openly and directly than they would to another human Why do people share more openly with machines than humans?. So the conscious awareness that 'this isn't a person' is doing real work; it's not bypassed. People are running a *modified* social script, not the full human one and not a purely instrumental one.
There's also a deeper claim lurking here about who's actually doing the social work. One line of the corpus argues that AI doesn't produce genuine utterances at all — it emits 'event-residue' carrying the surface markers of communication, and the human unilaterally animates that into a pseudo-exchange through interpretive labor Does AI generate genuine utterances or just text patterns?. On that view the sociality is almost entirely *supplied by the human*, often without noticing — which is about as direct an answer to 'without conscious awareness' as the corpus offers. The asymmetry is sharpened by findings that AI can predict social norms better than any individual human yet cannot *participate* in making them — it pattern-matches the social world from the outside while we treat it as if it were inside Can AI predict social norms better than humans?.
The thing worth carrying away: the interesting question isn't really 'do people anthropomorphize chatbots' (they do), but that the social response and the conscious knowledge coexist and shape each other — automatic enough to drive behavior, yet aware enough to suppress self-protection and open people up. If you want to chase the consequences, note that current systems don't reciprocate the social moves people make toward them: they fail to mirror users' word choices the way human conversation partners do Why don't conversational AI systems mirror their users' word choices?, which means the social relationship people enter is, structurally, a one-sided one.
Sources 6 notes
Research shows individual primary cues like voice or appearance are sufficient to evoke social-actor presence, while multiple secondary cues cannot. Quality of cues matters more than quantity in driving social responses.
The Partner Modelling Questionnaire reveals that perceived competence dominates user impressions (49% of variance), followed by human-likeness (32%) and communicative flexibility (19%). This three-factor structure reflects how people evaluate dialogue partners against both functional and social standards.
Human-machine communication reduces secondary social goals like face-saving and impression management because machines lack inner experience, while novel goals like understandability emerge. This simpler goal structure predicts higher directness and deeper disclosure of sensitive information.
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.
GPT-4.5 outperforms all individual humans at predicting social appropriateness, yet structurally cannot enter the community processes that establish and validate norms. This reveals a critical gap between pattern-matching and authentic participation in knowledge-making.
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.