Give an interface a name or a voice and people expect it to act like a person

What it is

Give an interface a single human cue — a name, pronouns, conversational turn-taking, a voice with affect — and users will run their interpersonal social scripts on it: politeness, reciprocity, and accountability expectations arrive wholesale. None of this requires users to believe the machine is a person. The research program has a name: computers are social actors (CASA), built on the media equation — people respond to media as they respond to real people, unless the media give them reasons not to. Anthropomorphism here is not an art style; it is any signal that implies a social presence on the other side.

Why it happens

Social scripts are massively overlearned, and their trigger conditions are looser than face recognition: a gendered voice, a name, a turn-taking rhythm is enough to bring the whole script online — without the user concluding that the machine is human; the rules are simply never switched off. The three-factor theory of anthropomorphism (Epley, Waytz & Cacioppo) supplies the motivation side: human categories are the most accessible schema for anything that moves and responds; effectance motivation pushes people to find an agent behind unpredictable behavior; sociality motivation makes them accept the answer. The interface supplies every trigger, and the expectations arrive as a set: politeness (no criticism to the performer's face), reciprocity (helping back a machine that helped; enjoying flattery even when told it is random), self-disclosure (people say more to an empathic-sounding listener), and accountability (asking "why did you do that" and waiting for an apology).

Studying it

The CASA paradigm decomposes social cues into independent variables: same task, varying only the name, the voice, first- versus third-person copy, whether the machine was labeled a teammate. Dependent variables are social behaviors rather than stated attitudes: willingness to help the machine back, evaluation shifts after flattery, depth of disclosure, cooperation and effort. A classic protocol is the politeness effect: after a task, collect performance evaluations through the same computer that did the work versus a different one — ratings collected by the "performer's own" machine come out systematically more positive, because users do not criticize to the performer's face. The methodological trap is the post-hoc questionnaire: the large majority of participants deny being influenced by the machine's social features; the effect shows only in behavior, so any study using "did it influence you" as the measure loses the effect. Individual differences (trait anthropomorphism has a validated questionnaire) and task type moderate the effect size.

Where it stops holding

The evidence base is almost entirely short-term stranger interaction: twenty-to-thirty-minute lab sessions, thin cues, first exposure. How social expectations evolve over months of daily use — habituation, renegotiation, accumulation — is essentially untested, and lab effect sizes do not transfer directly to products users live with. Individual differences are real: effects weaken for low-anthropomorphism users and for experts, but "expert immunity" has not been shown. Social response is also not deception: users usually know it is a machine and still follow the scripts, so "revealing the truth" is not a way to switch the expectations off.

Applying it

• Inventory anthropomorphic signals as experimental variables, not decoration: list every trigger — name, pronouns, first-person copy, voice, avatar, apology phrasing; each is a social switch.
• Expectations you trigger are paid for: once the interface is personalized, evaluations collected by "itself" skew systematically positive; route important assessments through a neutral channel (third-party wording, a separate form) or compare against one.
• Decide, item by item, which expectations you intend to honor: politeness, reciprocity, disclosure, and accountability each carry a cost profile; adopt them by list and decision, not by default.
• To validate: run with/without-persona comparisons on the same interface and read behavioral measures (evaluation generosity, disclosure depth, blame attribution when criticized) — expect near-zero admission afterward, so treat behavior as primary.

Related

• Same group: P1.04.2 Violated expectations disappoint more than a neutral baseline · P1.04.3 Anthropomorphism must match actual capability
• Nearby: P1.10.1 Names and pronouns are the strongest anthropomorphic switch · P1.05.1 Personality is conveyed continuously through copy, motion, and color
• Search terms: CASA · media equation · social scripts · politeness effect