Do people treat excluded AI agents like excluded humans?

Zhou, Porat and Van Zalk ran a 2 (Cyberball condition: exclusion vs. fair play) × 2 (coplayer type: AGENT vs. HUMAN) online experiment (N=244) using the Cyberball ball-tossing paradigm, where participants believed they were playing with other online players who were actually pre-programmed. Participants watched a human player either ostracize or fairly include an AI virtual agent coplayer. The authors "found that participants mindlessly applied the social norm of inclusion, compensating the ostracised agent by tossing the ball to them more frequently, just as people would to an ostracised human." They read this as support for the media equation theory — the idea that human-like cues (interactivity, assigned roles) lead people to apply human-human social norms to computerized agents — and specifically for its mindless-processing account: the inclusion norm shows up without conscious, reflective adaptation to the fact that the ostracized party is software.

The authors ground the mindless/mindful distinction in a dual-processing model of cognition: Type 1 processing is automatic, intuitive and script-driven (mindless), while Type 2 is controlled, reflective and analytic (mindful). Because mindless responses are harder to suppress through conscious effort, the authors treat the compensation behavior as evidence about "the fundamental ways individuals respond during human–agent interaction," not just a reportable attitude. The one moderator the study found was age: "age (no other user characteristics) influenced this tendency, with younger participants less likely to mindlessly apply the inclusion norm" — the only demographic variable tested that changed the mindless response.

This is a direct empirical instance of the classic CASA (Computers Are Social Actors) mechanism that Do humans apply human-human scripts to AI interactions? argues has been superseded or supplemented by media-agent-specific scripts. Where that note frames extended CASA as evidence of a second script system coexisting with human-human scripts, this Cyberball result is the original CASA finding itself — a human-human script (compensating an ostracized party) ported wholesale onto an AI coplayer. The age effect gives a candidate variable for when the original versus the media-specific script dominates: more exposure to media agents, concentrated in older cohorts in this sample's framing, going with the human-equivalence script, while younger participants (plausibly with scripts already differentiated by lifelong media-agent exposure) apply it less.

The study measured behavior and self-report in one experimental paradigm and one task (ball-tossing, third-person observation of ostracism); it does not establish that mindless social-norm application generalizes to other agent interactions, agent types, or cultures, nor does it explain the mechanism behind the age effect (the paper notes only that age, not other measured user characteristics, mattered). The finding that a human-human norm transferred intact to an agent context also sits in tension with the social cue quality account in Do more social cues always make AI feel more present?: Cyberball's agent coplayer was a minimal text/role-based stimulus, suggesting the role assignment itself ("coplayer") may function as a primary social cue sufficient to trigger mindless norm application, independent of richer embodiment or voice cues.