Asymmetric Lay Beliefs About the Mutual Exclusivity of Interpersonal and Human–AI Communication Preferences postprint

Abstract: As generative artificial intelligence (AI) increasingly becomes a communication partner, interpersonal communication and human–AI communication have emerged as two important ways of satisfying communication needs. However, existing research has examined individuals’ own attitudes toward AI or interpersonal interaction, leaving unclear how observers infer the relationship between another person’s preferences for these two forms. To address this gap, the present research examines whether laypeople hold a mutual-exclusivity belief that liking one form implies disliking the other. We further investigate the consequences of this lay belief for AI product recommendations and social invitations, as well as its underlying mechanism, directional asymmetry, and boundary conditions. We propose a perceived need substitutability model in which observers infer relational and emotional needs from a target’s preference for interpersonal communication and instrumental needs from their preference for human–AI communication. Then, they judge whether the alternative form can satisfy those needs. Because people view AI as capable of meeting instrumental needs but less capable of satisfying relational and emotional needs than humans, a preference for interpersonal communication should produce a stronger negative inference about human–AI communication preference than the reverse.
Seven studies (total N = 2,193) tested these predictions. Studies 1a and 1b compared baseline conditions with conditions describing a target as preferring communication with either humans or AI and measured inferred preference for the alternative form of communication. The results showed that learning that a target preferred one form of communication significantly reduced the inferred preference for the other form, supporting the mutual-exclusivity belief.
Studies 2a and 2b tested whether these inferences affected decisions to target individuals with advertisements for AI social products or invite them to an in-person social activity. These studies replicated the mutual-exclusivity belief and showed its consequences: observers were less willing to target people who preferred interpersonal communication with advertisements for AI social products and less willing to invite people who preferred human–AI communication to in-person social activities.
Study 3 tested the predicted directional asymmetry in a 2 × 2 between-participants design while equating the strength of the provided preference information and clarifying that the alternative preference had not been assessed. A stated preference for interpersonal communication generated a stronger negative inference about AI communication than a stated preference for AI communication generated a negative inference about interpersonal communication.
Study 4 examined perceived need substitutability as a mechanism underlying this asymmetry. Interpersonal communication was seen as more capable of satisfying needs associated with human–AI communication than AI was of satisfying needs associated with interpersonal communication, and perceived need substitutability showed a significant indirect effect.
Study 5 tested topic objectivity as a boundary condition by comparing unspecified and objective topics. When communication concerned objective topics, the direction of the asymmetry reversed.
These findings extend the scope of human–AI interaction research from users’ own attitudes to third-party social inference. They identify asymmetric perceived need substitutability as a mechanism underlying observers’ inferences about others’ communication preferences and show how these inferences shape AI-related targeting and interpersonal inclusion.