Adoption in human-like AI system: A moderated-mediation approach

Abstract

Human-like AI systems increasingly interact with users as partners, but it is still unclear how perceived human-likeness leads to adoption and when transparency helps or dilutes that effect. We propose a sequential moderated-mediation model in which social presence (perceived human-likeness) increases certainty, certainty increases trust, and trust increases intention to use the system. Transparency is expected to weaken the social presence-to-certainty link. We tested the model in an online 2 × 3 × 2 between-subjects experiment that varied agency locus (human-programmed vs. AI-agency), transparency (none vs. placebo vs. genuine explanation), and decision context (fake-news verification vs. friending suggestions). Data from 491 U.S. adults recruited via Qualtrics Panels were analyzed with structural equation modelling and robustness checks using a propensity-score matched sample (N = 373). Results support the serial mechanism: the indirect effect of social presence on use intention through certainty and trust was significant (β = 0.124, p < .001). Transparency significantly dampened this pathway (moderated-mediation index = −0.093, p = .018), with the conditional indirect effect declining from 0.351 at low transparency to 0.166 at high transparency. These findings show that human-like cues mainly work by reducing uncertainty, but their value depends on how transparent the system is, offering guidance for designing AI partners.