친숙도의 역설: AI 버튜버 정체성이 경험성 지각을 통해 시청 행동에 미치는 조건부 효과

초록

Background and Purpose Virtual YouTubers (VTubers)—digital content creators who perform through animated avatars—have grown into a significant commercial and cultural phenomenon. A defining structural feature of VTubers is the separation between avatar and operator, referred to in VTuber culture as the nakano hito (“person inside”; Lu et al., 2021). Traditionally, this operator has been human, but the emergence of AI-powered VTubers such as Neuro-sama (Gach, 2023) has introduced a new variable: whether the entity behind the avatar is human or artificial intelligence. Prior research indicates that identity disclosure meaningfully alters viewer responses (Lim & Lee, 2023; Muniz et al., 2024), yet the psychological pathway through which this operates in VTuber contexts remains underspecified. This study applies Gray, Gray, and Wegner’s (2007) mind perception theory to examine how disclosed identity influences viewing behavioral intentions through experience perception—the perceived capacity to feel emotions and sensations. A moderated mediation model is tested, with VTuber familiarity as a moderator, alongside an exploratory examination of narrative structure (self-narrative vs. third-person narrative). Theoretical Framework Mind perception theory proposes that humans evaluate others along two dimensions: agency (the capacity to plan and act) and experience (the capacity to feel). Although disclosed identity may shape both dimensions, experience was selected as the focal mediator because subscription and recommendation intentions are affective, relational responses more closely tied to the perceived capacity to feel than to goal-directed agency (Waytz et al., 2010). In the VTuber context, the human/AI distinction serves as a categorical cue that shapes experience attribution: human operators suggest that genuine emotion lies behind the avatar, whereas AI operators may suppress such attribution. VTuber familiarity was introduced as a boundary condition, given that viewers of this relatively new medium differ substantially in prior exposure and thus in how they interpret identity information. Narrative structure was examined exploratorily, drawing on narrative transportation theory (Green & Brock, 2000) and the continuum model of impression formation (Fiske & Neuberg, 1990). Method A 2 (Identity: human vs. AI) × 2 (Narrative structure: self-narrative vs. third-person narrative) between-subjects online experiment was conducted with 300 Korean young adults aged 19–24 (M = 22.25, SD = 1.55), recruited through a professional survey panel. Participants were randomly assigned to one of four conditions and exposed to an approximately 3-minute VTuber video featuring a custom character (“Rossy”). Identity was manipulated through persistent on-screen text above the video frame and repeated in-video captions; avatar appearance, background music, and screen layout were held constant across all conditions. Narrative structure was manipulated by varying whether the VTuber spoke from first-person experience or provided third-person informational content. Experience and agency were measured using Gray et al.’s (2007) mind perception scale (experience α = .98; agency α = .95). Behavioral intention was assessed with three items adapted from Sundar et al. (2009; α = .95). VTuber familiarity was measured via a single-item 7-point scale prior to stimulus exposure. Moderated mediation was tested using OLS regression with 5,000-iteration percentile-based bootstrap confidence intervals, and the Johnson–Neyman technique was used to identify the precise range of familiarity over which the conditional effect was significant. Results Randomization checks confirmed group equivalence across all demographic and familiarity variables (all ps > .05). H1 (supported). A 2×2 ANOVA revealed a significant main effect of identity on experience perception, F(1, 296) = 29.47, p < .001, η² = .090. The human condition (M = 4.70, SD = 1.78) significantly exceeded the AI condition (M = 3.57, SD = 1.82). Neither narrative structure nor the identity × narrative structure interaction reached significance. H2 (supported). Experience perception significantly and positively predicted behavioral intention (B = 0.338, SE = 0.047, p < .001). Notably, a positive direct effect of AI identity (coded 1; human = 0) on behavioral intention was also observed (B = 0.573, p = .001), yielding an inconsistent mediation structure (MacKinnon et al., 2000) in which a negative indirect path via experience and a separate positive direct pathway coexist—rendering the total effect non-significant (r = .060, p = .297). H3 and H3a (supported). Familiarity significantly moderated the identity → experience relationship (B = −0.362, SE = 0.149, p = .016). Counter to expectation, higher familiarity was associated with larger gaps in experience perception between human and AI conditions. Johnson–Neyman analysis identified a transition point at familiarity score 2.941 (below the sample mean of 4.41), indicating that the conditional effect was significant for the majority of participants. The index of moderated mediation was −0.122 (95% CI [−0.238, −0.021]), with significant conditional indirect effects at all three levels of familiarity (−1SD: −0.211; M: −0.380; +1SD: −0.549). A supplementary analysis using agency as the mediator yielded a parallel pattern, and the very high correlation between experience and agency (r = .880) suggests that the two dimensions, though statistically distinguishable, were closely intertwined in this VTuber context—indicating that identity cues shaped mind attribution as a whole rather than experience alone. RQ. Narrative structure produced no significant effect on either experience perception or behavioral intention, in the ANOVAs or in the moderated regression analysis. Discussion and Implications Holding avatar appearance constant, this study shows that disclosed identity—rather than visual features—drives differential experience attribution in VTuber contexts, extending mind perception theory to a new class of digital media entity. The paradoxical moderation pattern is interpreted through the concept of category sensitivity: viewers more familiar with VTuber culture appear to possess refined schemas for the human/AI distinction, rendering them more—not less—responsive to identity cues. This parallels social identity theory's account of how category salience amplifies intergroup differentiation (Tajfel & Turner, 1979), though this interpretation is post-hoc and warrants direct testing in future research. The inconsistent mediation structure indicates that AI VTubers elicit a dual response: reduced emotional connection via lowered experience perception, and simultaneously maintained or heightened engagement via an independent pathway—possibly related to content utility or novelty, though this mechanism was not directly measured in the present study. The Johnson–Neyman threshold (2.941) may serve as an exploratory reference point for audience segmentation rather than a definitive practical criterion. Limitations include the absence of direct manipulation checks, potential confounding in the narrative structure manipulation, and single-item measurement of familiarity. In addition, although confirmatory factor analysis supported a two-factor structure over a one-factor model, the strong correlation between experience and agency indicates limited discriminant validity in this context, suggesting that the two mind-perception dimensions should be measured and modeled with greater care in future research. Future studies should also directly measure the mechanism underlying the positive direct effect.

키워드