Recalled AI Experiences Shift Evaluations of AI but Not of Humans or Experts

Recalled AI Experiences Shift Evaluations of AI but Not of Humans or Experts | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Recalled AI Experiences Shift Evaluations of AI but Not of Humans or Experts Kasumi Dan, Takahiro Hoshino This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9267872/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Generative artificial intelligence (AI) is spreading rapidly in everyday life, raising concerns that valuing AI may undermine the perceived value or necessity of humans while benefiting only certain groups. Yet little work tests whether shifts in AI attitudes spill over to evaluations of humans or experts. In a 3x3 between-subjects experiment (N = 987), participants recalled satisfying, dissatisfying, or neutral AI experiences and then evaluated one target (AI, humans, or experts). Recalling satisfying versus dissatisfying experiences shifted trust in AI, perceived competence, continuance intention, and willingness to pay for a plan, with stronger decreases after dissatisfying recall. In contrast, evaluations of humans and experts (trust, competence, and willingness to rely across contexts) showed no significant differences across priming conditions. Thus, we find no evidence that AI attitudes translate into zero-sum devaluation of humans or experts, suggesting largely independent evaluative systems and supporting coexistence of technological and human trust in practice. Biological sciences/Psychology Social science/Psychology Social science/Science technology and society attitudes toward AI priming effects AI usage intention human–AI evaluation algorithm aversion 1. Introduction Recent advances in artificial intelligence (AI), particularly conversational generative AI, have rapidly expanded its user base [1, 2]. As AI has become integrated into society, evidence shows that it improves work quality and efficiency [3, 4] and enhances well-being and quality of life [5, 6, 7]. Yet its development has been accompanied by anxiety, perceived threat, and resistance, and concerns about its negative societal consequences persist [8, 9]. AI-related anxiety and perceived threat have long been discussed, from economic concerns about job displacement [10] to broader fears that AI will dominate or destabilize humanity [11, 12]. A prominent concern is that AI may diminish the value, dignity, or necessity of humans, either collectively or individually [13]. Recent research suggests that anxiety about AI adoption extends beyond economic concerns. A more persistent concern is that individual abilities or social standing may be devalued and AI judgments trusted more than human judgments, undermining human credibility and necessity as decision-making agents. These concerns have been documented as threats to identity, self-worth, and perceived human necessity [14, 15, 16]. Specifically, exposure to AI adoption elicits occupational identity threats, including fears that AI will override individual judgment or diminish others’ respect for them. Likewise, automation threats reduce self-esteem and increase psychological instability [17, 18]. In extreme cases, AI-related developments evoke annihilation anxiety—the fear that humans may no longer be necessary [15]. However, these concerns mainly reflect anticipated declines in one’s own value or credibility and thus remain at the level of perceived anxiety. What remains largely unexamined is whether, as AI advances, people have begun to evaluate others more negatively or to see humans in general as less necessary. In other words, it remains unclear whether people make zero-sum judgments in which increased trust in and reliance on AI are accompanied by a corresponding decrease in human evaluation. Experimental findings suggest this possibility. Prior research has shown that people occasionally follow algorithmic advice over human advice, regardless of objective accuracy or quality, implying that human judgments, including the judgment of an individual and those of others, may be considered inferior to AI judgments [19, 20, 21, 22]. However, because these studies rely on simultaneous comparisons, they cannot distinguish whether preferring AI entails a devaluation of humans or whether AI is simply evaluated more positively while evaluations of humans remain unchanged. Therefore, the present study considered changes in evaluations of AI and evaluations of humans as conceptually independent. It examined whether increased trust in and reliance on AI are accompanied by a tradeoff in which evaluations of humans decline. In this study, participants recalled positive, negative, or neutral AI experiences and then evaluated one target—AI, humans, or experts—on trust, perceived competence, and willingness to rely across multiple task contexts, testing whether effects differed between AI-appropriate and human-appropriate tasks. Because trust and perceived competence are core evaluation dimensions [23, 24], and willingness to rely indicates behavioral translation [25, 26], these were the primary measures. Although the primary goal was to examine potential tradeoffs between AI and human evaluations, we also validated whether recalling satisfying or dissatisfying AI experiences directly shifts overall attitudes toward AI, including continuance intentions and willingness to pay. Establishing such baseline attitude shifts enabled us to assess whether changes in AI evaluations spillover to other targets. 2. Background and Hypotheses 2.1. Effects of AI Perceptions on Evaluations of Humans in General AI is increasingly viewed as a socially comparable outgroup with the potential to impact how humans evaluate themselves and others, rather than only as a tool [15]. Empirical research suggests that AI intelligence functions as a social reference point for human self-understanding. Santoro and Monin [27] show that when people learn that AI is acquiring abilities traditionally considered aspects of “human nature,” such as logical reasoning or communication, they update what they view as essential to being human by emphasizing other traits believed to be distinctively human (e.g., having a personality or feeling love). Similarly, exposure to AI-generated work leads people to evaluate their own creativity and that of humans in general more positively, as creativity is perceived as a skill that AI lacks [28, 29]. In contrast, Kim and McGill [30] found that exposure to autonomous agents with high socio-emotional capabilities leads people to perceive AI as more humanlike while simultaneously judging real humans as lower in humanness. Similarly, Osborne and Bailey [31] showed that reading highly empathetic advice generated by large language models reduces evaluations of the authenticity of the advice of an individual, which is defined as being grounded in the internal values and human qualities of that individual. These studies suggest that people evaluate humans in relation to AI, and that this comparison is bidirectional. When AI highlights what it lacks, perceived human value can rise [27, 28, 29]. When AI’s humanlike qualities and competence are salient, however, perceptions of human authenticity or humanness can decline [30, 31]. Thus, emphasizing AI’s limitations tends to elevate human value, whereas emphasizing its humanlike capabilities tends to diminish it. This pattern concerns not only what humans and AI can both do, but also what remains distinctively human. As long as some capacities are still seen as uniquely human, human value may not be immediately threatened. But as AI acquires abilities that increasingly overlap with human capacities, the domain of the “distinctively human,” and thus a basis of human value, is likely to contract. Malone et al. [32] describe such overlap as a structure that can intensify zero-sum trust dynamics, in which greater trust in AI may come at the expense of trust in humans. Related work suggests that as AI replaces human roles, people may lose opportunities to display their abilities and value, undermining human dignity and perceived necessity [33, 34, 35]. Overall, these findings support a zero-sum view in which gains in AI evaluation may trade off against evaluations of humans. Therefore, when people compare AI with humans, satisfaction with AI may lead them to perceive a reduction in uniquely human capabilities, thereby lowering their evaluations of humans. In contrast, dissatisfaction with AI may heighten the salience of uniquely human capabilities and increase their evaluations of humans. Based on this reasoning, we propose the following hypothesis. H1a. Recalling a satisfying experience with AI will increase trust, perceived competence, and the willingness to rely on AI, while decreasing trust, perceived competence, and the willingness to rely on humans. H1b. Recalling a dissatisfying experience with AI will decrease trust, perceived competence, and the willingness to rely on AI, while increasing trust, perceived competence, and the willingness to rely on humans. 2.2. Effects of AI Perceptions on Evaluations of Human Experts The previous section argued that evaluations of AI and humans may form a zero-sum relationship. Yet people do not view all humans as equally replaceable by AI. Research has long distinguished experts from laypeople [36, 37, 38], and people trust these groups differently: expert opinions are seen as more accurate than one’s own, and expert advice is followed more than nonexpert advice [39, 40, 41]. Experts thus occupy socially recognized positions distinct from the general public [42], implying that human evaluations are stratified rather than uniform. A similar stratified view appears in research on technological change. Computerization and automation disproportionately threaten low-skill, low-wage work, whereas jobs requiring advanced skills or judgment are less affected [43]. Their negative effects on employment and wages fall mainly on routine manual and nonexpert work [44]. More broadly, technology replaces routine, repetitive labor but often complements nonroutine cognitive work, increasing its relative value [45, 46]. Technological progress therefore differentiates between those displaced and those—often experts—whose value is maintained or enhanced through integration with technology. Because AI is also a technology, this logic should extend to AI. Discussions of AI’s social impact likewise adopt a stratified perspective [47], describing AI as extending automation while shifting human work toward either low-skilled nonroutine tasks or high-skilled work requiring professional judgment [48]. Even when AI enters nonroutine cognitive domains such as diagnosis and programming, it need not displace professionals; it can complement their work and raise the value of expertise [49]. Consistent with this view, people still prefer human experts to AI in highly expert domains [50, 51, 52]. AI development may therefore stratify evaluation targets by their relation to AI rather than lowering evaluations of all humans alike. Experts, in particular, may preserve or even increase their value as AI advances. Recent work further suggests that the perceived competence of highly agentic AI is associated with higher competence evaluations of its designers [53]. If experts are seen as those who design, manage, or skillfully use AI, then greater trust in and reliance on AI may spill over into more favorable evaluations of experts. Related research supports this possibility: people project impressions of AI from their evaluations of its users, and trust in technological systems often extends to trust in designers and operators [54, 55]. AI is thus not perceived as an isolated entity but as implicitly linked to the people behind it. From this perspective, experts can be understood as an AI-adjacent group. Satisfaction with AI should therefore increase evaluations of experts, whereas dissatisfaction should decrease them. Based on this reasoning, we propose the following hypothesis. H2a. Recalling a satisfying experience with AI will increase trust, perceived competence, and willingness to rely on experts. H2b. Recalling a dissatisfying experience with AI will decrease trust, perceived competence, and willingness to rely on experts. 2.3. Task-Type Moderation Attitudes toward AI vary systematically with the task characteristics. AI is generally trusted more in tasks with clear criteria, high objectivity, or routine structures, and less in subjective tasks requiring creativity, value judgment, or personalization [19, 56, 57]. In domains involving moral judgment or empathy, resistance to machine decision-making remains strong, even when the output quality is high [58, 59, 60]. Overall, people appear to distinguish between “AI-appropriate tasks” and “tasks that humans should perform,” and base trust and usage intentions on this distinction. The key distinction lies in the asymmetry of choice: AI-appropriate tasks are those that both AI and humans can perform, whereas human-appropriate tasks are those that only humans should perform. In AI-appropriate tasks, using AI is relatively acceptable, and the choice between delegating tasks to AI or humans is more open. Consequently, evaluations are likely to depend less on normative beliefs about who should perform the task and more on performance expectations. Therefore, information regarding prior satisfaction or dissatisfaction with AI should carry a greater weight, leading to larger shifts in attitude. Consequently, human evaluations in general and expert evaluations, as described in H1 and H2, are also more likely to change in these contexts. By contrast, in human-appropriate tasks, the firm belief that humans should perform such tasks limits the perceived legitimacy of delegating them to AI. Therefore, even when AI satisfaction or dissatisfaction is recalled, entrenched preferences for humans and low expectations of AI are likely to constrain attitude changes. Overall, the effects proposed in H1 and H2 are expected to be moderated by task context, with stronger effects for AI-appropriate tasks and weaker effects for human-appropriate tasks. H3. The magnitude of these effects will vary by task context: Stronger in AI-appropriate tasks (e.g., information or technical tasks) and weaker in human-appropriate tasks (e.g., emotional or creative contexts). In addition to the three hypotheses, we examined whether recalling satisfying or dissatisfying AI experiences influences broader attitudes toward AI, specifically, continuance intentions and willingness to pay for a paid plan. These measures capture more reliable and behaviorally proximal attitudes, which serve as prerequisites for the aforementioned evaluative spillover effects. Although this analysis is independent of the central hypothesis, it is informative for understanding the acceptance of generative AI. 3. Materials and Methods 3.1. Research Design The study was approved by the Institutional Review Board of the Institute for Economic Studies, Keio University (Receipt No. 25005), and all procedures were performed in accordance with the Declaration of Helsinki and relevant institutional guidelines and regulations. This work was supported by the Japan Society for the Promotion of Science under Grant JP22H03552. It was registered prior to data collection. We employed a 3 (AI priming: satisfying vs. dissatisfying vs. neutral) × 3 (evaluation target: AI vs. humans vs. experts) fully between-subjects experimental design with participants randomly assigned to one of the nine conditions. The independent variables were the AI experience recall (priming condition) and the evaluation target. The dependent variables were general trust, perceived competence toward the target, and willingness to rely on the target across multiple task contexts. Trust and perceived competence were measured as general evaluations independent of the context, whereas willingness to rely was assessed separately for each task context. Task contexts were classified as AI-appropriate or human-appropriate based on a prior pilot study, which enabled us to test whether the effects of AI experience recall were moderated by task type. In addition, all participants reported continuance intentions toward generative AI and willingness to pay for a paid AI plan, which served as supplementary measures of overall AI attitudes and as a validity check for the priming manipulation. 3.2. Participants A preregistered power analysis using G*Power 3.1 (f = 0.20, α = 0.05, power = 0.80; one-way ANOVA) indicated a required sample size of approximately 83 participants per condition. As the effects of the three priming conditions (satisfying, dissatisfying, and neutral) were analyzed separately for each evaluation target (AI, humans, and experts), the target sample size was set at 744. To account for potentially smaller effect sizes, we planned to recruit participants to obtain approximately 900–1,170 valid responses. Participants were recruited from a large Japanese online panel. Eligible participants, Japanese-speaking adults (18+ years old) with prior experience using generative AI, were randomly assigned to one of the nine conditions. They received compensation in accordance with the standards set by the panel provider. After providing informed consent, 1,606 participants completed the survey. Participants who failed the attention check were excluded from the analysis. The attention check item was “When I want to kill time, I would like to talk to (AI/humans/experts). Note: Regardless of your actual opinion, please select ‘1’ for this item.” After exclusion, the final analytical sample comprised 987 participants. To assess demographic balance, gender distribution, and age, these variables were compared across priming conditions (satisfying, dissatisfying, and neutral) within each evaluation target condition (AI, humans, and experts). No significant differences were observed, neither in any target condition (gender: χ ² (2) = 0.13–2.40, ps = 0.301–0.939; age: F(2, 304–342) = 0.09–0.71, ps = 0.493–0.910), nor in the overall sample (gender: χ ² (2) = 0.88, p = 0.643; age: F(2, 984) = 0.43, p = 0.649). These results indicate minimal demographic imbalance across conditions. The sample size and demographic characteristics of each condition are presented in Table 1. 3.3. Procedure After reading an explanation about data usage, confidentiality, and privacy protection, participants provided written informed consent electronically before participating in the study. They were randomly assigned to one of nine conditions (3 AI priming × 3 evaluation targets). For the priming manipulation, participants first recalled their past experiences with generative AI in an open-ended format. In the satisfying condition, they described three positive aspects of their AI use; in the dissatisfying condition, three negative aspects; and in the neutral condition, three general uses of generative AI. At this stage, AI was defined as conversational generative AI based on large language models, with examples provided (e.g., ChatGPT), while nontarget systems (e.g., voice assistants such as Siri) were explicitly excluded to ensure a shared understanding across participants. The full description of the priming materials is provided in Supplementary Table S1. After the priming task, participants evaluated a single target (AI, humans, or experts) for general trust and perceived competence. Subsequently, they reported their willingness to rely on the target across multiple task scenarios, including emotional support and writing tasks; the attention check item was embedded within these measures. In the expert condition only, participants were also presented with a definition of experts as individuals with advanced knowledge and experience in a specific domain who provide advice, judgment, or support in society [61], with examples such as scientists, researchers, physicians, lawyers, and consultants. Finally, all participants completed measures of continuance intention toward generative AI and willingness to pay for a paid AI plan. 3.4. Measures The primary dependent variables were trust, perceived competence, and willingness to rely on an evaluation target (AI, humans, or experts). In addition, overall attitudes toward generative AI were measured using continuance intention and willingness to pay for a paid AI plan. All items were rated on 7-point Likert scales, and mean scores were computed for each construct. The full wording of the item is presented in Supplementary Table S2. 3.4.1. Trust General trust in the evaluation target was measured using a shortened version of the disposition to trust scale [62]. Although initially developed to assess trust in humans, the scale has been adapted for machines by replacing the referent while preserving the item structure (Disposition to Trust Machines [63]), rendering it suitable for the parallel assessment of AI and humans. In the present study, the same item structure was used across conditions, with the target replaced by “AI,” “humans,” or “experts.” The responses ranged from 1 (“strongly disagree”) to 7 (“strongly agree”), with higher scores indicating greater trust. 3.4.2. Perceived Competence Perceived competence was measured using items from the agency–competence (AC) sub-dimension of the agency–communion framework [64]. This construct captures the extent to which the target is perceived as competent. The item wording was maintained constant across conditions, with the subject replaced by the relevant target. All items were rated on a 7-point Likert scale ranging from 1 (“not very competent/efficient/intelligent”) to 7 (“very competent/efficient/intelligent”), with higher scores indicating greater perceived competence. 3.4.3. Willingness to Rely Willingness to rely was measured across eight task contexts (four AI-appropriate and four human-appropriate) classified in a prior pilot study. For each context, participants indicated their agreement with statements about the evaluation target using a 7-point Likert scale (1 = strongly disagree; 7 = strongly agree). Analyses were conducted at two levels: first, at the level of each specific task context, and second, at the task-type level, by averaging the four contexts classified as AI-appropriate and the four classified as human-appropriate. Higher scores indicated a stronger intention to rely on the target in a given context. A pilot study was conducted to validate the classification of the eight task contexts. Participants (N = 60; 31.7% women; mean age = 46.2 years) rated each context on a 7-point scale ranging from 1 (“humans”) to 7 (“AI”) in response to the question of whether humans or AI should perform the task. One-sample t-tests comparing the mean of each item to the scale midpoint (4) showed that contexts designated as AI-appropriate were rated significantly toward AI. In contrast, those designated as human-appropriate were rated significantly toward humans, consistent with the prior classification of researchers. Thus, the assignment of each context to AI-appropriate or human-appropriate tasks was validated by the judgments of participants. Supplementary Table S3 presents the full item wordings and statistical results. 3.4.4. AI Continuance Intention As a standard measure across all conditions, overall attitudes toward AI were assessed using a continuance-intention scale adapted from prior studies on the technology acceptance model [65, 66]. Items assessed the intention to continue using AI over the approaching months or years and were rated on a 7-point Likert scale (1 = strongly disagree; 7 = strongly agree). The original scale targeted general technologies; in this study, the items were adapted to refer specifically to generative AI tools. Higher scores indicated stronger intentions to continue using AI. 3.4.5. Willingness to Pay for a Paid AI Plan As a standard measure across all conditions, the willingness to pay for a paid generative AI plan was measured using items adapted from prior research on subscription intentions for online services [67], with references replaced with target generative AI tools. The items assessed willingness to pay for a paid plan and the perceived value of the plan, using a 7-point Likert scale (1 = strongly disagree; 7 = strongly agree). Higher scores indicated a stronger willingness to subscribe to a paid AI plan. All scales that were not initially developed for this study were translated into Japanese using a standard translation–back-translation procedure. Two bilingual researchers independently translated the items and reconciled discrepancies, and a third bilingual researcher conducted back-translation. This process was repeated until semantic equivalence with the original items was achieved. 3.5. Statistical analysis To examine whether the AI priming manipulation affected overall attitudes toward AI, one-way ANOVAs were conducted on AI continuance intention and willingness to pay for a paid AI plan, both measured across all conditions, with the priming condition (satisfying vs. dissatisfying vs. neutral) as the independent variable. Although the measures share a standard item structure, they differ in referents, and measurement invariance across targets cannot be assumed. Therefore, the preregistered primary analyses were performed separately for each evaluation target (AI, humans, and experts) using one-way ANOVAs to test the effect of the priming condition (satisfying, dissatisfying, and neutral) within each target. Bonferroni-adjusted post hoc comparisons were conducted following significant omnibus effects. After completion of the preregistered analyses, four additional exploratory analyses were conducted to clarify the interpretation and robustness of the results. First, because separate target-specific tests do not by themselves establish whether the magnitude of the priming effect differs across targets, we conducted exploratory follow-up analyses on trust and perceived competence. In these follow-up analyses, scores were standardized within target prior to estimation, and models included priming condition, target, and their interaction. Planned contrasts focused on the satisfying-versus-dissatisfying comparison and the corresponding AI-versus-human and AI-versus-expert difference-of-differences. Second, to quantify evidence for and against satisfying-versus-dissatisfying differences in human and expert evaluations, we conducted exploratory Bayesian independent-samples t-tests. These analyses were conducted for trust, perceived competence, and the aggregate willingness-to-rely indices. We report BF01 using the default Cauchy prior (r = 0.707). Third, to test the preregistered task-type moderation hypothesis more directly, willingness-to-rely responses were also reanalyzed in exploratory item-level mixed-effects models within each target condition. These models included fixed effects of priming condition, task type, and their interaction, with random intercepts for participant and scenario. Finally, as an exploratory robustness check, the principal ANOVAs were repeated in the full sample prior to attention-check exclusion. 3.6. Declaration of generative AI and AI-assisted technologies in the manuscript preparation process During the preparation of this manuscript, the authors made limited use of generative AI tools (ChatGPT, including Deep Research functionality) to support exploratory literature searches and to assist with English-language editing and phrasing. All relevant literature was independently located, read, and evaluated by the authors. The study design, analyses, interpretation of results, and final wording of the manuscript were conducted entirely by the authors, who take full responsibility for the integrity and content of the published work. 4. Results 4.1. Baseline Effects of AI Experience Recall Baseline effects of the priming manipulation on overall attitudes toward AI are presented in Table 2. AI continuance intention showed a significant main effect of priming, F(2, 984) = 6.37, p = 0.002, η² = 0.013. The mean scores were in the following order: satisfying > neutral > dissatisfying. Bonferroni-adjusted comparisons indicated significant differences between the satisfying and dissatisfying conditions and between the neutral and dissatisfying conditions, but not between the satisfying and neutral conditions. A similar pattern emerged for willingness to pay for a paid AI plan, with a significant main effect of priming, F(2, 984) = 4.69, p = 0.009, and η² = 0.009. The mean scores were in the same order: satisfying > neutral > dissatisfying; however, post hoc tests revealed a significant difference only between the satisfying and dissatisfying conditions. Overall, these results indicate that recalling satisfying versus dissatisfying AI experiences influences AI-related behavioral intentions, with negative recall exerting a powerful downward effect. Building on these baseline attitude shifts, the following section examines whether changes in evaluations of AI spillover to evaluations of other targets. 4.2. Effects of AI Experience Recall on Evaluations of Humans and Experts The primary preregistered target-specific analyses are presented in Table 3. 4.2.1. AI Condition Regarding trust in AI, a significant priming effect was observed, F(2, 332) = 3.28, p = 0.039, and η² = 0.019. The mean scores were in the following order: satisfying > neutral > dissatisfying; however, post hoc tests showed a significant difference only between the satisfying and dissatisfying conditions, with no significant difference between the satisfying and neutral conditions. Perceived competence of AI also showed a significant priming effect, F(2, 332) = 10.11, p < 0.001, and η² = 0.057. Post hoc comparisons indicated that the dissatisfied condition was rated significantly lower than both the satisfied and neutral conditions, whereas the satisfied and neutral conditions did not differ significantly. For the more behaviorally proximal measure of willingness to rely, priming effects were largely non-significant across individual task contexts. An exception was observed in the “complex discussion” context, where a significant effect emerged, F(2, 332) = 3.48, p = 0.032, andη² = 0.021, with willingness to rely being lower in the dissatisfying condition (M = 4.00 and SD = 1.69) than in the satisfying condition (M = 4.57 and SD = 1.57). By contrast, no significant priming effects were found for the aggregated indices of AI-appropriate or human-appropriate tasks. Overall, these results suggest that recalling satisfying versus dissatisfying AI experiences was sufficient to shift attitudinal evaluations of AI, particularly trust and perceived competence. In contrast, the priming effects on the willingness to rely were minimal. 4.2.2. Human and Expert Conditions No significant priming effects were observed in either the human or expert condition for trust, perceived competence, or willingness to rely in any task context (all ps > 0.05). Therefore, recalling satisfying or dissatisfying AI experiences did not significantly change evaluations of humans in general or of experts. 4.3. Exploratory follow-up analyses clarifying target specificity and null spillover To test whether the priming contrast was larger for AI than for humans or experts, we conducted exploratory general linear models (Table 4). For trust, the priming-by-target interaction was not significant, p = 0.390. The satisfying-versus-dissatisfying contrast was significant for AI (b = 0.34, p = 0.011) but not for humans (b = 0.09, p = 0.506) or experts (b = 0.14, p = 0.290), and the AI-versus-human and AI-versus-expert difference-of-differences were not significant (ps = 0.196 and 0.277). For perceived competence, the interaction was significant, p = 0.019. The same contrast was significant for AI (b = 0.55, p < 0.001) but not for humans (b = 0.06, p = 0.639) or experts (b = -0.02, p = 0.901), and was larger for AI than for humans (b = 0.49, p = 0.010) and experts (b = 0.57, p = 0.002). We conducted exploratory Bayesian t-tests for human and expert evaluations because non-significant spillover effects alone do not support the null (Table 5). Bayes factors provided moderate evidence for the null for humans (trust: BF01 = 5.38; competence: BF01 = 6.05) and experts (trust: BF01 = 4.29; competence: BF01 = 7.05). For willingness-to-rely outcomes, most comparisons favored the null, although evidence was weak for willingness to rely on humans in AI-appropriate tasks (BF01 = 1.50). Overall, these analyses indicated that the priming effect was target-specific for perceived competence and did not spill over to human or expert evaluations. 4.4. Exploratory follow-up analyses of task-type moderation and robustness To test the preregistered task-type moderation hypothesis, we conducted exploratory item-level linear mixed-effects models for each target. Contrary to H3, the priming-by-task-type interaction was significant only for AI (p = 0.028), reflecting a larger satisfying-versus-dissatisfying contrast in human-appropriate tasks (b = 0.37, p = 0.021) than in AI-appropriate tasks (b = 0.17, p = 0.287). For humans, the interaction was not significant (p = 0.060), although the satisfying-versus-dissatisfying contrast in AI-appropriate tasks trended negative (b = -0.29, p = 0.055). For experts, neither the interaction nor the within-task-type contrasts was significant (ps ≥ 0.369). Thus, the preregistered moderation hypothesis was not supported. We repeated the principal ANOVAs in the full sample prior to attention-check exclusion. Reanalyses of completers (N = 1,606) yielded the same pattern: priming effects on AI trust, F(2, 546) = 3.10, p = 0.046, and perceived competence, F(2, 546) = 9.31, p 0.34). Although an omnibus priming effect emerged for expert trust, F(2, 528) = 3.52, p = 0.030, this was driven by the neutral-versus-dissatisfying comparison (p = 0.025) rather than the satisfying-versus-dissatisfying contrast (p = 0.611). Accordingly, the main conclusions were not dependent on attention-check exclusion. 4.5. Summary of Hypothesis Tests Although evaluations of AI were affected by priming manipulation, evaluations of humans in general were not, providing no support for H1, which predicts a tradeoff between evaluations of AI and humans. Similarly, because AI experience recall did not significantly affect evaluations of experts, H2, which predicts a corresponding spillover to experts, was not supported. Exploratory direct-comparison analyses showed that the target specificity of the priming effect was clearest for perceived competence, and exploratory Bayesian analyses provided moderate support for the absence of corresponding effects on human and expert trust and competence. Finally, exploratory item-level follow-up analyses did not support the preregistered task context moderation predicted in H3. 5. Discussion This study examined concerns that the rise of generative AI may undermine human values or necessity [13], focusing on whether such concerns are reflected in actual evaluative tendencies. To avoid considering humans as a homogeneous category, we distinguished between humans in general and experts who are theoretically less susceptible to replacement and more closely associated with the design and use of AI. Using an experiential priming manipulation that induced recall of satisfying or dissatisfying AI experiences, we assessed evaluations of AI, humans in general, and experts to test whether evaluations of AI and humans operate in a zero-sum manner and whether evaluations of AI spillover to experts as an AI-adjacent group. 5.1. Interpreting Changes in Evaluations of AI The results indicate that recalling positive or negative experiences with generative AI can shift general attitudes toward AI, including continuance intentions, willingness to subscribe to paid plans, and evaluations of trust and competence. Notably, this effect was more pronounced as a decline following a dissatisfying recall than as an increase following a satisfying recall. In other words, recalling satisfying experiences did not lead to significantly more positive AI evaluations than in the neutral condition, whereas recalling dissatisfying experiences was associated with significantly more negative evaluations. Although this asymmetry is not explicitly hypothesized, it is consistent with prior research on algorithm aversion. Previous studies have suggested that people are sensitive to algorithmic errors and may discount algorithmic judgments after observing even a single failure, occasionally favoring less accurate human judgments instead [26]. From this perspective, the more substantial negative shift observed in the dissatisfying condition in addition to the limited positive change in the satisfying condition, may reflect an asymmetry in how people update their attitudes toward AI based on negative versus positive experiences. 5.2. Implications of the Absence of Spillover Effects As mentioned previously, the priming manipulation influenced AI evaluation, particularly trust and perceived competence. However, no significant priming effects were observed for evaluations of humans in general or experts; thus, the hypotheses predicting spillover from AI evaluations to human or expert evaluations were not supported. Recalling satisfaction or dissatisfaction with AI altered attitudes toward it but did not produce corresponding changes in human or expert evaluations. Exploratory follow-up analyses sharpened this interpretation. Direct-comparison analyses showed that the target specificity of the priming effect was clearest for perceived competence, for which the satisfying-versus-dissatisfying contrast was significantly larger for AI than for humans or experts. Exploratory Bayesian analyses further provided moderate support for the absence of corresponding effects on human and expert trust and perceived competence. This pattern was also robust in reanalyses of the full sample prior to attention-check exclusion. This pattern suggests that evaluations of AI and of humans or experts may not be linked in a zero-sum or tightly coupled manner, but may instead be psychologically independent. Within the scope of the present findings, existential concerns that AI advancement will directly undermine human trustworthiness, competence, and necessity do not appear to be reflected in people’s evaluative judgments. Similarly, the results do not support a simple linkage in which more favorable evaluations of AI translate into higher evaluations of experts associated with AI, suggesting that AI and experts may be perceived as distinct evaluative targets. Several possible explanations exist for the absence of spillovers. One possibility is that people still prefer many tasks to be performed by humans [68, 69], leading them to maintain beliefs in uniquely human roles and values. Another is that, given warnings that anthropomorphizing AI, a mere machine, and comparing it with humans is inappropriate and harmful [70, 71], people may not regard AI as a social entity or may actively avoid doing so. Conversely, because people consider AI an autonomous agent [72], evaluations of experts as its designers or managers may not covary with evaluations of AI. Moreover, prior research suggests that exposure to AI development can elicit defensive reactions that uniquely reinforce beliefs about human values [73, 27]. Such processes may not have been detectable through self-reported measures, suggesting that zero-sum comparisons may operate at a more implicit level. Overall, these considerations suggest that the mechanisms underlying the present findings are likely complex, and future research is needed to clarify which factors play the most substantial role in shaping human and expert evaluations in the context of AI advancement. 5.3. Interpreting Findings on Willingness to Rely Another noteworthy finding is that recalling satisfying or dissatisfying AI experiences broadly shifted attitudes toward AI, but had limited effects on willingness to rely across task contexts. Exploratory item-level linear mixed-effects analyses likewise did not support the preregistered task-type moderation hypothesis. The priming-by-task-type interaction was significant only for AI, and this effect was larger in human-appropriate than in AI-appropriate tasks, contrary to prediction. For humans, the corresponding interaction was not significant, although the satisfying-versus-dissatisfying contrast in AI-appropriate tasks trended in the negative direction. For experts, neither the interaction nor the within-task-type contrasts were significant. A more plausible explanation is that, in contrast to other measures, the willingness-to-rely measure requires respondents to consider concrete situations. Prior research indicates that relating messages to the context of an individual increases elaboration [74], weakening priming effects [75]. Therefore, willingness-to-rely judgments may be more stable than general attitudes and less susceptible to a brief recall manipulation, while also being more strongly anchored in pre-existing beliefs about task appropriateness. 5.4. Limitations This study had several limitations. Although exploratory follow-up analyses helped clarify the null spillover findings, they were not preregistered and should therefore be interpreted cautiously. In addition, while multiple mechanisms may underlie the present findings, the study design did not permit their direct testing. Our conclusions are thus limited to suggesting that increases in AI evaluations do not necessarily accompany changes in human evaluations and do not support stronger claims about enduring beliefs in uniquely human values. Relatedly, the priming manipulation focused on satisfaction or dissatisfaction with AI use rather than more existential dimensions such as agency, autonomy, or sociality. Recalling use experiences may have encouraged participants to view AI as a convenient tool, reducing the likelihood of perceiving it as a social entity comparable to humans. Future studies should examine whether similar patterns emerge when recall manipulations target AI’s autonomy or humanity. Finally, although experts were introduced as a group theoretically positioned to benefit from AI complementarity, we did not assess whether participants perceived them as AI designers, managers, or value-enhancing agents, so these stratified effects may not have been fully captured. 5.5. Contributions Despite these limitations, this study makes several contributions to the literature. Using a sufficiently large sample and a preregistered design, we demonstrated that recalling dissatisfying AI experiences leads to broad declines in attitudes toward AI, including willingness to pay for a paid plan. This finding suggests that in promoting AI services, reducing errors and emphasizing safety and reliability may be more important than highlighting superior performance alone, providing practical implications for AI service design and marketing. In addition, because a simple recall-based priming manipulation shifted attitudes and behavioral intentions toward AI, this approach may be helpful for future experimental research examining AI-related attitudes. The primary contribution of this study is its approach to the widely discussed concern that AI undermines human values, replaces humans, or exacerbates inequality. Rather than focusing on whether people merely feel such anxieties, we examined how they evaluated humans in general and experts after forming positive or negative attitudes toward AI. The exploratory follow-up analyses sharpened this interpretation by showing that target specificity was clearest for perceived competence and by providing moderate Bayesian support for the absence of corresponding effects on human and expert trust and competence. Thus, the results showed that AI-related experiences altered evaluations of AI itself, but did not produce evidence of a broad zero-sum devaluation of humans or experts. These findings suggest that some contemporary concerns about the negative societal impact of AI may be overstated and necessitate a more nuanced view of how AI development relates to human values. Finally, although the societal effects of AI are undoubtedly complex and cannot be reduced to simple positive or negative outcomes, we hope that this study contributes to a more balanced understanding of how humans and AI may coexist in practice. Declarations Acknowledgements The authors would like to thank the Keio Economic Research Institute for its support in English-language editing. We are also grateful to Ryoya Nakano for his valuable assistance with the Japanese translation of the measurement scales, as well as for his support in refining the English wording of the Japanese stimulus materials used in the study. In addition, we acknowledge the Japan Science and Technology Agency (JST) for providing administrative and institutional support related to the authors’ academic activities during the course of this research. Author Contributions Kasumi Dan contributed to conceptualization, methodology, data curation, formal analysis, investigation, software implementation, validation, and drafting of the original manuscript, as well as reviewing and editing the manuscript. Takahiro Hoshino contributed to conceptualization, investigation, supervision, project administration, funding acquisition, validation, and critical review and editing of the manuscript. Both authors approved the final version of the manuscript and agree to be accountable for all aspects of the work. Additional Information Competing interests The authors declare no competing interests. Funding Declaration This work was supported by the JSPS KAKENHI under Grant JP22H03552. Data Availability The anonymized quantitative dataset and analysis code are available at the Open Science Framework (OSF): https://osf.io/g3yjh/overview?view_only=3a76dd0e579740378040a6f39e7a61ff Open-ended responses are not publicly available because they may contain potentially identifiable information in free-text form. Access requests will be considered by the corresponding author on reasonable request, subject to ethical and privacy restrictions. References OpenAI. (2025). How people are using ChatGPT. https://openai.com/index/how-people-are-using-chatgpt/ Hussein, H., Gordon, M., Hodgkinson, C., Foreman, R., & Wagad, S. (2025). ChatGPT’s Impact Across Sectors: A Systematic Review of Key Themes and Challenges. Big Data and Cognitive Computing, 9(3), 56. https://doi.org/10.3390/bdcc9030056 Cui, H., & Yasseri, T. (2024). AI-enhanced collective intelligence. Patterns, 5(11), 101074. https://doi.org/10.1016/j.patter.2024.101074 Luo, X., Rechardt, A., Sun, G., Nejad, K. K., Yáñez, F., Yilmaz, B., Lee, K., Cohen, A. O., Borghesani, V., Pashkov, A., Marinazzo, D., Nicholas, J., Salatiello, A., Sucholutsky, I., Minervini, P., Razavi, S., Rocca, R., Yusifov, E., Okalova, T., . . . Love, B. C. (2025). Large language models surpass human experts in predicting neuroscience results. Nature Human Behaviour, 9(2), 305-315. https://doi.org/10.1038/s41562-024-02046-9 Li, H., Zhang, R., Lee, Y., Kraut, R. E., & Mohr, D. C. (2023). Systematic review and meta-analysis of AI-based conversational agents for promoting mental health and well-being. NPJ Digital Medicine, 6(1), 236. https://doi.org/10.1038/s41746-023-00979-5 Rehman, A. U., Behera, R. K., Islam, M. S., Abbasi, F. A., & Imtiaz, A. (2024). Assessing the usage of ChatGPT on life satisfaction among higher education students: The moderating role of subjective health. Technology in Society, 78, 102655. https://doi.org/10.1016/j.techsoc.2024.102655 Feng, X., Tian, L., Ho, G. W. K., Yorke, J., & Hui, V. (2025). The Effectiveness of AI Chatbots in Alleviating Mental Distress and Promoting Health Behaviors Among Adolescents and Young Adults: Systematic Review and Meta-Analysis. Journal of Medical Internet Research, 27, e79850. https://doi.org/10.2196/79850 Wang, Y.-Y., & Wang, Y.-S. (2022). Development and validation of an artificial intelligence anxiety scale: An initial application in predicting motivated learning behavior. Interactive Learning Environments, 30(4), 619–634. https://doi.org/10.1080/10494820.2019.1674887 Dong, M., Conway, J. R., Bonnefon, J., Shariff, A., & Rahwan, I. (2024). Fears about artificial intelligence across 20 countries and six domains of application. The American Psychologist, https://doi.org/10.1037/amp0001454 Jo, H., & Park, D.-H. (2025). The fear of being replaced by generative AI: An examination of influential factors among office workers. Technological Forecasting and Social Change, 220, 124326. https://doi.org/10.1016/j.techfore.2025.124326 De Freitas, J., Agarwal, S., Schmitt, B., & Haslam, N. (2023). Psychological factors underlying attitudes toward AI tools. Nature Human Behaviour, 7(11), 1845-1854. https://doi.org/10.1038/s41562-023-01734-2 Bales, A. (2025). A polycrisis threat model for AI. AI & Society, 40(8), 6277–6289. https://doi.org/10.1007/s00146-025-02371-1 Selenko, E., Bankins, S., Shoss, M., Warburton, J., & Restubog, S. L. D. (2022). Artificial intelligence and the future of work: A functional-identity perspective. Current Directions in Psychological Science, 31(3), 272–279. https://doi.org/10.1177/09637214221091823 Alkhalifah, J. M., Bedaiwi, A. M., Shaikh, N., Seddiq, W., & Meo, S. A. (2024). Existential anxiety about artificial intelligence (AI)- is it the end of humanity era or a new chapter in the human revolution: questionnaire-based observational study. Frontiers in Psychiatry, 15, 1368122. https://doi.org/10.3389/fpsyt.2024.1368122 Frenkenberg, A., & Hochman, G. (2025). It’s Scary to Use It, It’s Scary to Refuse It: The Psychological Dimensions of AI Adoption—Anxiety, Motives, and Dependency. Systems, 13(2), 82. https://doi.org/10.3390/systems13020082 Shonhe, L., & Min, Q. (2025). Human-machine collaboration: Exploring professional identity threat within the records and information management community. ASLIB Journal of Information Management. Advance online publication. https://doi.org/10.1108/AJIM-08-2024-0673 Goethals, F., & Ziegelmayer, J. L. (2022). Anxiety buffers and the threat of extreme automation: a terror management theory perspective. Information Technology & People, 35(1), 96-118. https://doi.org/10.1108/ITP-06-2019-0304 Mirbabaie, M., Brünker, F., Möllmann Frick, Nicholas R. J., & Stieglitz, S. (2022). The rise of artificial intelligence – understanding the AI identity threat at the workplace. Electronic Markets, 32(1), 73-99. https://doi.org/10.1007/s12525-021-00496-x Castelo, N., Bos, M. W., & Lehmann, D. R. (2019). Task-dependent algorithm aversion. Journal of Marketing Research, 56(5), 809–825. https://doi.org/10.1177/0022243719851788 Logg, J. M., Minson, J. A., & Moore, D. A. (2019). Algorithm appreciation: People prefer algorithmic to human judgment. Organizational Behavior and Human Decision Processes, 151, 90–103. https://doi.org/10.1016/j.obhdp.2018.12.005 You, S., Cathy, L. Y., & Li, X. (2022). Algorithmic versus Human Advice: Does Presenting Prediction Performance Matter for Algorithm Appreciation? Journal of Management Information Systems, 39(2), 336-365. https://doi.org/10.1080/07421222.2022.2063553 Klingbeil, A., Grützner, C., & Schreck, P. (2024). Trust and reliance on AI: An experimental study on the extent and costs of overreliance on AI. Computers in Human Behavior, 160, 108352. https://doi.org/10.1016/j.chb.2024.108352 Fiske, S. T. (2018). Stereotype Content: Warmth and Competence Endure. Current Directions in Psychological Science, 27(2), 67-73. https://doi.org/10.1177/0963721417738825 Glikson, E., & Woolley, A. W. (2020). Human trust in artificial intelligence: Review of empirical research. Academy of Management Annals, 14(2), 627–660. https://doi.org/10.5465/annals.2018.0057 Lee, J. D., & See, K. A. (2004). Trust in Automation: Designing for Appropriate Reliance. Human Factors, 46(1), https://doi.org/10.1518/hfes.46.1.50_30392 Dietvorst, B. J., Simmons, J. P., & Massey, C. (2015). Algorithm aversion: people erroneously avoid algorithms after seeing them err. Journal of Experimental Psychology.General, 144(1), 114-26. https://doi.org/10.1037/xge0000033 Santoro, E., & Monin, B. (2023). The AI Effect: People rate distinctively human attributes as more essential to being human after learning about artificial intelligence advances. Journal of Experimental Social Psychology, 107, 104464. https://doi.org/10.1016/j.jesp.2023.104464 Horton Jr, C. B., White, M. W., & Iyengar, S. S. (2023). Bias against AI art can enhance perceptions of human creativity. Scientific Reports, 13(1), 19001. https://doi.org/10.1038/s41598-023-45202-3 Reich, T., & Teeny, J. D. (2025). Does artificial intelligence cause artificial confidence? Generative artificial intelligence as an emerging social referent. Journal of Personality and Social Psychology, https://doi.org/10.1037/pspa0000450 Kim, H.-Y., & McGill, A. L. (2025). AI-induced dehumanization. Journal of Consumer Psychology, 35(3), 363–381. https://doi.org/10.1002/jcpy.1441 Osborne, M. R., & Bailey, E. R. (2025). Me vs. the machine? Subjective evaluations of human- and AI-generated advice. Scientific Reports, 15, 3980. https://doi.org/10.1038/s41598-025-86623-6 Malone, E., Afroogh, S., & D'Cruz, J. (2025). When Trust Is Zero Sum: Automation's Threat to Epistemic Agency. Ethics and Information Technology, 27(2), 29. https://doi.org/10.1007/s10676-025-09838-1 Bankins, S., & Formosa, P. (2023). The Ethical Implications of Artificial Intelligence (AI) For Meaningful Work: JBE. Journal of Business Ethics, 185(4), 725-740. https://doi.org/10.1007/s10551-023-05339-7 Dang, J., & Liu, L. (2025a). Dehumanization risks associated with artificial intelligence use. The American Psychologist, https://doi.org/10.1037/amp0001542 Ferdman, A. (2025). AI deskilling is a structural problem. AI & Society. https://doi.org/10.1007/s00146-025-02686-z Gieryn, T. F. (1983). Boundary-work and the demarcation of science from non-science: Strains and interests in professional ideologies of scientists. American Sociological Review, 48(6), 781–795. https://doi.org/10.2307/2095325 Haas, P. M. (1992). Epistemic communities and international-policy coordination: Introduction. International Organization, 46(1), 1–35. https://doi.org/10.1017/S0020818300001442 Grundmann, R. (2017). The Problem of Expertise in Knowledge Societies. Minerva: A Review of Science, Learning and Policy, 55(1), 25-48. https://doi.org/10.1007/s11024-016-9308-7 Harvey, N., & Fischer, I. (1997). Taking advice: Accepting help, improving judgment, and sharing responsibility. Organizational Behavior and Human Decision Processes, 70(2), 117-133. https://doi.org/10.1006/obhd.1997.2697 Meshi, D., Biele, G., Korn, C. W., & Heekeren, H. R. (2012). How Expert Advice Influences Decision Making. PLoS One, 7(11). https://doi.org/10.1371/journal.pone.0049748 Bailey, P. E., Leon, T., Ebner, N. C., Moustafa, A. A., & Weidemann, G. (2023). A meta-analysis of the weight of advice in decision-making. Current Psychology, 42, 24516–24541. https://doi.org/10.1007/s12144-022-03573-2 Keating, L. A., & Walker, B. W. (2025). Lay theories of expertise: A mixed-methods exploration. Journal of Management Studies. Advance online publication. https://doi.org/10.1111/joms.13188 Frey, C. B., & Osborne, M. A. (2017). The future of employment: How susceptible are jobs to computerisation? . Technological Forecasting and Social Change, 114, 254-269,278-280. https://doi.org/10.1016/j.techfore.2016.08.019 Acemoglu, D., & Restrepo, P. (2020). Robots and jobs: Evidence from U.S. labor markets. Journal of Political Economy, 128(6), 2188–2244. https://doi.org/10.1086/705716 Autor, D. H., Levy, F., & Murnane, R. J. (2003). The Skill Content of Recent Technological Change: An Empirical Exploration. The Quarterly Journal of Economics, 118(4), 1279-1333. https://doi.org/10.1162/003355303322552801 Autor, D. H. (2015). Why Are There Still So Many Jobs? The History and Future of Workplace Automation. The Journal of Economic Perspectives, 29(3), 3-30. https://doi.org/10.1257/jep.29.3.3 Jacobs, J. (2024). The artificial intelligence shock and socio-political polarization. Technological Forecasting and Social Change, 199, 123006. https://doi.org/10.1016/j.techfore.2023.123006 Tschang, F. T., & Almirall, E. (2021). Artificial Intelligence as Augmenting Automation: Implications for Employment. The Academy of Management Perspectives, 35(4), 642. https://doi.org/10.5465/amp.2019.0062 Septiandri, A. A., Constantinides, M., & Quercia, D. (2024). The potential impact of AI innovations on US occupations. PNAS Nexus, 3(9), 320. https://doi.org/10.1093/pnasnexus/pgae320 Novozhilova, E., Mays, K., & Katz, J. E. (2024). Looking towards an automated future: U.S. attitudes towards future artificial intelligence instantiations and their effect. Humanities & Social Sciences Communications, 11(1), 132. https://doi.org/10.1057/s41599-024-02625-1 Busch, F., Hoffmann, L., Xu, L., Zhang, L. J., Hu, B., García-Juárez, I., Toapanta-Yanchapaxi, L., Gorelik, N., Gorelik, V., Rodriguez-Granillo, G., Ferrarotti, C., Cuong, N. N., Thi, C. A. P., Tuncel, M., Kaya, G., Solis-Barquero, S., Mendez Avila, M.,C., Ivanova, N. G., Kitamura, F. C., . . . Zhang, S. (2025). Multinational Attitudes Toward AI in Health Care and Diagnostics Among Hospital Patients. JAMA Network Open, 8(6) https://doi.org/10.1001/jamanetworkopen.2025.14452 Jin, M., Yang, Z., Freling, T. L., & Janakiraman, N. (2025). The human superiority effect in advice taking: A multimethod exploration and implications for policy makers and governmental organizations. Journal of Public Policy & Marketing, 44(3), 350–369. https://doi.org/10.1177/07439156251320314 Vanneste, B. S., & Puranam, P. (2025). Artificial intelligence, trust, and perceptions of agency. Academy of Management Review, 50(4), 726–744. https://doi.org/10.5465/amr.2022.0041 Dang, J., & Liu, L. (2025b). Public trust in artificial intelligence users. Current Opinion in Psychology, 66, 102148. https://doi.org/10.1016/j.copsyc.2025.102148 Lumineau, F., Schilke, O., & Wang, W. (2023). Organizational trust in the age of the fourth industrial revolution: Shifts in the nature, production, and targets of trust. Journal of Management Inquiry, 32(1), 21–34. https://doi.org/10.1177/10564926221127852 Longoni, C., & Cian, L. (2022). Artificial intelligence in utilitarian vs. hedonic contexts: The “word-of-machine” effect. Journal of Marketing, 86(1), 91–108. Qin, X., Zhou, X., Chen, C., Wu, D., Zhou, H., Dong, X., Cao, L., & Lu, J. G. (2025). AI aversion or appreciation? A capability-personalization framework and a meta-analytic review. Psychological Bulletin, 151(5), 580-599. https://doi.org/10.1037/bul0000477 Dietvorst, B. J., & Bartels, D. M. (2022). Consumers object to algorithms making morally relevant tradeoffs because of algorithms’ consequentialist decision strategies. Journal of Consumer Psychology, 32(3), 406–424. https://doi.org/10.1002/jcpy.1266 Yin, Y., Jia, N., & Wakslak, C. J. (2024). AI can help people feel heard, but an AI label diminishes this impact. Proceedings of the National Academy of Sciences of the United States of America, 121(14) https://doi.org/10.1073/pnas.2319112121 Myers, S., & Everett, J. A. C. (2025). People expect artificial moral advisors to be more utilitarian and distrust utilitarian moral advisors. Cognition, 256, 106028. https://doi.org/10.1016/j.cognition.2024.106028 Merkley, E. (2020). Anti-intellectualism, populism, and motivated resistance to expert consensus. Public Opinion Quarterly, 84(1), 24–48. https://doi.org/10.1093/poq/nfz053 Gefen, D. (2000). E-commerce: The role of familiarity and trust. Omega, 28(6), 725–737. https://doi.org/10.1016/S0305-0483(00)00021-9 Riedl, R., Hogeterp, S. A., & Reuter, M. (2024). Do patients prefer a human doctor, artificial intelligence, or a blend, and is this preference dependent on medical discipline? Empirical evidence and implications for medical practice. Frontiers in Psychology, 15, 1422177. https://doi.org/10.3389/fpsyg.2024.1422177 Abele, A. E., Hauke, N., Peters, K., Louvet, E., Szymkow, A., & Duan, Y. (2016). Facets of the Fundamental Content Dimensions: Agency with Competence and Assertiveness-Communion with Warmth and Morality. Frontiers in Psychology, 7, 1810. https://doi.org/10.3389/fpsyg.2016.01810 Bhattacherjee, A. (2001). Understanding information systems continuance: An expectation-confirmation model. MIS Quarterly, 25(3), 351–370. https://doi.org/10.2307/3250921 Venkatesh, V., Morris, M. G., Davis, G. B., & Davis, F. D. (2003). User acceptance of information technology: Toward a unified view. MIS Quarterly, 27(3), 425–478. https://doi.org/10.2307/30036540 Chang, V., Yang, Y., Xu, Q. A., & Xiong, C. (2021). Factors influencing consumer intention to subscribe to the premium music streaming services in China. Journal of Global Information Management, 29(6), 1–25. https://doi.org/10.4018/JGIM.20211101.oa17 Mahmud, H., Islam, A. N., Ahmed, S. I., & Smolander, K. (2022). What influences algorithmic decision-making? A systematic literature review on algorithm aversion. Technological Forecasting and Social Change, 175, 1. https://doi.org/10.1016/j.techfore.2021.121390 Afroogh, S., Akbari, A., Malone, E., Kargar, M., & Alambeigi, H. (2024). Trust in AI: Progress, challenges, and future directions. Humanities & Social Sciences Communications, 11(1), Article 1568. https://doi.org/10.1057/s41599-024-04044-8 Placani, A. (2024). Anthropomorphism in AI: Hype and fallacy. AI and Ethics, 4, 691–698. https://doi.org/10.1007/s43681-024-00419-4 Peter, S., Riemer, K., & West, J. D. (2025). The benefits and dangers of anthropomorphic conversational agents. Proceedings of the National Academy of Sciences of the United States of America, 122(22). https://doi.org/10.1073/pnas.2415898122 Bonnefon, J., Rahwan, I., & Shariff, A. (2024). The Moral Psychology of Artificial Intelligence. Annual Review of Psychology, 75, 653-675. https://doi.org/10.1146/annurev-psych-030123-113559 Cha, Y.-J., Baek, S., Ahn, G., Lee, H., Lee, B., Shin, J.-E., & Jang, D. (2020). Compensating for the loss of human distinctiveness: The use of social creativity under human–machine comparisons. Computers in Human Behavior, 103, 80–90. https://doi.org/10.1016/j.chb.2019.08.027 Petty, R. E., Cacioppo, J. T., & Goldman, R. (1981). Personal Involvement as a Determinant of Argument-Based Persuasion. Journal of Personality and Social Psychology, 41(5), 847. https://doi.org/10.1037/0022-3514.41.5.847 Thompson, E. P., Roman, R. J., Moskowitz, G. B., Chaiken, S., & Bargh, J. A. (1994). Accuracy motivation attenuates covert priming: The systematic reprocessing of social information. Journal of Personality and Social Psychology, 66(3), 474. https://doi.org/10.1037/0022-3514.66.3.474 Tables Table1. Demographics of our Sample Demographics Satisfying–AI (n = 111) Satisfying–Human (n = 110) Satisfying–Expert (n = 116) Dissatisfying–AI (n = 118) Dissatisfying–Human (n = 104) Dissatisfying–Expert (n = 126) Neutral–AI (n = 106) Neutral–Human (n = 93) Neutral–Expert (n = 103) All (n = 987) Age 18-24 9.0% 23.6% 8.6% 7.6% 20.2% 13.5% 6.6% 25.8% 7.8% 13.4% 25-34 20.7% 24.5% 22.4% 22.0% 25.0% 18.3% 21.7% 24.7% 24.3% 22.5% 35-44 27.9% 28.2% 31.0% 24.6% 23.1% 29.4% 29.2% 17.2% 26.2% 26.5% 45-54 23.4% 15.5% 21.6% 27.1% 22.1% 24.6% 20.8% 21.5% 25.2% 22.5% 55-64 15.3% 8.2% 12.1% 16.9% 7.7% 12.7% 17.0% 9.7% 13.6% 12.7% 65-74 2.7% 0.0% 4.3% 0.8% 1.9% 0.8% 3.8% 1.1% 1.9% 1.9% 75 + 0.9% 0.0% 0.0% 0.8% 0.0% 0.8% 0.9% 0.0% 1.0% 0.5% Mean 41.9 35.0 41.0 42.2 37.0 40.5 42.8 36.4 41.2 39.9 Gender Male 61.3% 51.8% 67.2% 55.9% 60.6% 65.1% 63.2% 61.3% 66.0% 61.4% Female 38.7% 48.2% 32.8% 44.1% 39.4% 34.9% 36.8% 38.7% 34.0% 38.6% Others 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% Demographic characteristics of participants across the nine experimental conditions in a 3 (AI experience recall: satisfying vs. dissatisfying vs. neutral) × 3 (evaluation target: AI vs. humans vs. experts) between-subjects design (N = 987). Values represent percentages within each condition unless otherwise noted; mean age is reported in years. Table 2. Effects of AI Experience Recall on AI Continuance Intention and Willingness to Pay Measure Priming Condition Satisfying AI experience (n = 337, M ± SD) Dissatisfying AI experience (n = 348, M ± SD) Neutral AI experience (n = 302, M ± SD) F(2, 984) p η² Post-hoc (Bonferroni) Intention to Continue Using AI 5.87 ± 1.07 5.55 ± 1.35 5.78 ± 1.18 6.37 0.002 0.013 Dissatisfying < Satisfying Dissatisfying < Neutral Willingness to Pay for a Paid AI Plan 3.49 ± 1.54 3.14 ± 1.56 3.37 ± 1.51 4.69 0.009 0.009 Dissatisfying < Satisfying Effects of AI experience recall (satisfying vs. dissatisfying vs. neutral) on overall attitudes toward generative AI. Means (M) and standard deviations (SD) are reported for AI continuance intention and willingness to pay for a paid AI plan (7-point scales). One-way ANOVAs tested the main effect of priming condition across the full sample (N = 987), followed by Bonferroni-adjusted post hoc comparisons. Table 3. Effects of AI Experience Recall on Evaluations of AI, Humans, and Experts Target/Measure Priming Condition Satisfying AI experience (M ± SD) Dissatisfying AI experience (M ± SD) Neutral AI experience (M ± SD) F p η² Post-hoc (Bonferroni) AI F(2, 332) Trust 4.32 ± 1.22 3.90 ± 1.19 4.13 ± 1.32 3.28 0.039 0.019 Dissatisfying < Satisfying Perceived Competence 5.52 ± 0.86 4.96 ± 1.10 5.40 ± 0.99 10.11 < 0.001 0.057 Dissatisfying < Satisfying Dissatisfying < Neutral Willingness to Rely (AI-appropriate) 5.28 ± 1.04 5.11 ± 1.09 5.27 ± 1.15 0.85 0.430 0.005 – Willingness to Rely (Human-appropriate) 3.86 ± 1.36 3.50 ± 1.39 3.56 ± 1.38 2.26 0.106 0.013 – Humans F(2, 304) Trust 3.83 ± 1.23 3.72 ± 1.24 4.11 ± 1.34 2.47 0.086 0.016 – Perceived Competence 3.95 ± 1.14 3.87 ± 1.17 3.90 ± 1.17 0.11 0.897 0.001 – Willingness to Rely (AI-appropriate) 3.52 ± 1.17 3.81 ± 1.20 3.72 ± 1.23 1.64 0.195 0.011 – Willingness to Rely (Human-appropriate) 4.82 ± 1.22 4.85 ± 1.03 4.92 ± 1.03 0.24 0.791 0.002 – Experts F(2, 342) Trust 4.65 ± 1.19 4.49 ± 1.21 4.80 ± 1.11 1.94 0.145 0.011 – Perceived Competence 4.96 ± 1.08 4.97 ± 1.00 5.04 ± 1.08 0.2 0.820 0.001 – Willingness to Rely (AI-appropriate) 4.50 ± 1.18 4.44 ± 1.24 4.53 ± 0.98 0.18 0.839 0.001 – Willingness to Rely (Human-appropriate) 4.34 ± 1.09 4.21 ± 1.25 4.33 ± 0.91 0.51 0.603 0.003 – Effects of AI experience recall (satisfying vs. dissatisfying vs. neutral) on evaluations of three targets (AI, humans, and experts). Means (M) and standard deviations (SD) are reported for trust, perceived competence, and willingness to rely (7-point scales). One-way ANOVAs were conducted separately within each evaluation target to test the effect of priming condition. Willingness to rely is presented for aggregated AI-appropriate and human-appropriate task contexts. Bonferroni-adjusted post hoc comparisons are reported where significant. Table 4. Direct comparison of the satisfying–dissatisfying contrast across targets Panel A. Omnibus interaction test Measure Priming × Target df p Trust F = 1.03 (4, 978) 0.390 Perceived competence F = 2.97 (4, 978) 0.019 Panel B. Planned contrasts (target-standardized scores) Measure AI Sat − Dis Human Sat − Dis Expert Sat − Dis AI − Human AI − Expert Trust 0.34 (0.011) 0.09 (0.506) 0.14 (0.290) 0.25 (0.196) 0.20 (0.277) Perceived competence 0.55 (< 0.001) 0.06 (0.639) −0.02 (0.901) 0.49 (0.010) 0.57 (0.002) Follow-up analyses on target-standardized trust and perceived competence scores directly comparing the planned satisfying-versus-dissatisfying priming contrast across AI, human, and expert targets. Panel A reports Type III tests from models including priming condition, target, and their interaction. Panel B reports the satisfying-versus-dissatisfying contrast within each target and the corresponding difference-of-differences comparing AI with humans and experts.Scores were standardized within target before analysis; positive estimates indicate higher evaluations in the satisfying condition than in the dissatisfying condition. Raw means and standard deviations for each priming condition are reported in Table 3. Table 5. Bayesian t-tests for satisfying vs. dissatisfying comparisons in human and expert evaluations Target Outcome Mean diff. (Sat − Dis) BF01 NHST p Human Trust 0.12 5.38 0.494 Human Perceived competence 0.07 6.05 0.642 Human Rely (AI-appropriate) −0.29 1.50 0.075 Human Rely (Human-appropriate) −0.03 6.60 0.856 Expert Trust 0.16 4.29 0.303 Expert Perceived competence −0.02 7.05 0.901 Expert Rely (AI-appropriate) 0.05 6.75 0.743 Expert Rely (Human-appropriate) 0.13 5.05 0.396 Bayesian independent-samples t-tests comparing the satisfying and dissatisfying priming conditions for trust, perceived competence, and aggregate willingness-to-rely outcomes in the human and expert target conditions. Mean differences are reported as satisfying minus dissatisfying; positive values indicate higher evaluations in the satisfying condition. BF01 quantifies evidence for the null hypothesis of no difference. Bayes factors were computed using the default Cauchy prior ( ). BF01 values greater than 3 indicate at least moderate evidence for the null hypothesis. NHST p values are shown for reference. AI evaluations are not included because these Bayesian analyses were intended to quantify the absence of spillover to non-AI targets. Additional Declarations No competing interests reported. Supplementary Files AIvsHumanSupplementary.pdf Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-9267872","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":629993560,"identity":"fe541990-1eb1-453e-9823-7a1bf2040d3f","order_by":0,"name":"Kasumi Dan","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA4klEQVRIiWNgGAWjYBACCTBpYCMnz9588GEDmH+AKC1pxoY9x5INSdDCcDix4UaOmWQDMQ6TbD/+TOpGATNjY88Zs8oZFRZyDIxn8VsjzZNjJp1jwMbMzt5WdnPDGQljBoZzCXi1yDHksAG18LAx9hzedvNhm0RiA8MZA/xa+J8/A2qR4GG4kWBW+PAfEVqkJRJADjOQYLiRYsa4sYEILZIz3hhb5xgkGIACWXLGMQljNkJ+kTif/vB2zp//9fOBUfmxp6ZOjl+CQIhhAjaJMyTqYGDg7yFZyygYBaNgFAxvAAAHDEgtoxf+qwAAAABJRU5ErkJggg==","orcid":"","institution":"Keio University","correspondingAuthor":true,"prefix":"","firstName":"Kasumi","middleName":"","lastName":"Dan","suffix":""},{"id":629993562,"identity":"939118b2-f8d4-4c3a-b32b-cf11bd518857","order_by":1,"name":"Takahiro Hoshino","email":"","orcid":"","institution":"Keio University","correspondingAuthor":false,"prefix":"","firstName":"Takahiro","middleName":"","lastName":"Hoshino","suffix":""}],"badges":[],"createdAt":"2026-03-30 13:53:34","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-9267872/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9267872/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":108491364,"identity":"7e7e1e7d-f88f-4652-b0f5-1970a39bf7de","added_by":"auto","created_at":"2026-05-05 09:53:32","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":455730,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9267872/v1/76bf0944-1bb2-4132-8f90-93a52deb5b6d.pdf"},{"id":108077018,"identity":"718caf0a-389b-4025-b883-3976301dea31","added_by":"auto","created_at":"2026-04-29 07:05:57","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":267861,"visible":true,"origin":"","legend":"","description":"","filename":"AIvsHumanSupplementary.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9267872/v1/2261090c31934d0524a51c5a.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Recalled AI Experiences Shift Evaluations of AI but Not of Humans or Experts","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eRecent advances in artificial intelligence (AI), particularly conversational generative AI, have rapidly expanded its user base [1, 2]. As AI has become integrated into society, evidence shows that it improves work quality and efficiency [3, 4] and enhances well-being and quality of life [5, 6, 7]. Yet its development has been accompanied by anxiety, perceived threat, and resistance, and concerns about its negative societal consequences persist [8, 9].\u003c/p\u003e\n\u003cp\u003eAI-related anxiety and perceived threat have long been discussed, from economic concerns about job displacement [10] to broader fears that AI will dominate or destabilize humanity [11, 12]. A prominent concern is that AI may diminish the value, dignity, or necessity of humans, either collectively or individually [13].\u003c/p\u003e\n\u003cp\u003eRecent research suggests that anxiety about AI adoption extends beyond economic concerns. A more persistent concern is that individual abilities or social standing may be devalued and AI judgments trusted more than human judgments, undermining human credibility and necessity as decision-making agents. These concerns have been documented as threats to identity, self-worth, and perceived human necessity [14, 15, 16]. Specifically, exposure to AI adoption elicits occupational identity threats, including fears that AI will override individual judgment or diminish others\u0026rsquo; respect for them. Likewise, automation threats reduce self-esteem and increase psychological instability [17, 18]. In extreme cases, AI-related developments evoke annihilation anxiety\u0026mdash;the fear that humans may no longer be necessary [15].\u003c/p\u003e\n\u003cp\u003eHowever, these concerns mainly reflect anticipated declines in one\u0026rsquo;s own value or credibility and thus remain at the level of perceived anxiety. What remains largely unexamined is whether, as AI advances, people have begun to evaluate others more negatively or to see humans in general as less necessary. In other words, it remains unclear whether people make zero-sum judgments in which increased trust in and reliance on AI are accompanied by a corresponding decrease in human evaluation.\u003c/p\u003e\n\u003cp\u003eExperimental findings suggest this possibility. Prior research has shown that people occasionally follow algorithmic advice over human advice, regardless of objective accuracy or quality, implying that human judgments, including the judgment of an individual and those of others, may be considered inferior to AI judgments [19, 20, 21, 22]. However, because these studies rely on simultaneous comparisons, they cannot distinguish whether preferring AI entails a devaluation of humans or whether AI is simply evaluated more positively while evaluations of humans remain unchanged.\u003c/p\u003e\n\u003cp\u003eTherefore, the present study considered changes in evaluations of AI and evaluations of humans as conceptually independent. It examined whether increased trust in and reliance on AI are accompanied by a tradeoff in which evaluations of humans decline.\u003c/p\u003e\n\u003cp\u003eIn this study, participants recalled positive, negative, or neutral AI experiences and then evaluated one target\u0026mdash;AI, humans, or experts\u0026mdash;on trust, perceived competence, and willingness to rely across multiple task contexts, testing whether effects differed between AI-appropriate and human-appropriate tasks. Because trust and perceived competence are core evaluation dimensions [23, 24], and willingness to rely indicates behavioral translation [25, 26], these were the primary measures.\u003c/p\u003e\n\u003cp\u003eAlthough the primary goal was to examine potential tradeoffs between AI and human evaluations, we also validated whether recalling satisfying or dissatisfying AI experiences directly shifts overall attitudes toward AI, including continuance intentions and willingness to pay. Establishing such baseline attitude shifts enabled us to assess whether changes in AI evaluations spillover to other targets.\u0026nbsp;\u003c/p\u003e"},{"header":"2. Background and Hypotheses","content":"\u003cp\u003e2.1. Effects of AI Perceptions on Evaluations of Humans in General\u003c/p\u003e\n\u003cp\u003eAI is increasingly viewed as a socially comparable outgroup with the potential to impact how humans evaluate themselves and others, rather than only as a tool [15]. Empirical research suggests that AI intelligence functions as a social reference point for human self-understanding. Santoro and Monin [27] show that when people learn that AI is acquiring abilities traditionally considered aspects of “human nature,” such as logical reasoning or communication, they update what they view as essential to being human by emphasizing other traits believed to be distinctively human (e.g., having a personality or feeling love). Similarly, exposure to AI-generated work leads people to evaluate their own creativity and that of humans in general more positively, as creativity is perceived as a skill that AI lacks [28, 29]. In contrast, Kim and McGill [30] found that exposure to autonomous agents with high socio-emotional capabilities leads people to perceive AI as more humanlike while simultaneously judging real humans as lower in humanness. Similarly, Osborne and Bailey [31] showed that reading highly empathetic advice generated by large language models reduces evaluations of the authenticity of the advice of an individual, which is defined as being grounded in the internal values and human qualities of that individual.\u003c/p\u003e\n\u003cp\u003eThese studies suggest that people evaluate humans in relation to AI, and that this comparison is bidirectional. When AI highlights what it lacks, perceived human value can rise [27, 28, 29]. When AI’s humanlike qualities and competence are salient, however, perceptions of human authenticity or humanness can decline [30, 31]. Thus, emphasizing AI’s limitations tends to elevate human value, whereas emphasizing its humanlike capabilities tends to diminish it.\u003c/p\u003e\n\u003cp\u003eThis pattern concerns not only what humans and AI can both do, but also what remains distinctively human. As long as some capacities are still seen as uniquely human, human value may not be immediately threatened. But as AI acquires abilities that increasingly overlap with human capacities, the domain of the “distinctively human,” and thus a basis of human value, is likely to contract.\u003c/p\u003e\n\u003cp\u003eMalone et al. [32] describe such overlap as a structure that can intensify zero-sum trust dynamics, in which greater trust in AI may come at the expense of trust in humans. Related work suggests that as AI replaces human roles, people may lose opportunities to display their abilities and value, undermining human dignity and perceived necessity [33, 34, 35]. Overall, these findings support a zero-sum view in which gains in AI evaluation may trade off against evaluations of humans.\u003c/p\u003e\n\u003cp\u003eTherefore, when people compare AI with humans, satisfaction with AI may lead them to perceive a reduction in uniquely human capabilities, thereby lowering their evaluations of humans. In contrast, dissatisfaction with AI may heighten the salience of uniquely human capabilities and increase their evaluations of humans. Based on this reasoning, we propose the following hypothesis.\u003c/p\u003e\n\u003cp\u003e \u003cstrong\u003eH1a.\u003c/strong\u003e Recalling a satisfying experience with AI will increase trust, perceived competence, and the willingness to rely on AI, while decreasing trust, perceived competence, and the willingness to rely on humans.\u003c/p\u003e\n\u003cp\u003e \u003cstrong\u003eH1b.\u0026nbsp;\u003c/strong\u003eRecalling a dissatisfying experience with AI will decrease trust, perceived competence, and the willingness to rely on AI, while increasing trust, perceived competence, and the willingness to rely on humans.\u003c/p\u003e\n\u003cp\u003e2.2. Effects of AI Perceptions on Evaluations of Human Experts\u003c/p\u003e\n\u003cp\u003eThe previous section argued that evaluations of AI and humans may form a zero-sum relationship. Yet people do not view all humans as equally replaceable by AI. Research has long distinguished experts from laypeople [36, 37, 38], and people trust these groups differently: expert opinions are seen as more accurate than one’s own, and expert advice is followed more than nonexpert advice [39, 40, 41]. Experts thus occupy socially recognized positions distinct from the general public [42], implying that human evaluations are stratified rather than uniform.\u003c/p\u003e\n\u003cp\u003eA similar stratified view appears in research on technological change. Computerization and automation disproportionately threaten low-skill, low-wage work, whereas jobs requiring advanced skills or judgment are less affected [43]. Their negative effects on employment and wages fall mainly on routine manual and nonexpert work [44]. More broadly, technology replaces routine, repetitive labor but often complements nonroutine cognitive work, increasing its relative value [45, 46]. Technological progress therefore differentiates between those displaced and those—often experts—whose value is maintained or enhanced through integration with technology.\u003c/p\u003e\n\u003cp\u003eBecause AI is also a technology, this logic should extend to AI. Discussions of AI’s social impact likewise adopt a stratified perspective [47], describing AI as extending automation while shifting human work toward either low-skilled nonroutine tasks or high-skilled work requiring professional judgment [48]. Even when AI enters nonroutine cognitive domains such as diagnosis and programming, it need not displace professionals; it can complement their work and raise the value of expertise [49]. Consistent with this view, people still prefer human experts to AI in highly expert domains [50, 51, 52]. AI development may therefore stratify evaluation targets by their relation to AI rather than lowering evaluations of all humans alike. Experts, in particular, may preserve or even increase their value as AI advances.\u003c/p\u003e\n\u003cp\u003eRecent work further suggests that the perceived competence of highly agentic AI is associated with higher competence evaluations of its designers [53]. If experts are seen as those who design, manage, or skillfully use AI, then greater trust in and reliance on AI may spill over into more favorable evaluations of experts. Related research supports this possibility: people project impressions of AI from their evaluations of its users, and trust in technological systems often extends to trust in designers and operators [54, 55]. AI is thus not perceived as an isolated entity but as implicitly linked to the people behind it. From this perspective, experts can be understood as an AI-adjacent group. Satisfaction with AI should therefore increase evaluations of experts, whereas dissatisfaction should decrease them. Based on this reasoning, we propose the following hypothesis.\u003c/p\u003e\n\u003cp\u003e \u003cstrong\u003eH2a.\u003c/strong\u003e Recalling a satisfying experience with AI will increase trust, perceived competence, and willingness to rely on experts.\u003c/p\u003e\n\u003cp\u003e \u003cstrong\u003eH2b.\u003c/strong\u003e Recalling a dissatisfying experience with AI will decrease trust, perceived competence, and willingness to rely on experts.\u003c/p\u003e\n\u003cp\u003e2.3. Task-Type Moderation\u003c/p\u003e\n\u003cp\u003eAttitudes toward AI vary systematically with the task characteristics. AI is generally trusted more in tasks with clear criteria, high objectivity, or routine structures, and less in subjective tasks requiring creativity, value judgment, or personalization [19, 56, 57]. In domains involving moral judgment or empathy, resistance to machine decision-making remains strong, even when the output quality is high [58, 59, 60]. Overall, people appear to distinguish between “AI-appropriate tasks” and “tasks that humans should perform,” and base trust and usage intentions on this distinction.\u003c/p\u003e\n\u003cp\u003eThe key distinction lies in the asymmetry of choice: AI-appropriate tasks are those that both AI and humans can perform, whereas human-appropriate tasks are those that only humans should perform.\u003c/p\u003e\n\u003cp\u003eIn AI-appropriate tasks, using AI is relatively acceptable, and the choice between delegating tasks to AI or humans is more open. Consequently, evaluations are likely to depend less on normative beliefs about who should perform the task and more on performance expectations. Therefore, information regarding prior satisfaction or dissatisfaction with AI should carry a greater weight, leading to larger shifts in attitude. Consequently, human evaluations in general and expert evaluations, as described in H1 and H2, are also more likely to change in these contexts.\u003c/p\u003e\n\u003cp\u003eBy contrast, in human-appropriate tasks, the firm belief that humans should perform such tasks limits the perceived legitimacy of delegating them to AI. Therefore, even when AI satisfaction or dissatisfaction is recalled, entrenched preferences for humans and low expectations of AI are likely to constrain attitude changes.\u003c/p\u003e\n\u003cp\u003eOverall, the effects proposed in H1 and H2 are expected to be moderated by task context, with stronger effects for AI-appropriate tasks and weaker effects for human-appropriate tasks.\u003c/p\u003e\n\u003cp\u003eH3. The magnitude of these effects will vary by task context: Stronger in AI-appropriate tasks (e.g., information or technical tasks) and weaker in human-appropriate tasks (e.g., emotional or creative contexts).\u003c/p\u003e\n\u003cp\u003eIn addition to the three hypotheses, we examined whether recalling satisfying or dissatisfying AI experiences influences broader attitudes toward AI, specifically, continuance intentions and willingness to pay for a paid plan. These measures capture more reliable and behaviorally proximal attitudes, which serve as prerequisites for the aforementioned evaluative spillover effects. Although this analysis is independent of the central hypothesis, it is informative for understanding the acceptance of generative AI.\u003c/p\u003e"},{"header":"3. Materials and Methods","content":"\u003cp\u003e3.1. Research Design\u003c/p\u003e\n\u003cp\u003eThe study was approved by the Institutional Review Board of the Institute for Economic Studies, Keio University (Receipt No. 25005), and all procedures were performed in accordance with the Declaration of Helsinki and relevant institutional guidelines and regulations. This work was supported by the\u0026nbsp;Japan Society for the Promotion of Science\u0026nbsp;under Grant\u0026nbsp;JP22H03552.\u0026nbsp;It was registered prior to data collection. We employed a 3 (AI priming: satisfying vs. dissatisfying vs. neutral) × 3 (evaluation target: AI vs. humans vs. experts) fully between-subjects experimental design with participants randomly assigned to one of the nine conditions.\u003c/p\u003e\n\u003cp\u003eThe independent variables were the AI experience recall (priming condition) and the evaluation target. The dependent variables were general trust, perceived competence toward the target, and willingness to rely on the target across multiple task contexts. Trust and perceived competence were measured as general evaluations independent of the context, whereas willingness to rely was assessed separately for each task context.\u003c/p\u003e\n\u003cp\u003eTask contexts were classified as AI-appropriate or human-appropriate based on a prior pilot study, which enabled us to test whether the effects of AI experience recall were moderated by task type. In addition, all participants reported continuance intentions toward generative AI and willingness to pay for a paid AI plan, which served as supplementary measures of overall AI attitudes and as a validity check for the priming manipulation.\u003c/p\u003e\n\u003cp\u003e3.2. Participants\u003c/p\u003e\n\u003cp\u003eA preregistered power analysis using G*Power 3.1 (f = 0.20, α = 0.05, power = 0.80; one-way ANOVA) indicated a required sample size of approximately 83 participants per condition. As the effects of the three priming conditions (satisfying, dissatisfying, and neutral) were analyzed separately for each evaluation target (AI, humans, and experts), the target sample size was set at 744. To account for potentially smaller effect sizes, we planned to recruit participants to obtain approximately 900–1,170 valid responses.\u003c/p\u003e\n\u003cp\u003eParticipants were recruited from a large Japanese online panel. Eligible participants, Japanese-speaking adults (18+ years old) with prior experience using generative AI, were randomly assigned to one of the nine conditions. They received compensation in accordance with the standards set by the panel provider. After providing informed consent, 1,606 participants completed the survey.\u003c/p\u003e\n\u003cp\u003eParticipants who failed the attention check were excluded from the analysis. The attention check item was “When I want to kill time, I would like to talk to (AI/humans/experts). Note: Regardless of your actual opinion, please select ‘1’ for this item.” After exclusion, the final analytical sample comprised 987 participants.\u003c/p\u003e\n\u003cp\u003eTo assess demographic balance, gender distribution, and age, these variables were compared across priming conditions (satisfying, dissatisfying, and neutral) within each evaluation target condition (AI, humans, and experts). No significant differences were observed, neither in any target condition (gender: χ\u003csup\u003e²\u003c/sup\u003e(2) = 0.13–2.40, ps = 0.301–0.939; age: F(2, 304–342) = 0.09–0.71, ps = 0.493–0.910), nor in the overall sample (gender: χ\u003csup\u003e²\u003c/sup\u003e(2) = 0.88, p = 0.643; age: F(2, 984) = 0.43, p = 0.649). These results indicate minimal demographic imbalance across conditions. The sample size and demographic characteristics of each condition are presented in Table 1.\u003c/p\u003e\n\u003cp\u003e3.3. Procedure\u003c/p\u003e\n\u003cp\u003eAfter reading an explanation about data usage, confidentiality, and privacy protection, participants provided written informed consent electronically before participating in the study. They were randomly assigned to one of nine conditions (3 AI priming × 3 evaluation targets). For the priming manipulation, participants first recalled their past experiences with generative AI in an open-ended format. In the satisfying condition, they described three positive aspects of their AI use; in the dissatisfying condition, three negative aspects; and in the neutral condition, three general uses of generative AI. At this stage, AI was defined as conversational generative AI based on large language models, with examples provided (e.g., ChatGPT), while nontarget systems (e.g., voice assistants such as Siri) were explicitly excluded to ensure a shared understanding across participants. The full description of the priming materials is provided in Supplementary Table S1.\u003c/p\u003e\n\u003cp\u003eAfter the priming task, participants evaluated a single target (AI, humans, or experts) for general trust and perceived competence. Subsequently, they reported their willingness to rely on the target across multiple task scenarios, including emotional support and writing tasks; the attention check item was embedded within these measures. In the expert condition only, participants were also presented with a definition of experts as individuals with advanced knowledge and experience in a specific domain who provide advice, judgment, or support in society [61], with examples such as scientists, researchers, physicians, lawyers, and consultants.\u003c/p\u003e\n\u003cp\u003eFinally, all participants completed measures of continuance intention toward generative AI and willingness to pay for a paid AI plan.\u003c/p\u003e\n\u003cp\u003e3.4. Measures\u003c/p\u003e\n\u003cp\u003eThe primary dependent variables were trust, perceived competence, and willingness to rely on an evaluation target (AI, humans, or experts). In addition, overall attitudes toward generative AI were measured using continuance intention and willingness to pay for a paid AI plan. All items were rated on 7-point Likert scales, and mean scores were computed for each construct. The full wording of the item is presented in Supplementary Table S2.\u003c/p\u003e\n\u003cp\u003e3.4.1. Trust\u003c/p\u003e\n\u003cp\u003eGeneral trust in the evaluation target was measured using a shortened version of the disposition to trust scale [62]. Although initially developed to assess trust in humans, the scale has been adapted for machines by replacing the referent while preserving the item structure (Disposition to Trust Machines [63]), rendering it suitable for the parallel assessment of AI and humans. In the present study, the same item structure was used across conditions, with the target replaced by “AI,” “humans,” or “experts.” The responses ranged from 1 (“strongly disagree”) to 7 (“strongly agree”), with higher scores indicating greater trust.\u003c/p\u003e\n\u003cp\u003e3.4.2. Perceived Competence\u003c/p\u003e\n\u003cp\u003ePerceived competence was measured using items from the agency–competence (AC) sub-dimension of the agency–communion framework [64]. This construct captures the extent to which the target is perceived as competent. The item wording was maintained constant across conditions, with the subject replaced by the relevant target. All items were rated on a 7-point Likert scale ranging from 1 (“not very competent/efficient/intelligent”) to 7 (“very competent/efficient/intelligent”), with higher scores indicating greater perceived competence.\u003c/p\u003e\n\u003cp\u003e3.4.3. Willingness to Rely\u003c/p\u003e\n\u003cp\u003eWillingness to rely was measured across eight task contexts (four AI-appropriate and four human-appropriate) classified in a prior pilot study. For each context, participants indicated their agreement with statements about the evaluation target using a 7-point Likert scale (1 = strongly disagree; 7 = strongly agree). Analyses were conducted at two levels: first, at the level of each specific task context, and second, at the task-type level, by averaging the four contexts classified as AI-appropriate and the four classified as human-appropriate. Higher scores indicated a stronger intention to rely on the target in a given context.\u003c/p\u003e\n\u003cp\u003eA pilot study was conducted to validate the classification of the eight task contexts. Participants (N = 60; 31.7% women; mean age = 46.2 years) rated each context on a 7-point scale ranging from 1 (“humans”) to 7 (“AI”) in response to the question of whether humans or AI should perform the task. One-sample t-tests comparing the mean of each item to the scale midpoint (4) showed that contexts designated as AI-appropriate were rated significantly toward AI. In contrast, those designated as human-appropriate were rated significantly toward humans, consistent with the prior classification of researchers. Thus, the assignment of each context to AI-appropriate or human-appropriate tasks was validated by the judgments of participants. Supplementary Table S3 presents the full item wordings and statistical results.\u003c/p\u003e\n\u003cp\u003e3.4.4. AI Continuance Intention\u003c/p\u003e\n\u003cp\u003eAs a standard measure across all conditions, overall attitudes toward AI were assessed using a continuance-intention scale adapted from prior studies on the technology acceptance model [65, 66]. Items assessed the intention to continue using AI over the approaching months or years and were rated on a 7-point Likert scale (1 = strongly disagree; 7 = strongly agree). The original scale targeted general technologies; in this study, the items were adapted to refer specifically to generative AI tools. Higher scores indicated stronger intentions to continue using AI.\u003c/p\u003e\n\u003cp\u003e3.4.5. Willingness to Pay for a Paid AI Plan\u003c/p\u003e\n\u003cp\u003eAs a standard measure across all conditions, the willingness to pay for a paid generative AI plan was measured using items adapted from prior research on subscription intentions for online services [67], with references replaced with target generative AI tools. The items assessed willingness to pay for a paid plan and the perceived value of the plan, using a 7-point Likert scale (1 = strongly disagree; 7 = strongly agree). Higher scores indicated a stronger willingness to subscribe to a paid AI plan.\u003c/p\u003e\n\u003cp\u003eAll scales that were not initially developed for this study were translated into Japanese using a standard translation–back-translation procedure. Two bilingual researchers independently translated the items and reconciled discrepancies, and a third bilingual researcher conducted back-translation. This process was repeated until semantic equivalence with the original items was achieved.\u003c/p\u003e\n\u003cp\u003e3.5. \u0026nbsp;Statistical analysis\u003c/p\u003e\n\u003cp\u003eTo examine whether the AI priming manipulation affected overall attitudes toward AI, one-way ANOVAs were conducted on AI continuance intention and willingness to pay for a paid AI plan, both measured across all conditions, with the priming condition (satisfying vs. dissatisfying vs. neutral) as the independent variable.\u003c/p\u003e\n\u003cp\u003eAlthough the measures share a standard item structure, they differ in referents, and measurement invariance across targets cannot be assumed. Therefore, the preregistered primary analyses were performed separately for each evaluation target (AI, humans, and experts) using one-way ANOVAs to test the effect of the priming condition (satisfying, dissatisfying, and neutral) within each target. Bonferroni-adjusted post hoc comparisons were conducted following significant omnibus effects.\u003c/p\u003e\n\u003cp\u003eAfter completion of the preregistered analyses, four additional exploratory analyses were conducted to clarify the interpretation and robustness of the results. First, because separate target-specific tests do not by themselves establish whether the magnitude of the priming effect differs across targets, we conducted exploratory follow-up analyses on trust and perceived competence. In these follow-up analyses, scores were standardized within target prior to estimation, and models included priming condition, target, and their interaction. Planned contrasts focused on the satisfying-versus-dissatisfying comparison and the corresponding AI-versus-human and AI-versus-expert difference-of-differences.\u003c/p\u003e\n\u003cp\u003eSecond, to quantify evidence for and against satisfying-versus-dissatisfying differences in human and expert evaluations, we conducted exploratory Bayesian independent-samples t-tests. These analyses were conducted for trust, perceived competence, and the aggregate willingness-to-rely indices. We report BF01 using the default Cauchy prior (r = 0.707).\u003c/p\u003e\n\u003cp\u003eThird, to test the preregistered task-type moderation hypothesis more directly, willingness-to-rely responses were also reanalyzed in exploratory item-level mixed-effects models within each target condition. These models included fixed effects of priming condition, task type, and their interaction, with random intercepts for participant and scenario.\u003c/p\u003e\n\u003cp\u003eFinally, as an exploratory robustness check, the principal ANOVAs were repeated in the full sample prior to attention-check exclusion.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e3.6. Declaration of generative AI and AI-assisted technologies in the manuscript preparation process\u003c/p\u003e\n\u003cp\u003eDuring the preparation of this manuscript, the authors made limited use of generative AI tools (ChatGPT, including Deep Research functionality) to support exploratory literature searches and to assist with English-language editing and phrasing. All relevant literature was independently located, read, and evaluated by the authors. The study design, analyses, interpretation of results, and final wording of the manuscript were conducted entirely by the authors, who take full responsibility for the integrity and content of the published work.\u003c/p\u003e"},{"header":"4. Results","content":"\u003cp\u003e4.1. Baseline Effects of AI Experience Recall\u003c/p\u003e\n\u003cp\u003eBaseline effects of the priming manipulation on overall attitudes toward AI are presented in Table 2. AI continuance intention showed a significant main effect of priming, F(2, 984) = 6.37, p = 0.002, η² = 0.013. The mean scores were in the following order: satisfying \u0026gt; neutral \u0026gt; dissatisfying. Bonferroni-adjusted comparisons indicated significant differences between the satisfying and dissatisfying conditions and between the neutral and dissatisfying conditions, but not between the satisfying and neutral conditions.\u003c/p\u003e\n\u003cp\u003eA similar pattern emerged for willingness to pay for a paid AI plan, with a significant main effect of priming, F(2, 984) = 4.69, p = 0.009, and η² = 0.009. The mean scores were in the same order: satisfying \u0026gt; neutral \u0026gt; dissatisfying; however, post hoc tests revealed a significant difference only between the satisfying and dissatisfying conditions.\u003c/p\u003e\n\u003cp\u003eOverall, these results indicate that recalling satisfying versus dissatisfying AI experiences influences AI-related behavioral intentions, with negative recall exerting a powerful downward effect. Building on these baseline attitude shifts, the following section examines whether changes in evaluations of AI spillover to evaluations of other targets.\u003c/p\u003e\n\u003cp\u003e4.2. Effects of AI Experience Recall on Evaluations of Humans and Experts\u003c/p\u003e\n\u003cp\u003eThe primary preregistered target-specific analyses are presented in Table 3.\u003c/p\u003e\n\u003cp\u003e4.2.1. AI Condition\u003c/p\u003e\n\u003cp\u003eRegarding trust in AI, a significant priming effect was observed, F(2, 332) = 3.28, p = 0.039, and η² = 0.019. The mean scores were in the following order: satisfying \u0026gt; neutral \u0026gt; dissatisfying; however, post hoc tests showed a significant difference only between the satisfying and dissatisfying conditions, with no significant difference between the satisfying and neutral conditions.\u003c/p\u003e\n\u003cp\u003ePerceived competence of AI also showed a significant priming effect, F(2, 332) = 10.11, p \u0026lt; 0.001, and η² = 0.057. Post hoc comparisons indicated that the dissatisfied condition was rated significantly lower than both the satisfied and neutral conditions, whereas the satisfied and neutral conditions did not differ significantly.\u003c/p\u003e\n\u003cp\u003eFor the more behaviorally proximal measure of willingness to rely, priming effects were largely non-significant across individual task contexts. An exception was observed in the “complex discussion” context, where a significant effect emerged, F(2, 332) = 3.48, p = 0.032, andη² = 0.021, with willingness to rely being lower in the dissatisfying condition (M = 4.00 and SD = 1.69) than in the satisfying condition (M = 4.57 and SD = 1.57). By contrast, no significant priming effects were found for the aggregated indices of AI-appropriate or human-appropriate tasks.\u003c/p\u003e\n\u003cp\u003eOverall, these results suggest that recalling satisfying versus dissatisfying AI experiences was sufficient to shift attitudinal evaluations of AI, particularly trust and perceived competence. In contrast, the priming effects on the willingness to rely were minimal.\u003c/p\u003e\n\u003cp\u003e4.2.2. Human and Expert Conditions\u003c/p\u003e\n\u003cp\u003eNo significant priming effects were observed in either the human or expert condition for trust, perceived competence, or willingness to rely in any task context (all ps \u0026gt; 0.05). Therefore, recalling satisfying or dissatisfying AI experiences did not significantly change evaluations of humans in general or of experts.\u003c/p\u003e\n\u003cp\u003e4.3. \u0026nbsp;Exploratory follow-up analyses clarifying target specificity and null spillover\u003c/p\u003e\n\u003cp\u003eTo test whether the priming contrast was larger for AI than for humans or experts, we conducted exploratory general linear models (Table 4). For trust, the priming-by-target interaction was not significant, p = 0.390. The satisfying-versus-dissatisfying contrast was significant for AI (b = 0.34, p = 0.011) but not for humans (b = 0.09, p = 0.506) or experts (b = 0.14, p = 0.290), and the AI-versus-human and AI-versus-expert difference-of-differences were not significant (ps = 0.196 and 0.277). For perceived competence, the interaction was significant, p = 0.019. The same contrast was significant for AI (b = 0.55, p \u0026lt; 0.001) but not for humans (b = 0.06, p = 0.639) or experts (b = -0.02, p = 0.901), and was larger for AI than for humans (b = 0.49, p = 0.010) and experts (b = 0.57, p = 0.002).\u003c/p\u003e\n\u003cp\u003eWe conducted exploratory Bayesian t-tests for human and expert evaluations because non-significant spillover effects alone do not support the null (Table 5). Bayes factors provided moderate evidence for the null for humans (trust: BF01 = 5.38; competence: BF01 = 6.05) and experts (trust: BF01 = 4.29; competence: BF01 = 7.05). For willingness-to-rely outcomes, most comparisons favored the null, although evidence was weak for willingness to rely on humans in AI-appropriate tasks (BF01 = 1.50). Overall, these analyses indicated that the priming effect was target-specific for perceived competence and did not spill over to human or expert evaluations.\u003c/p\u003e\n\u003cp\u003e4.4. \u0026nbsp;Exploratory follow-up analyses of task-type moderation and robustness\u003c/p\u003e\n\u003cp\u003eTo test the preregistered task-type moderation hypothesis, we conducted exploratory item-level linear mixed-effects models for each target. Contrary to H3, the priming-by-task-type interaction was significant only for AI (p = 0.028), reflecting a larger satisfying-versus-dissatisfying contrast in human-appropriate tasks (b = 0.37, p = 0.021) than in AI-appropriate tasks (b = 0.17, p = 0.287). For humans, the interaction was not significant (p = 0.060), although the satisfying-versus-dissatisfying contrast in AI-appropriate tasks trended negative (b = -0.29, p = 0.055). For experts, neither the interaction nor the within-task-type contrasts was significant (ps ≥ 0.369). Thus, the preregistered moderation hypothesis was not supported.\u003c/p\u003e\n\u003cp\u003eWe repeated the principal ANOVAs in the full sample prior to attention-check exclusion. Reanalyses of completers (N = 1,606) yielded the same pattern: priming effects on AI trust, F(2, 546) = 3.10, p = 0.046, and perceived competence, F(2, 546) = 9.31, p \u0026lt; 0.001, remained significant, whereas all human outcomes remained non-significant (all ps \u0026gt; 0.34). Although an omnibus priming effect emerged for expert trust, F(2, 528) = 3.52, p = 0.030, this was driven by the neutral-versus-dissatisfying comparison (p = 0.025) rather than the satisfying-versus-dissatisfying contrast (p = 0.611). Accordingly, the main conclusions were not dependent on attention-check exclusion.\u003c/p\u003e\n\u003cp\u003e4.5. Summary of Hypothesis Tests\u003c/p\u003e\n\u003cp\u003eAlthough evaluations of AI were affected by priming manipulation, evaluations of humans in general were not, providing no support for H1, which predicts a tradeoff between evaluations of AI and humans. Similarly, because AI experience recall did not significantly affect evaluations of experts, H2, which predicts a corresponding spillover to experts, was not supported. Exploratory direct-comparison analyses showed that the target specificity of the priming effect was clearest for perceived competence, and exploratory Bayesian analyses provided moderate support for the absence of corresponding effects on human and expert trust and competence. Finally, exploratory item-level follow-up analyses did not support the preregistered task context moderation predicted in H3.\u003c/p\u003e"},{"header":"5. Discussion","content":"\u003cp\u003eThis study examined concerns that the rise of generative AI may undermine human values or necessity [13], focusing on whether such concerns are reflected in actual evaluative tendencies. To avoid considering humans as a homogeneous category, we distinguished between humans in general and experts who are theoretically less susceptible to replacement and more closely associated with the design and use of AI. Using an experiential priming manipulation that induced recall of satisfying or dissatisfying AI experiences, we assessed evaluations of AI, humans in general, and experts to test whether evaluations of AI and humans operate in a zero-sum manner and whether evaluations of AI spillover to experts as an AI-adjacent group.\u003c/p\u003e\n\u003cp\u003e5.1. Interpreting Changes in Evaluations of AI\u003c/p\u003e\n\u003cp\u003eThe results indicate that recalling positive or negative experiences with generative AI can shift general attitudes toward AI, including continuance intentions, willingness to subscribe to paid plans, and evaluations of trust and competence. Notably, this effect was more pronounced as a decline following a dissatisfying recall than as an increase following a satisfying recall. In other words, recalling satisfying experiences did not lead to significantly more positive AI evaluations than in the neutral condition, whereas recalling dissatisfying experiences was associated with significantly more negative evaluations.\u003c/p\u003e\n\u003cp\u003eAlthough this asymmetry is not explicitly hypothesized, it is consistent with prior research on algorithm aversion. Previous studies have suggested that people are sensitive to algorithmic errors and may discount algorithmic judgments after observing even a single failure, occasionally favoring less accurate human judgments instead [26]. From this perspective, the more substantial negative shift observed in the dissatisfying condition in addition to the limited positive change in the satisfying condition, may reflect an asymmetry in how people update their attitudes toward AI based on negative versus positive experiences.\u003c/p\u003e\n\u003cp\u003e5.2. Implications of the Absence of Spillover Effects\u003c/p\u003e\n\u003cp\u003eAs mentioned previously, the priming manipulation influenced AI evaluation, particularly trust and perceived competence. However, no significant priming effects were observed for evaluations of humans in general or experts; thus, the hypotheses predicting spillover from AI evaluations to human or expert evaluations were not supported. Recalling satisfaction or dissatisfaction with AI altered attitudes toward it but did not produce corresponding changes in human or expert evaluations. Exploratory follow-up analyses sharpened this interpretation. Direct-comparison analyses showed that the target specificity of the priming effect was clearest for perceived competence, for which the satisfying-versus-dissatisfying contrast was significantly larger for AI than for humans or experts. Exploratory Bayesian analyses further provided moderate support for the absence of corresponding effects on human and expert trust and perceived competence. This pattern was also robust in reanalyses of the full sample prior to attention-check exclusion.\u003c/p\u003e\n\u003cp\u003eThis pattern suggests that evaluations of AI and of humans or experts may not be linked in a zero-sum or tightly coupled manner, but may instead be psychologically independent. Within the scope of the present findings, existential concerns that AI advancement will directly undermine human trustworthiness, competence, and necessity do not appear to be reflected in people’s evaluative judgments. Similarly, the results do not support a simple linkage in which more favorable evaluations of AI translate into higher evaluations of experts associated with AI, suggesting that AI and experts may be perceived as distinct evaluative targets.\u003c/p\u003e\n\u003cp\u003eSeveral possible explanations exist for the absence of spillovers. One possibility is that people still prefer many tasks to be performed by humans [68, 69], leading them to maintain beliefs in uniquely human roles and values. Another is that, given warnings that anthropomorphizing AI, a mere machine, and comparing it with humans is inappropriate and harmful [70, 71], people may not regard AI as a social entity or may actively avoid doing so. Conversely, because people consider AI an autonomous agent [72], evaluations of experts as its designers or managers may not covary with evaluations of AI. Moreover, prior research suggests that exposure to AI development can elicit defensive reactions that uniquely reinforce beliefs about human values [73, 27]. Such processes may not have been detectable through self-reported measures, suggesting that zero-sum comparisons may operate at a more implicit level. Overall, these considerations suggest that the mechanisms underlying the present findings are likely complex, and future research is needed to clarify which factors play the most substantial role in shaping human and expert evaluations in the context of AI advancement.\u003c/p\u003e\n\u003cp\u003e5.3. Interpreting Findings on Willingness to Rely\u003c/p\u003e\n\u003cp\u003eAnother noteworthy finding is that recalling satisfying or dissatisfying AI experiences broadly shifted attitudes toward AI, but had limited effects on willingness to rely across task contexts. Exploratory item-level linear mixed-effects analyses likewise did not support the preregistered task-type moderation hypothesis. The priming-by-task-type interaction was significant only for AI, and this effect was larger in human-appropriate than in AI-appropriate tasks, contrary to prediction. For humans, the corresponding interaction was not significant, although the satisfying-versus-dissatisfying contrast in AI-appropriate tasks trended in the negative direction. For experts, neither the interaction nor the within-task-type contrasts were significant.\u003c/p\u003e\n\u003cp\u003eA more plausible explanation is that, in contrast to other measures, the willingness-to-rely measure requires respondents to consider concrete situations. Prior research indicates that relating messages to the context of an individual increases elaboration [74], weakening priming effects [75]. Therefore, willingness-to-rely judgments may be more stable than general attitudes and less susceptible to a brief recall manipulation, while also being more strongly anchored in pre-existing beliefs about task appropriateness.\u003c/p\u003e\n\u003cp\u003e5.4. Limitations\u003c/p\u003e\n\u003cp\u003eThis study had several limitations. Although exploratory follow-up analyses helped clarify the null spillover findings, they were not preregistered and should therefore be interpreted cautiously. In addition, while multiple mechanisms may underlie the present findings, the study design did not permit their direct testing. Our conclusions are thus limited to suggesting that increases in AI evaluations do not necessarily accompany changes in human evaluations and do not support stronger claims about enduring beliefs in uniquely human values. Relatedly, the priming manipulation focused on satisfaction or dissatisfaction with AI use rather than more existential dimensions such as agency, autonomy, or sociality. Recalling use experiences may have encouraged participants to view AI as a convenient tool, reducing the likelihood of perceiving it as a social entity comparable to humans. Future studies should examine whether similar patterns emerge when recall manipulations target AI’s autonomy or humanity. Finally, although experts were introduced as a group theoretically positioned to benefit from AI complementarity, we did not assess whether participants perceived them as AI designers, managers, or value-enhancing agents, so these stratified effects may not have been fully captured.\u003c/p\u003e\n\u003cp\u003e5.5. Contributions\u003c/p\u003e\n\u003cp\u003eDespite these limitations, this study makes several contributions to the literature. Using a sufficiently large sample and a preregistered design, we demonstrated that recalling dissatisfying AI experiences leads to broad declines in attitudes toward AI, including willingness to pay for a paid plan. This finding suggests that in promoting AI services, reducing errors and emphasizing safety and reliability may be more important than highlighting superior performance alone, providing practical implications for AI service design and marketing. In addition, because a simple recall-based priming manipulation shifted attitudes and behavioral intentions toward AI, this approach may be helpful for future experimental research examining AI-related attitudes.\u003c/p\u003e\n\u003cp\u003eThe primary contribution of this study is its approach to the widely discussed concern that AI undermines human values, replaces humans, or exacerbates inequality. Rather than focusing on whether people merely feel such anxieties, we examined how they evaluated humans in general and experts after forming positive or negative attitudes toward AI. The exploratory follow-up analyses sharpened this interpretation by showing that target specificity was clearest for perceived competence and by providing moderate Bayesian support for the absence of corresponding effects on human and expert trust and competence. Thus, the results showed that AI-related experiences altered evaluations of AI itself, but did not produce evidence of a broad zero-sum devaluation of humans or experts. These findings suggest that some contemporary concerns about the negative societal impact of AI may be overstated and necessitate a more nuanced view of how AI development relates to human values.\u003c/p\u003e\n\u003cp\u003eFinally, although the societal effects of AI are undoubtedly complex and cannot be reduced to simple positive or negative outcomes, we hope that this study contributes to a more balanced understanding of how humans and AI may coexist in practice.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003eAcknowledgements\u003c/p\u003e\n\u003cp\u003eThe authors would like to thank the Keio Economic Research Institute for its support in English-language editing. We are also grateful to Ryoya Nakano for his valuable assistance with the Japanese translation of the measurement scales, as well as for his support in refining the English wording of the Japanese stimulus materials used in the study. In addition, we acknowledge the Japan Science and Technology Agency (JST) for providing administrative and institutional support related to the authors’ academic activities during the course of this research.\u003c/p\u003e\n\u003cp\u003eAuthor Contributions\u003c/p\u003e\n\u003cp\u003eKasumi Dan contributed to conceptualization, methodology, data curation, formal analysis, investigation, software implementation, validation, and drafting of the original manuscript, as well as reviewing and editing the manuscript. Takahiro Hoshino contributed to conceptualization, investigation, supervision, project administration, funding acquisition, validation, and critical review and editing of the manuscript. Both authors approved the final version of the manuscript and agree to be accountable for all aspects of the work.\u003c/p\u003e\n\u003cp\u003eAdditional Information\u003c/p\u003e\n\u003cp\u003eCompeting interests\u003c/p\u003e\n\u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e\n\u003cp\u003eFunding Declaration\u003c/p\u003e\n\u003cp\u003eThis work was supported by the JSPS KAKENHI under Grant JP22H03552.\u003c/p\u003e\n\u003cp\u003eData Availability\u003c/p\u003e\n\u003cp\u003eThe anonymized quantitative dataset and analysis code are available at the Open Science Framework (OSF): https://osf.io/g3yjh/overview?view_only=3a76dd0e579740378040a6f39e7a61ff\u003c/p\u003e\n\u003cp\u003eOpen-ended responses are not publicly available because they may contain potentially identifiable information in free-text form. Access requests will be considered by the corresponding author on reasonable request, subject to ethical and privacy restrictions.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eOpenAI. (2025). How people are using ChatGPT. https://openai.com/index/how-people-are-using-chatgpt/\u003c/li\u003e\n\u003cli\u003eHussein, H., Gordon, M., Hodgkinson, C., Foreman, R., \u0026amp; Wagad, S. (2025). ChatGPT\u0026rsquo;s Impact Across Sectors: A Systematic Review of Key Themes and Challenges. Big Data and Cognitive Computing, 9(3), 56. https://doi.org/10.3390/bdcc9030056\u003c/li\u003e\n\u003cli\u003eCui, H., \u0026amp; Yasseri, T. (2024). AI-enhanced collective intelligence. Patterns, 5(11), 101074. https://doi.org/10.1016/j.patter.2024.101074\u003c/li\u003e\n\u003cli\u003eLuo, X., Rechardt, A., Sun, G., Nejad, K. K., Y\u0026aacute;\u0026ntilde;ez, F., Yilmaz, B., Lee, K., Cohen, A. O., Borghesani, V., Pashkov, A., Marinazzo, D., Nicholas, J., Salatiello, A., Sucholutsky, I., Minervini, P., Razavi, S., Rocca, R., Yusifov, E., Okalova, T., . . . Love, B. C. (2025). Large language models surpass human experts in predicting neuroscience results. Nature Human Behaviour, 9(2), 305-315. https://doi.org/10.1038/s41562-024-02046-9\u003c/li\u003e\n\u003cli\u003eLi, H., Zhang, R., Lee, Y., Kraut, R. E., \u0026amp; Mohr, D. C. (2023). Systematic review and meta-analysis of AI-based conversational agents for promoting mental health and well-being. NPJ Digital Medicine, 6(1), 236. https://doi.org/10.1038/s41746-023-00979-5\u003c/li\u003e\n\u003cli\u003eRehman, A. U., Behera, R. K., Islam, M. S., Abbasi, F. A., \u0026amp; Imtiaz, A. (2024). Assessing the usage of ChatGPT on life satisfaction among higher education students: The moderating role of subjective health. Technology in Society, 78, 102655. https://doi.org/10.1016/j.techsoc.2024.102655\u003c/li\u003e\n\u003cli\u003eFeng, X., Tian, L., Ho, G. W. K., Yorke, J., \u0026amp; Hui, V. (2025). The Effectiveness of AI Chatbots in Alleviating Mental Distress and Promoting Health Behaviors Among Adolescents and Young Adults: Systematic Review and Meta-Analysis. Journal of Medical Internet Research, 27, e79850. https://doi.org/10.2196/79850\u003c/li\u003e\n\u003cli\u003eWang, Y.-Y., \u0026amp; Wang, Y.-S. (2022). Development and validation of an artificial intelligence anxiety scale: An initial application in predicting motivated learning behavior. Interactive Learning Environments, 30(4), 619\u0026ndash;634. https://doi.org/10.1080/10494820.2019.1674887\u003c/li\u003e\n\u003cli\u003eDong, M., Conway, J. R., Bonnefon, J., Shariff, A., \u0026amp; Rahwan, I. (2024). Fears about artificial intelligence across 20 countries and six domains of application. The American Psychologist, https://doi.org/10.1037/amp0001454\u003c/li\u003e\n\u003cli\u003eJo, H., \u0026amp; Park, D.-H. (2025). The fear of being replaced by generative AI: An examination of influential factors among office workers. Technological Forecasting and Social Change, 220, 124326. https://doi.org/10.1016/j.techfore.2025.124326\u003c/li\u003e\n\u003cli\u003eDe Freitas, J., Agarwal, S., Schmitt, B., \u0026amp; Haslam, N. (2023). Psychological factors underlying attitudes toward AI tools. Nature Human Behaviour, 7(11), 1845-1854. https://doi.org/10.1038/s41562-023-01734-2\u003c/li\u003e\n\u003cli\u003eBales, A. (2025). A polycrisis threat model for AI. AI \u0026amp; Society, 40(8), 6277\u0026ndash;6289. https://doi.org/10.1007/s00146-025-02371-1\u003c/li\u003e\n\u003cli\u003eSelenko, E., Bankins, S., Shoss, M., Warburton, J., \u0026amp; Restubog, S. L. D. (2022).\u003cbr\u003e Artificial intelligence and the future of work: A functional-identity perspective. Current Directions in Psychological Science, 31(3), 272\u0026ndash;279.\u003cbr\u003e https://doi.org/10.1177/09637214221091823\u003c/li\u003e\n\u003cli\u003eAlkhalifah, J. M., Bedaiwi, A. M., Shaikh, N., Seddiq, W., \u0026amp; Meo, S. A. (2024). Existential anxiety about artificial intelligence (AI)- is it the end of humanity era or a new chapter in the human revolution: questionnaire-based observational study. Frontiers in Psychiatry, 15, 1368122. https://doi.org/10.3389/fpsyt.2024.1368122\u003c/li\u003e\n\u003cli\u003eFrenkenberg, A., \u0026amp; Hochman, G. (2025). It\u0026rsquo;s Scary to Use It, It\u0026rsquo;s Scary to Refuse It: The Psychological Dimensions of AI Adoption\u0026mdash;Anxiety, Motives, and Dependency. Systems, 13(2), 82. https://doi.org/10.3390/systems13020082\u003c/li\u003e\n\u003cli\u003eShonhe, L., \u0026amp; Min, Q. (2025). Human-machine collaboration: Exploring professional identity threat within the records and information management community. ASLIB Journal of Information Management. Advance online publication. https://doi.org/10.1108/AJIM-08-2024-0673\u003c/li\u003e\n\u003cli\u003eGoethals, F., \u0026amp; Ziegelmayer, J. L. (2022). Anxiety buffers and the threat of extreme automation: a terror management theory perspective. Information Technology \u0026amp; People, 35(1), 96-118. https://doi.org/10.1108/ITP-06-2019-0304\u003c/li\u003e\n\u003cli\u003eMirbabaie, M., Br\u0026uuml;nker, F., M\u0026ouml;llmann Frick, Nicholas R. J., \u0026amp; Stieglitz, S. (2022). The rise of artificial intelligence \u0026ndash; understanding the AI identity threat at the workplace. Electronic Markets, 32(1), 73-99. https://doi.org/10.1007/s12525-021-00496-x\u003c/li\u003e\n\u003cli\u003eCastelo, N., Bos, M. W., \u0026amp; Lehmann, D. R. (2019).\u003cbr\u003e Task-dependent algorithm aversion. Journal of Marketing Research, 56(5), 809\u0026ndash;825. https://doi.org/10.1177/0022243719851788\u003c/li\u003e\n\u003cli\u003eLogg, J. M., Minson, J. A., \u0026amp; Moore, D. A. (2019). Algorithm appreciation: People prefer algorithmic to human judgment. Organizational Behavior and Human Decision Processes, 151, 90\u0026ndash;103. https://doi.org/10.1016/j.obhdp.2018.12.005\u003c/li\u003e\n\u003cli\u003eYou, S., Cathy, L. Y., \u0026amp; Li, X. (2022). Algorithmic versus Human Advice: Does Presenting Prediction Performance Matter for Algorithm Appreciation? Journal of Management Information Systems, 39(2), 336-365. https://doi.org/10.1080/07421222.2022.2063553\u003c/li\u003e\n\u003cli\u003eKlingbeil, A., Gr\u0026uuml;tzner, C., \u0026amp; Schreck, P. (2024). Trust and reliance on AI: An experimental study on the extent and costs of overreliance on AI. Computers in Human Behavior, 160, 108352. https://doi.org/10.1016/j.chb.2024.108352\u003c/li\u003e\n\u003cli\u003eFiske, S. T. (2018). Stereotype Content: Warmth and Competence Endure. Current Directions in Psychological Science, 27(2), 67-73. https://doi.org/10.1177/0963721417738825\u003c/li\u003e\n\u003cli\u003eGlikson, E., \u0026amp; Woolley, A. W. (2020). Human trust in artificial intelligence: Review of empirical research. Academy of Management Annals, 14(2), 627\u0026ndash;660. https://doi.org/10.5465/annals.2018.0057\u003c/li\u003e\n\u003cli\u003eLee, J. D., \u0026amp; See, K. A. (2004). Trust in Automation: Designing for Appropriate Reliance. Human Factors, 46(1), https://doi.org/10.1518/hfes.46.1.50_30392\u003c/li\u003e\n\u003cli\u003eDietvorst, B. J., Simmons, J. P., \u0026amp; Massey, C. (2015). Algorithm aversion: people erroneously avoid algorithms after seeing them err. Journal of Experimental Psychology.General, 144(1), 114-26. https://doi.org/10.1037/xge0000033\u003c/li\u003e\n\u003cli\u003eSantoro, E., \u0026amp; Monin, B. (2023). The AI Effect: People rate distinctively human attributes as more essential to being human after learning about artificial intelligence advances. Journal of Experimental Social Psychology, 107, 104464. https://doi.org/10.1016/j.jesp.2023.104464\u003c/li\u003e\n\u003cli\u003eHorton Jr, C. B., White, M. W., \u0026amp; Iyengar, S. S. (2023). Bias against AI art can enhance perceptions of human creativity. Scientific Reports, 13(1), 19001. https://doi.org/10.1038/s41598-023-45202-3\u003c/li\u003e\n\u003cli\u003eReich, T., \u0026amp; Teeny, J. D. (2025). Does artificial intelligence cause artificial confidence? Generative artificial intelligence as an emerging social referent. Journal of Personality and Social Psychology, https://doi.org/10.1037/pspa0000450\u003c/li\u003e\n\u003cli\u003eKim, H.-Y., \u0026amp; McGill, A. L. (2025). AI-induced dehumanization. Journal of Consumer Psychology, 35(3), 363\u0026ndash;381. https://doi.org/10.1002/jcpy.1441\u003c/li\u003e\n\u003cli\u003eOsborne, M. R., \u0026amp; Bailey, E. R. (2025). Me vs. the machine? Subjective evaluations of human- and AI-generated advice. Scientific Reports, 15, 3980. https://doi.org/10.1038/s41598-025-86623-6\u003c/li\u003e\n\u003cli\u003eMalone, E., Afroogh, S., \u0026amp; D\u0026apos;Cruz, J. (2025). When Trust Is Zero Sum: Automation\u0026apos;s Threat to Epistemic Agency. Ethics and Information Technology, 27(2), 29. https://doi.org/10.1007/s10676-025-09838-1\u003c/li\u003e\n\u003cli\u003eBankins, S., \u0026amp; Formosa, P. (2023). The Ethical Implications of Artificial Intelligence (AI) For Meaningful Work: JBE. Journal of Business Ethics, 185(4), 725-740. https://doi.org/10.1007/s10551-023-05339-7\u003c/li\u003e\n\u003cli\u003eDang, J., \u0026amp; Liu, L. (2025a). Dehumanization risks associated with artificial intelligence use. The American Psychologist, https://doi.org/10.1037/amp0001542\u003c/li\u003e\n\u003cli\u003eFerdman, A. (2025). AI deskilling is a structural problem. AI \u0026amp; Society. https://doi.org/10.1007/s00146-025-02686-z\u003c/li\u003e\n\u003cli\u003eGieryn, T. F. (1983). Boundary-work and the demarcation of science from non-science: Strains and interests in professional ideologies of scientists. American Sociological Review, 48(6), 781\u0026ndash;795. https://doi.org/10.2307/2095325\u003c/li\u003e\n\u003cli\u003eHaas, P. M. (1992). Epistemic communities and international-policy coordination: Introduction. International Organization, 46(1), 1\u0026ndash;35. https://doi.org/10.1017/S0020818300001442\u003c/li\u003e\n\u003cli\u003eGrundmann, R. (2017). The Problem of Expertise in Knowledge Societies. Minerva: A Review of Science, Learning and Policy, 55(1), 25-48. https://doi.org/10.1007/s11024-016-9308-7\u003c/li\u003e\n\u003cli\u003eHarvey, N., \u0026amp; Fischer, I. (1997). Taking advice: Accepting help, improving judgment, and sharing responsibility. Organizational Behavior and Human Decision Processes, 70(2), 117-133. https://doi.org/10.1006/obhd.1997.2697\u003c/li\u003e\n\u003cli\u003eMeshi, D., Biele, G., Korn, C. W., \u0026amp; Heekeren, H. R. (2012). How Expert Advice Influences Decision Making. PLoS One, 7(11). https://doi.org/10.1371/journal.pone.0049748\u003c/li\u003e\n\u003cli\u003eBailey, P. E., Leon, T., Ebner, N. C., Moustafa, A. A., \u0026amp; Weidemann, G. (2023). A meta-analysis of the weight of advice in decision-making. Current Psychology, 42, 24516\u0026ndash;24541. https://doi.org/10.1007/s12144-022-03573-2\u003c/li\u003e\n\u003cli\u003eKeating, L. A., \u0026amp; Walker, B. W. (2025). Lay theories of expertise: A mixed-methods exploration. Journal of Management Studies. Advance online publication. https://doi.org/10.1111/joms.13188\u003c/li\u003e\n\u003cli\u003eFrey, C. B., \u0026amp; Osborne, M. A. (2017). The future of employment: How susceptible are jobs to computerisation? . Technological Forecasting and Social Change, 114, 254-269,278-280. https://doi.org/10.1016/j.techfore.2016.08.019\u003c/li\u003e\n\u003cli\u003eAcemoglu, D., \u0026amp; Restrepo, P. (2020). Robots and jobs: Evidence from U.S. labor markets. Journal of Political Economy, 128(6), 2188\u0026ndash;2244. https://doi.org/10.1086/705716\u003c/li\u003e\n\u003cli\u003eAutor, D. H., Levy, F., \u0026amp; Murnane, R. J. (2003). The Skill Content of Recent Technological Change: An Empirical Exploration. The Quarterly Journal of Economics, 118(4), 1279-1333. https://doi.org/10.1162/003355303322552801\u003c/li\u003e\n\u003cli\u003eAutor, D. H. (2015). Why Are There Still So Many Jobs? The History and Future of Workplace Automation. The Journal of Economic Perspectives, 29(3), 3-30. https://doi.org/10.1257/jep.29.3.3\u003c/li\u003e\n\u003cli\u003eJacobs, J. (2024). The artificial intelligence shock and socio-political polarization. Technological Forecasting and Social Change, 199, 123006. https://doi.org/10.1016/j.techfore.2023.123006\u003c/li\u003e\n\u003cli\u003eTschang, F. T., \u0026amp; Almirall, E. (2021). Artificial Intelligence as Augmenting Automation: Implications for Employment. The Academy of Management Perspectives, 35(4), 642. https://doi.org/10.5465/amp.2019.0062\u003c/li\u003e\n\u003cli\u003eSeptiandri, A. A., Constantinides, M., \u0026amp; Quercia, D. (2024). The potential impact of AI innovations on US occupations. PNAS Nexus, 3(9), 320. https://doi.org/10.1093/pnasnexus/pgae320\u003c/li\u003e\n\u003cli\u003eNovozhilova, E., Mays, K., \u0026amp; Katz, J. E. (2024). Looking towards an automated future: U.S. attitudes towards future artificial intelligence instantiations and their effect. Humanities \u0026amp; Social Sciences Communications, 11(1), 132. https://doi.org/10.1057/s41599-024-02625-1\u003c/li\u003e\n\u003cli\u003eBusch, F., Hoffmann, L., Xu, L., Zhang, L. J., Hu, B., Garc\u0026iacute;a-Ju\u0026aacute;rez, I., Toapanta-Yanchapaxi, L., Gorelik, N., Gorelik, V., Rodriguez-Granillo, G., Ferrarotti, C., Cuong, N. N., Thi, C. A. P., Tuncel, M., Kaya, G., Solis-Barquero, S., Mendez Avila, M.,C., Ivanova, N. G., Kitamura, F. C., . . . Zhang, S. (2025). Multinational Attitudes Toward AI in Health Care and Diagnostics Among Hospital Patients. JAMA Network Open, 8(6) https://doi.org/10.1001/jamanetworkopen.2025.14452\u003c/li\u003e\n\u003cli\u003eJin, M., Yang, Z., Freling, T. L., \u0026amp; Janakiraman, N. (2025). The human superiority effect in advice taking: A multimethod exploration and implications for policy makers and governmental organizations. Journal of Public Policy \u0026amp; Marketing, 44(3), 350\u0026ndash;369. https://doi.org/10.1177/07439156251320314\u003c/li\u003e\n\u003cli\u003eVanneste, B. S., \u0026amp; Puranam, P. (2025). Artificial intelligence, trust, and perceptions of agency. Academy of Management Review, 50(4), 726\u0026ndash;744. https://doi.org/10.5465/amr.2022.0041\u003c/li\u003e\n\u003cli\u003eDang, J., \u0026amp; Liu, L. (2025b). Public trust in artificial intelligence users. Current Opinion in Psychology, 66, 102148. https://doi.org/10.1016/j.copsyc.2025.102148\u003c/li\u003e\n\u003cli\u003eLumineau, F., Schilke, O., \u0026amp; Wang, W. (2023). Organizational trust in the age of the fourth industrial revolution: Shifts in the nature, production, and targets of trust. Journal of Management Inquiry, 32(1), 21\u0026ndash;34. https://doi.org/10.1177/10564926221127852\u003c/li\u003e\n\u003cli\u003eLongoni, C., \u0026amp; Cian, L. (2022). Artificial intelligence in utilitarian vs. hedonic contexts: The \u0026ldquo;word-of-machine\u0026rdquo; effect. Journal of Marketing, 86(1), 91\u0026ndash;108.\u003c/li\u003e\n\u003cli\u003eQin, X., Zhou, X., Chen, C., Wu, D., Zhou, H., Dong, X., Cao, L., \u0026amp; Lu, J. G. (2025). AI aversion or appreciation? A capability-personalization framework and a meta-analytic review. Psychological Bulletin, 151(5), 580-599. https://doi.org/10.1037/bul0000477\u003c/li\u003e\n\u003cli\u003eDietvorst, B. J., \u0026amp; Bartels, D. M. (2022). Consumers object to algorithms making morally relevant tradeoffs because of algorithms\u0026rsquo; consequentialist decision strategies. Journal of Consumer Psychology, 32(3), 406\u0026ndash;424. https://doi.org/10.1002/jcpy.1266\u003c/li\u003e\n\u003cli\u003eYin, Y., Jia, N., \u0026amp; Wakslak, C. J. (2024). AI can help people feel heard, but an AI label diminishes this impact. Proceedings of the National Academy of Sciences of the United States of America, 121(14) https://doi.org/10.1073/pnas.2319112121\u003c/li\u003e\n\u003cli\u003eMyers, S., \u0026amp; Everett, J. A. C. (2025). People expect artificial moral advisors to be more utilitarian and distrust utilitarian moral advisors. Cognition, 256, 106028. https://doi.org/10.1016/j.cognition.2024.106028\u003c/li\u003e\n\u003cli\u003eMerkley, E. (2020). Anti-intellectualism, populism, and motivated resistance to expert consensus. Public Opinion Quarterly, 84(1), 24\u0026ndash;48. https://doi.org/10.1093/poq/nfz053\u003c/li\u003e\n\u003cli\u003eGefen, D. (2000). E-commerce: The role of familiarity and trust. Omega, 28(6), 725\u0026ndash;737. https://doi.org/10.1016/S0305-0483(00)00021-9\u003c/li\u003e\n\u003cli\u003eRiedl, R., Hogeterp, S. A., \u0026amp; Reuter, M. (2024).\u003cbr\u003e Do patients prefer a human doctor, artificial intelligence, or a blend, and is this preference dependent on medical discipline? Empirical evidence and implications for medical practice. Frontiers in Psychology, 15, 1422177. https://doi.org/10.3389/fpsyg.2024.1422177\u003c/li\u003e\n\u003cli\u003eAbele, A. E., Hauke, N., Peters, K., Louvet, E., Szymkow, A., \u0026amp; Duan, Y. (2016). Facets of the Fundamental Content Dimensions: Agency with Competence and Assertiveness-Communion with Warmth and Morality. Frontiers in Psychology, 7, 1810. https://doi.org/10.3389/fpsyg.2016.01810\u003c/li\u003e\n\u003cli\u003eBhattacherjee, A. (2001). Understanding information systems continuance: An expectation-confirmation model. MIS Quarterly, 25(3), 351\u0026ndash;370. https://doi.org/10.2307/3250921\u003c/li\u003e\n\u003cli\u003eVenkatesh, V., Morris, M. G., Davis, G. B., \u0026amp; Davis, F. D. (2003). User acceptance of information technology: Toward a unified view. MIS Quarterly, 27(3), 425\u0026ndash;478. https://doi.org/10.2307/30036540\u003c/li\u003e\n\u003cli\u003eChang, V., Yang, Y., Xu, Q. A., \u0026amp; Xiong, C. (2021). Factors influencing consumer intention to subscribe to the premium music streaming services in China. Journal of Global Information Management, 29(6), 1\u0026ndash;25. https://doi.org/10.4018/JGIM.20211101.oa17\u003c/li\u003e\n\u003cli\u003eMahmud, H., Islam, A. N., Ahmed, S. I., \u0026amp; Smolander, K. (2022). What influences algorithmic decision-making? A systematic literature review on algorithm aversion. Technological Forecasting and Social Change, 175, 1. https://doi.org/10.1016/j.techfore.2021.121390\u003c/li\u003e\n\u003cli\u003eAfroogh, S., Akbari, A., Malone, E., Kargar, M., \u0026amp; Alambeigi, H. (2024). Trust in AI: Progress, challenges, and future directions. Humanities \u0026amp; Social Sciences Communications, 11(1), Article 1568. https://doi.org/10.1057/s41599-024-04044-8\u003c/li\u003e\n\u003cli\u003ePlacani, A. (2024). Anthropomorphism in AI: Hype and fallacy. AI and Ethics, 4, 691\u0026ndash;698. https://doi.org/10.1007/s43681-024-00419-4\u003c/li\u003e\n\u003cli\u003ePeter, S., Riemer, K., \u0026amp; West, J. D. (2025). The benefits and dangers of anthropomorphic conversational agents. Proceedings of the National Academy of Sciences of the United States of America, 122(22). https://doi.org/10.1073/pnas.2415898122\u003c/li\u003e\n\u003cli\u003eBonnefon, J., Rahwan, I., \u0026amp; Shariff, A. (2024). The Moral Psychology of Artificial Intelligence. Annual Review of Psychology, 75, 653-675. https://doi.org/10.1146/annurev-psych-030123-113559\u003c/li\u003e\n\u003cli\u003eCha, Y.-J., Baek, S., Ahn, G., Lee, H., Lee, B., Shin, J.-E., \u0026amp; Jang, D. (2020). Compensating for the loss of human distinctiveness: The use of social creativity under human\u0026ndash;machine comparisons. Computers in Human Behavior, 103, 80\u0026ndash;90. https://doi.org/10.1016/j.chb.2019.08.027\u003c/li\u003e\n\u003cli\u003ePetty, R. E., Cacioppo, J. T., \u0026amp; Goldman, R. (1981). Personal Involvement as a Determinant of Argument-Based Persuasion. Journal of Personality and Social Psychology, 41(5), 847. https://doi.org/10.1037/0022-3514.41.5.847\u003c/li\u003e\n\u003cli\u003eThompson, E. P., Roman, R. J., Moskowitz, G. B., Chaiken, S., \u0026amp; Bargh, J. A. (1994). Accuracy motivation attenuates covert priming: The systematic reprocessing of social information. Journal of Personality and Social Psychology, 66(3), 474. https://doi.org/10.1037/0022-3514.66.3.474\u003c/li\u003e\n\u003c/ol\u003e"},{"header":"Tables","content":"\u003cp\u003eTable1. Demographics of our Sample\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"747\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eDemographics\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eSatisfying\u0026ndash;AI (n = 111)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eSatisfying\u0026ndash;Human (n = 110)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eSatisfying\u0026ndash;Expert (n = 116)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eDissatisfying\u0026ndash;AI (n = 118)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eDissatisfying\u0026ndash;Human (n = 104)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eDissatisfying\u0026ndash;Expert (n = 126)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eNeutral\u0026ndash;AI (n = 106)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eNeutral\u0026ndash;Human (n = 93)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eNeutral\u0026ndash;Expert (n = 103)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eAll (n = 987)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eAge\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e18-24\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e9.0%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e23.6%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e8.6%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e7.6%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e20.2%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e13.5%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e6.6%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e25.8%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e7.8%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e13.4%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e25-34\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e20.7%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e24.5%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e22.4%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e22.0%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e25.0%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e18.3%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e21.7%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e24.7%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e24.3%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e22.5%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e35-44\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e27.9%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e28.2%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e31.0%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e24.6%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e23.1%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e29.4%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e29.2%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e17.2%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e26.2%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e26.5%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e45-54\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e23.4%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e15.5%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e21.6%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e27.1%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e22.1%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e24.6%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e20.8%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e21.5%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e25.2%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e22.5%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e55-64\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e15.3%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e8.2%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e12.1%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e16.9%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e7.7%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e12.7%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e17.0%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e9.7%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e13.6%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e12.7%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e65-74\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e2.7%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.0%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e4.3%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.8%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.9%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.8%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e3.8%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.1%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.9%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.9%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e75 +\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.9%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.0%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.0%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.8%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.0%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.8%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.9%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.0%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.0%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.5%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eMean\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e41.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e35.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e41.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e42.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e37.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e40.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e42.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e36.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e41.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e39.9\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eGender\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eMale\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e61.3%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e51.8%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e67.2%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e55.9%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e60.6%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e65.1%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e63.2%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e61.3%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e66.0%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e61.4%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eFemale\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e38.7%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e48.2%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e32.8%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e44.1%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e39.4%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e34.9%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e36.8%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e38.7%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e34.0%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e38.6%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eOthers\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.0%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.0%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.0%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.0%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.0%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.0%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.0%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.0%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.0%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.0%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eDemographic characteristics of participants across the nine experimental conditions in a 3 (AI experience recall: satisfying vs. dissatisfying vs. neutral) \u0026times; 3 (evaluation target: AI vs. humans vs. experts) between-subjects design (N = 987). Values represent percentages within each condition unless otherwise noted; mean age is reported in years.\u003c/p\u003e\n\u003cp\u003eTable 2. Effects of AI Experience Recall on AI Continuance Intention and Willingness to Pay\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"747\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eMeasure\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003ePriming Condition\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eSatisfying AI experience (n = 337, M \u0026plusmn; SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eDissatisfying AI experience (n = 348, M \u0026plusmn; SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eNeutral AI experience (n = 302, M \u0026plusmn; SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eF(2, 984)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003ep\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026eta;\u0026sup2;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003ePost-hoc (Bonferroni)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eIntention to Continue Using AI\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e5.87 \u0026plusmn; 1.07\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e5.55 \u0026plusmn; 1.35\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e5.78 \u0026plusmn; 1.18\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e6.37\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.002\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.013\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eDissatisfying \u0026lt; Satisfying\u003c/p\u003e\n \u003cp\u003eDissatisfying \u0026lt; Neutral\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eWillingness to Pay for a Paid AI Plan\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e3.49 \u0026plusmn; 1.54\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e3.14 \u0026plusmn; 1.56\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e3.37 \u0026plusmn; 1.51\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e4.69\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.009\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.009\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eDissatisfying \u0026lt; Satisfying\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eEffects of AI experience recall (satisfying vs. dissatisfying vs. neutral) on overall attitudes toward generative AI. Means (M) and standard deviations (SD) are reported for AI continuance intention and willingness to pay for a paid AI plan (7-point scales). One-way ANOVAs tested the main effect of priming condition across the full sample (N = 987), followed by Bonferroni-adjusted post hoc comparisons.\u003c/p\u003e\n\u003cp\u003eTable 3. Effects of AI Experience Recall on Evaluations of AI, Humans, and Experts\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"756\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eTarget/Measure\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003ePriming Condition\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eSatisfying AI experience (M \u0026plusmn; SD)\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eDissatisfying AI experience (M \u0026plusmn; SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eNeutral AI experience (M \u0026plusmn; SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eF\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003ep\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026eta;\u0026sup2;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003ePost-hoc (Bonferroni)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u003cem\u003eAI\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eF(2, 332)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eTrust\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e4.32 \u0026plusmn; 1.22\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e3.90 \u0026plusmn; 1.19\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e4.13 \u0026plusmn; 1.32\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e3.28\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.039\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.019\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eDissatisfying \u0026lt; Satisfying\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePerceived Competence\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e5.52 \u0026plusmn; 0.86\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e4.96 \u0026plusmn; 1.10\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e5.40 \u0026plusmn; 0.99\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e10.11\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026lt; 0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.057\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eDissatisfying \u0026lt; Satisfying\u003c/p\u003e\n \u003cp\u003eDissatisfying \u0026lt; Neutral\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eWillingness to Rely (AI-appropriate)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e5.28 \u0026plusmn; 1.04\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e5.11 \u0026plusmn; 1.09\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e5.27 \u0026plusmn; 1.15\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.85\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.430\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.005\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026ndash;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eWillingness to Rely (Human-appropriate)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e3.86 \u0026plusmn; 1.36\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e3.50 \u0026plusmn; 1.39\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e3.56 \u0026plusmn; 1.38\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e2.26\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.106\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.013\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026ndash;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u003cem\u003eHumans\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eF(2, 304)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eTrust\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e3.83 \u0026plusmn; 1.23\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e3.72 \u0026plusmn; 1.24\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e4.11 \u0026plusmn; 1.34\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e2.47\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.086\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.016\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026ndash;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePerceived Competence\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e3.95 \u0026plusmn; 1.14\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e3.87 \u0026plusmn; 1.17\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e3.90 \u0026plusmn; 1.17\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.11\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.897\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026ndash;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eWillingness to Rely (AI-appropriate)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e3.52 \u0026plusmn; 1.17\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e3.81 \u0026plusmn; 1.20\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e3.72 \u0026plusmn; 1.23\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.64\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.195\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.011\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026ndash;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eWillingness to Rely (Human-appropriate)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e4.82 \u0026plusmn; 1.22\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e4.85 \u0026plusmn; 1.03\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e4.92 \u0026plusmn; 1.03\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.24\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.791\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.002\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026ndash;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u003cem\u003eExperts\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eF(2, 342)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eTrust\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e4.65 \u0026plusmn; 1.19\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e4.49 \u0026plusmn; 1.21\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e4.80 \u0026plusmn; 1.11\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.94\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.145\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.011\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026ndash;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePerceived Competence\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e4.96 \u0026plusmn; 1.08\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e4.97 \u0026plusmn; 1.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e5.04 \u0026plusmn; 1.08\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.820\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026ndash;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eWillingness to Rely (AI-appropriate)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e4.50 \u0026plusmn; 1.18\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e4.44 \u0026plusmn; 1.24\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e4.53 \u0026plusmn; 0.98\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.18\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.839\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026ndash;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eWillingness to Rely (Human-appropriate)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e4.34 \u0026plusmn; 1.09\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e4.21 \u0026plusmn; 1.25\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e4.33 \u0026plusmn; 0.91\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.51\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.603\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.003\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026ndash;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eEffects of AI experience recall (satisfying vs. dissatisfying vs. neutral) on evaluations of three targets (AI, humans, and experts). Means (M) and standard deviations (SD) are reported for trust, perceived competence, and willingness to rely (7-point scales). One-way ANOVAs were conducted separately within each evaluation target to test the effect of priming condition. Willingness to rely is presented for aggregated AI-appropriate and human-appropriate task contexts. Bonferroni-adjusted post hoc comparisons are reported where significant.\u003c/p\u003e\n\u003cp\u003eTable 4. Direct comparison of the satisfying\u0026ndash;dissatisfying contrast across targets\u003c/p\u003e\n\u003cp\u003ePanel A. Omnibus interaction test\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eMeasure\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003ePriming \u0026times; Target\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003edf\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003ep\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eTrust\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eF = 1.03\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e(4, 978)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.390\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003ePerceived competence\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eF = 2.97\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e(4, 978)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.019\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003ePanel B. Planned contrasts (target-standardized scores)\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eMeasure\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eAI\u003cbr\u003e\u0026nbsp;Sat \u0026minus; Dis\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eHuman\u003cbr\u003e\u0026nbsp;Sat \u0026minus; Dis\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eExpert\u003cbr\u003e\u0026nbsp;Sat \u0026minus; Dis\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eAI \u0026minus; Human\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eAI \u0026minus; Expert\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eTrust\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.34 (0.011)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.09 (0.506)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.14 (0.290)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.25 (0.196)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.20 (0.277)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003ePerceived competence\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.55 (\u0026lt; 0.001)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.06 (0.639)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026minus;0.02 (0.901)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.49 (0.010)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.57 (0.002)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eFollow-up analyses on target-standardized trust and perceived competence scores directly comparing the planned satisfying-versus-dissatisfying priming contrast across AI, human, and expert targets. Panel A reports Type III tests from models including priming condition, target, and their interaction. Panel B reports the satisfying-versus-dissatisfying contrast within each target and the corresponding difference-of-differences comparing AI with humans and experts.Scores were standardized within target before analysis; positive estimates indicate higher evaluations in the satisfying condition than in the dissatisfying condition. Raw means and standard deviations for each priming condition are reported in Table 3.\u003c/p\u003e\n\u003cp\u003eTable 5. Bayesian t-tests for satisfying vs. dissatisfying comparisons in human and expert evaluations\u003c/p\u003e\n \u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eTarget\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eOutcome\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eMean diff.\u003cbr\u003e\u0026nbsp;(Sat \u0026minus; Dis)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eBF01\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eNHST p\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eHuman\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eTrust\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.12\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e5.38\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.494\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eHuman\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003ePerceived competence\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.07\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e6.05\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.642\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eHuman\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eRely (AI-appropriate)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026minus;0.29\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e1.50\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.075\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eHuman\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eRely (Human-appropriate)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026minus;0.03\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e6.60\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.856\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eExpert\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eTrust\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.16\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e4.29\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.303\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eExpert\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003ePerceived competence\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026minus;0.02\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e7.05\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.901\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eExpert\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eRely (AI-appropriate)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e6.75\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.743\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eExpert\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eRely (Human-appropriate)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.13\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e5.05\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.396\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003eBayesian independent-samples t-tests comparing the satisfying and dissatisfying priming conditions for trust, perceived competence, and aggregate willingness-to-rely outcomes in the human and expert target conditions. Mean differences are reported as satisfying minus dissatisfying; positive values indicate higher evaluations in the satisfying condition. BF01 quantifies evidence for the null hypothesis of no difference.\u0026nbsp;Bayes factors were computed using the default Cauchy prior (\u003cimg width=\"59\" height=\"24\" src=\"data:image/png;base64,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\" v:shapes=\"_x0000_i1025\" alt=\"image\"\u003e). BF01 values greater than 3 indicate at least moderate evidence for the null hypothesis. NHST p values are shown for reference. AI evaluations are not included because these Bayesian analyses were intended to quantify the absence of spillover to non-AI targets.\u003c/p\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"attitudes toward AI, priming effects, AI usage intention, human–AI evaluation, algorithm aversion","lastPublishedDoi":"10.21203/rs.3.rs-9267872/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9267872/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eGenerative artificial intelligence (AI) is spreading rapidly in everyday life, raising concerns that valuing AI may undermine the perceived value or necessity of humans while benefiting only certain groups. Yet little work tests whether shifts in AI attitudes spill over to evaluations of humans or experts. In a 3x3 between-subjects experiment (N\u0026thinsp;=\u0026thinsp;987), participants recalled satisfying, dissatisfying, or neutral AI experiences and then evaluated one target (AI, humans, or experts). Recalling satisfying versus dissatisfying experiences shifted trust in AI, perceived competence, continuance intention, and willingness to pay for a plan, with stronger decreases after dissatisfying recall. In contrast, evaluations of humans and experts (trust, competence, and willingness to rely across contexts) showed no significant differences across priming conditions. Thus, we find no evidence that AI attitudes translate into zero-sum devaluation of humans or experts, suggesting largely independent evaluative systems and supporting coexistence of technological and human trust in practice.\u003c/p\u003e","manuscriptTitle":"Recalled AI Experiences Shift Evaluations of AI but Not of Humans or Experts","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-04-29 07:05:52","doi":"10.21203/rs.3.rs-9267872/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"bd555969-0fb4-4947-825b-d1ec3599f505","owner":[],"postedDate":"April 29th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":67047623,"name":"Biological sciences/Psychology"},{"id":67047624,"name":"Social science/Psychology"},{"id":67047625,"name":"Social science/Science technology and society"}],"tags":[],"updatedAt":"2026-04-29T07:05:52+00:00","versionOfRecord":[],"versionCreatedAt":"2026-04-29 07:05:52","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9267872","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9267872","identity":"rs-9267872","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}