Knowledge-based representations of artificial intelligence and divine agents: A developmental study across Japan and the United States

Knowledge-based representations of artificial intelligence and divine agents: A developmental study across Japan and the United States | 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 Knowledge-based representations of artificial intelligence and divine agents: A developmental study across Japan and the United States Yusuke Moriguchi, Ryoichi Watanabe, Yusuke Hori, Bolivar Reyes-Jaquez This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6824602/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 11 You are reading this latest preprint version Abstract Advances in generative artificial intelligence (AI) have raised foundational questions about how we conceptualise such entities alongside traditional agents like humans and divine beings. Research demonstrates that both children and adults tend to attribute similar mental properties to AI and divine beings across cultures. However, fundamental differences exist in the nature of knowledge attributed to humans, AI and divine entities. Here we showed that Japanese children and adults across developmental stages conceptually cluster AI with divine entities in knowledge-based representational space, whereas U.S. participants maintain clear categorical boundaries between these agents using representational similarity analyses focusing on knowledge attributes. Our findings stand in contrast with previous cross-cultural investigations reporting shared basic agent structures across cultures. These results offer implications for theoretical models of agent categorisation, cross-cultural AI design, and our understanding of how humans reason about increasingly intelligent artificial minds in culturally contingent ways. Social science/Cultural and media studies Social science/Psychology artificial intelligence divine entities children agent perception development Figures Figure 1 Figure 2 Figure 3 Figure 4 Introduction Understanding how humans perceive and interact with artificial intelligence (AI) and robots has become an important area of research 1 , 2 . As AI and robots increasingly become part of our daily lives, it is essential to explore how humans conceptualise these advanced technologies. Research demonstrates that both adults and children attribute mental states—including thoughts, goals, and intentions—to AI systems and robots 3 – 7 . The literature on agent perception typically examines two primary dimensions: experience (capacity for sensations and emotions) and agency (ability for planning, thinking, and self-control) 6 , 8 , 9 . Previous studies have investigated how individuals attribute experience and agency to AI, robots, humans, animals, and divine entities, and shown that AI/robots and divine entities are attributed agency but perceived as lacking experience—a pattern observed across diverse cultural contexts 6 , 10 . However, the emergence of generative AI technologies such as ChatGPT and Gemini may challenge these traditional patterns of agent perception 11 – 13 . The ability of these AI systems with superhuman capabilities, such as super-knowledge, could blur or extend the established boundaries between humans, divine entities, and artificial minds, prompting a fundamental reconsideration of how we attribute mental states and capabilities to artificial agents 14 – 18 . Theorists have proposed that there are fundamental differences between the knowledge of AI and humans 11 , 19 – 21 . While LLMs acquire knowledge through next-word prediction as an instrument, humans develop structured world models that preserve representations of physical reality. In religious traditions, God is understood to possess omniscience—complete and perfect knowledge that surpasses human comprehension 22 , 23 . God’s knowledge can be conceptualised as a form of worldly knowledge—a complete and perfect understanding of the world’s true nature and causal relationships. This distinction between instrumental and worldly knowledge provides a novel framework for understanding how children and adults represent different types of agents. Differences in knowledge between agents raises important developmental and cultural questions. Research suggests that younger children from Western and non-Western societies initially anthropomorphise both AI and divine entities, attributing human-like knowledge limitations to them 4 , 24 – 26 . Developmental progression reveals increasing differentiation: divine entities become viewed as possessing complete worldly knowledge, whereas AI systems are perceived as tools having extensive knowledge 27 – 30 . However, the developmental trajectory of the conceptual relationships between AI and divine entities remains uncharacterised. Additionally, cultural factors can influence mental capacity attributions to non-human agents. Cross-cultural investigations indicate that while the basic agent structure is similar across both Western (e.g. USA) and non-Western (e.g. Japan) populations 6 , 10 , attitudes toward and anthropomorphic tendencies regarding AI can vary depending on cultural context 31 , 32 . Moreover, cultures with animistic or polytheistic traditions (such as Japan) may foster greater acceptance of human-like machines, whereas Western cultural frameworks might engender more cautious attitudes toward artificial agents 33 . These cultural differences could influence how participants conceptualise entities with different types of knowledge capabilities, potentially affecting the perceived relationship between AI systems and divine entities. The present investigation examines whether conceptualisations of knowledge in AI and divine entities vary between Japanese and U.S. populations, and, if so, when such cultural differences emerge in human development. By focusing specifically on knowledge attribution—rather than exclusively on experience or agency—we address a critical aspect of AI and divine entity comparisons. To investigate these issues, we employed two distinct but complementary approaches. First, we conducted a comprehensive evaluation using traditional questioning methods to compare how children perceive divine entities, AI, and humans across multiple dimensions, including biological attributes (e.g. ‘Can X breathe?’), psychological capabilities (e.g. ‘Can X feel sad?’), and extraordinary capabilities (e.g. ‘Can X know everything?’) (Experiments 1) 5 . This holistic assessment provided insights into how these agents are conceptualised across different domains. Building on these findings, our second approach specifically focused on the dimension of knowledge, employing a structural method using representational similarity analyses, a technique that has gained prominence in cognitive neuroscience research, particularly in studies of perception and categorisation (Experiments 2 and 3) 34 , 35 . This structural approach built on our previous work, in which we developed a similarity rating task suitable for young children 36 . In this study, we applied this methodology to analyse the patterns of similarity ratings among diverse entities, including humans, AI, and divine entities. The similarity scores were then compiled into a similarity matrix, from which we visualised the structure using dimensional reduction techniques, such as multi-dimensional scaling (Fig. 2 ). Considering a diverse range of entities in our similarity ratings helped us gain a structural understanding of how children categorise and relate different types of agents. From developmental perspectives, we can observe how children’s initial representations of AI and divine entities develop over time, whether their early tendency to cluster these entities with humans shifts with age, and how their conceptual framework for understanding different types of agents becomes more sophisticated. While such representational structures have been examined in adults 6 , our study uniquely captures how these conceptual organisations emerge and transform throughout development. We first examined these aspects in the Japanese context and then conducted the same experiments in the U.S. context to assess whether the results in Japan could be generalised. Results Data were collected in Japan first, followed by the U.S. However, to further improve the paper’s readability, findings are presented based on subtopics, rather than chronology. Attribution of Biological, Psychological, and Extraordinary Properties In Experiment 1, we recruited 73 Japanese preschool children and examined whether they similarly attributed biological (hungry, breathing), psychological (happy, sad), and extraordinary properties (super-knowledge, super-problem solving) to a human adult, AI, and God [a] , using a traditional questioning method 5 . Moreover, we preregistered this experiment and collected data from 36 U.S. preschool children. For example, children were asked questions such as, ‘Can a person solve any difficult problem?’ and ‘Can a person know everything?’ for extraordinary properties, and responded “yes” or “no” for each question. As shown in Fig. 1 , both Japanese and U.S. children differentiated each agent in terms of biological and psychological properties, but equally attributed extraordinary properties to each agent. We analysed whether children’s ‘yes’ responses differed as a function of the agent (A person, AI, and God); property (biological, psychological, and extraordinary); and culture (Japan vs. U.S.) using a mixed analysis of variance (ANOVA). We found a significant interaction between agent and property [F(4, 428) = 26.895, p < .001, η 2 G = 0.065]. Post-hoc comparison using Shaffer’s methods revealed significant differences between agents in each property (ps < .025), but the effect size was smaller for extraordinary properties (η 2 G = 0.019) compared to biological (η 2 G = 0.288) and psychological properties (η 2 G = 0.249). Specifically, across cultures children attributed extraordinary properties to God more than AI (p = .019), but no other differences were found (ps > .088). No significant interactions between agent, property, and culture were observed [F (4, 428) = 0.299, p < .001, η 2 G = 0.000]. Thus, both Japanese and U.S. children attributed the extraordinary properties to each agent in a relatively similar manner, compared to the biological and psychological properties. However, the small differences between agents in extraordinary properties do not mean that the children regarded a person, AI, and God as similar in terms of knowledge. Consequently, more direct comparisons are conducted in subsequent experiments. Similarity Structure of Agents Based on Knowledge Next, we directly examined whether children and adults believe that AI and God are similar in terms of knowledge using the similarity rating method. In Experiment 2, participants reported the subjective similarity between different pairs of agents. To fully understand the cognitive structure underlying children‘s agent perception, we extended our analysis beyond just the target items (a human lady, AI, and God) to include non-target items, such as a human girl, a robot, and a frog, based on a previous study 6 (for each image, Fig. 2 ). Two agents (e.g. human lady and God) were presented simultaneously each trial. Participants evaluated how similar they thought various pairs of living creatures and artefacts were to each other, in terms of how much knowledge they have about things. Next, they chose one from the following four options: ‘Dissimilar’, ‘Slightly dissimilar’, ‘Slightly similar’, or ‘Similar’. In Japan, we recruited 61 preschool and 54 school-aged Japanese children and 40 adults, and administered a similarity task to investigate how participants evaluated the relationship between agents. We also recruited 32 preschool and 38 school-aged U.S. children and 80 U.S. adults (40 religious and 38 non-religious) after the pre-registration. To visualise participants’ responses, we first present the dissimilarity matrices in Fig. 2 (panels A, B), while considering the mean for each agent comparison for a particular age group in each culture. White boxes represent the dissimilarity between agents, whereas black boxes represent the similarity across agents. For visual inspection, we focus on the blue cluster in Fig. 2 A (blue square in the Japanese category). The blue cluster represents the similarity between AI, robot and God. The similarities between AI and robot were generally high, which is not surprising. Importantly, similarities between AI/robot and God differed across cultures. That is, the Japanese participants rated the AI/robot and God as similar compared to other pairs, whereas such patterns were not observed in the U.S. population. To obtain a better qualitative understanding, we visualised the geometric relationships between agents using two-dimensional multi-dimensional scaling (MDS), which maintained the relational structures of agent dissimilarities (Fig. 2 , panels C, D). In MDS, the nearer the agents were, the more similar. There was a striking similarity between AI/robot and God in Japanese children and adults (blue circles in the Japanese category), but not in U.S. participants. The position of God clearly differed across age and cultural factors. Next, we examined whether the overall structure of the agents was similar across the developmental and cultural groups. Group-level dissimilarity matrices obtained from children and adults within/between the cultural groups appeared very comparable. Indeed, all the correlation coefficients between all pairs were > 0.72 (Spearman‘s ρ). The highest correlation was observed between Japanese adults and Japanese school children (ρ = .92), while the lowest correlation coefficient emerged between Japanese adults and U.S. preschool children (ρ = .72). Within each culture, all the correlation coefficients between any pair of age groups were > .89 in Japan and > .81 in the USA. These results indicate that the overall structure of agents was similar across the groups. However, visual inspection suggests that U.S. participants represented AI/robots differently from God, whereas Japanese participants regarded AI/robots and God as similar (Fig. 2 ). To quantify this result, we specifically focused on the similarity between AI/robots and God and examined whether the average AI-God and robot-God similarity scores (not including AI-robot similarity due to the high similarity) were distinctly higher than the other similarity scores. Specifically, we averaged the similarity scores of all pairs except the AI-God, robot-God, AI-robot, and the identical pairs used as controls (e.g. robot-robot), which were regarded as similarity scores of other pairs. We conducted mixed ANOVAs with the agent (AI/robot-God vs. other pairs); age group (preschool, school, and adult); and culture (Japan vs. the U.S.) as predictors of similarity scores (Fig. 3 , panel A). We found a significant interaction between agent and culture [F(1, 297) = 48.013, p < .001, η 2 G = 0.041]. Post-hoc analysis using Shaffer’s method revealed that the similarity between AI/robot and God was higher than that for other pairs in the Japanese population, but the opposite pattern was observed in the U.S. population (ps < .05). The interaction between the three factors was not significant [F(2, 297) = 0.081, p = .921, η 2 G = 0.000]. The results suggest that Japanese children and adults evaluated AI/robots and God as more similar than any other pairs, while U.S. children and adults evaluated AI/robots and God as more dissimilar than any other pairs. The analysis of age and cultural effects on each pair is reported in Supplemental analysis and Figure S1 . Similarity Matrices of Animacy The analyses above demonstrate that Japanese participants, but not U.S. participants, perceived similarities between AI/robots and God in terms of knowledge. Building on these findings and the biological attribution results from Experiment 1, we conducted Experiment 3 to examine whether this perceived similarity between AI/robots and God is specific to these agents’ knowledge characteristics, or extends to their biological/animate nature. As shown in Experiment 1 (Fig. 1 ), participants attributed different biological characteristics to AI and God. This suggests that while these agents might share knowledge qualities, they may be perceived differently in terms of their animate nature. Indeed, it is well known that children have a naive biology theory (e.g. animate/inanimate) distinct from a naive psychology theory (e.g. knowledge of things) 37 – 39 . To directly test this hypothesis, we conducted a similarity judgement task that focused specifically on animacy in Experiment 3. Since US participants did not perceive similarities between AI/robots and God in Experiment 2, we focused solely on Japanese participants in Experiment 3. We recruited 23 preschool children, 55 school-aged children, and 40 adults and administered the same similarity task using the same stimuli as in Experiment 2. However, this time, participants were explicitly asked to judge similarities based on animacy. If the similarity between AI/robots and God observed in Experiment 2 is indeed specific to knowledge, we would expect different similarity patterns in Experiment 3 that reflect the distinct biological/animate nature of these agents, as suggested by the findings of Experiment 1. We developed dissimilarity matrices and performed MDS in Experiment 3 and considered the mean for each agent comparison for a particular age group (Fig. 4 , panel A). The red clusters and red circles in Fig. 4 represent the similarity between animate things (e.g. a human adult and a frog). The red clusters/circles are present in school-aged children and adults but is weak in preschool children. We also conducted the same analyses in Experiment 2 and compared the similarity between AI/robots and God to the average similarity rating between other pairs in the Japanese population. Results revealed that the similarity between AI/robots and God was significantly lower than the similarity of other pairs [F(1, 113) = 7.698, p = .006, η 2 G = 0.041] in Experiment 3, in contrast to Experiment 2 (Fig. 3 B left). No other significant effects were observed. Next, we focused on the red cluster and assessed whether the similarity between animate agents was higher than that of other pairs by comparing the average similarity between animate agents (a human lady, a girl, a baby, a chimpanzee, and a frog) to the average similarity rating between other pairs, except the identical pairs. We then conducted mixed ANOVAs with agent (animate vs. other pairs) and age group (preschool, school, and adult) as independent factors (Fig. 3 B, right). We found a significant main effect of agent, showing that the similarity between animate pairs was higher than that between other pairs [F(1, 113) = 193.732, p < .001, η 2 G = 0.041]. We also found an interaction between agent and age group [F(2, 113) = 10.525, p < .001, η 2 G = 0.041]. Post-hoc analyses using Shaffer’s method revealed that the similarity between animate pairs was higher than that between other pairs in each age group (ps <. 001). However, the effect sizes in adults (Cohen’s d = 1.826) were higher than those in preschool (Cohen’s d = 0.699) and school-aged children (Cohen’s d = 0.998). The results suggest that both children and adults rated the similarity of agents in multiple domains, such as knowledge and animacy, depending on the tasks, and perceived AI/robots and God as similar in terms of knowledge, but not animacy. In sum, the perceived similarity between AI/robot and God is specific to their knowledge characteristics rather than extending to their biological/animate nature. Discussion This study examined how children and adults evaluate different agents in terms of knowledge. Specifically, we addressed how AI is situated in terms of knowledge among existing agents, such as divine entities, and how this position is influenced by developmental and cultural factors. First, we used a more traditional questioning method to examine the issue and found that both Japanese and U.S. children discriminate AI from divine entities and a human in terms of biological and psychological properties, but they attribute the extraordinary property to AI, divine entities, and a person in a relatively similar way. We then used a similarity rating method and found that although the overall structure of knowledge was similar across developmental and cultural groups, some differences were observed. Specifically, Japanese children and adults regarded AI/robots as similar to God, but U.S. children and adults regarded AI/robots as distinct from God. Experiment 3 confirmed that Japanese children and adults rated AI/robot and God as similar in terms of knowledge, but not animacy. This study provides novel insights into how humans represent various agents and their relationships. Previous studies of agent perception showed that participants attributed agency, but not experience, to AI/robots and divine entities similarly across diverse cultural contexts 6 , 10 . However, while our analysis revealed that the overall structure of agent relationships showed remarkable consistency across age groups and cultures, interesting cultural differences emerged, specifically in how AI and divine entities were integrated into these otherwise stable representational structures. In the Japanese context, our analysis demonstrated that participants represented AI with notably similar patterns to their representation of divine entities, particularly in terms of the knowledge attributed. The results suggests that the Japanese participants assimilated AI into the same cognitive cluster as divine entities. This finding aligns with previous research indicating that individuals from East Asian cultures tend to understand robots through mental state attribution without using humans as a reference point, whereas those from Western cultures tend to evaluate robots based on their similarity to humans and maintain distinct categories for different types of agents 31 , 32 . In our study, Japanese participants rated AI and divine entities as similar, even though an adult lady was not necessarily regarded as similar to these agents. In contrast, U.S. participants showed a markedly different pattern, in which AI were represented as a distinct category, significantly different from both human agents and divine entities. Japanese participants may have conceptualised AI as a supernatural-like agent, whereas U.S. participants maintained a more distinct ‘AI as technology’ category boundary. This finding enriches mind perception theories by demonstrating that a non-human, non-biological agent with superhuman capabilities can either be assimilated into the high-agency mind category (approaching God-like status) or kept separate as a distinct technological entity, depending on cultural context. The observed cross-cultural differences in the conceptual integration of AI and divine entities likely reflect broader cultural frameworks that shape how such agents are understood. In Japan, Shinto traditions that imbue non-human entities with spirit, combined with widespread media portrayals of harmonious human–robot relationships (e.g., in animation), may predispose individuals to assimilate AI into spiritually or mentally agentive categories 31 , 32 . In contrast, U.S. cultural models—often grounded in monotheistic religious traditions that position God as a unique, omnipotent being—encourage sharper categorical boundaries between divine and technological agents. Moreover, Western narratives and media promote specific impressions about AI 40 , reinforcing the perception of AI as a powerful yet fundamentally tool-like objects. While this study did not directly assess religiosity, media exposure, or AI familiarity, future research could examine how these factors contribute to cross-cultural variation in the cognitive organisation of emerging agents such as AI. Such inquiry would help clarify whether cultural pathways operate through differential exposure, belief systems, or motivational stances toward technology and divinity. Interestingly, these cultural patterns in AI representation remained relatively stable across age groups in both cultures. The results were inconsistent with previous research showing that children attribute certain mental states to AI (especially conversational agents) and moral consideration, but this tendency declines between ages 4 to 11 years 30 . The inconsistencies may be partly due to differences in the research methods (similarity judgement vs. questioning) and contents (knowledge vs. emotion/moral). Moreover, the youngest participants in our study were approximately 5–6-years-old, whereas the previous studies included children as young as 3–4-years-old. Thus, age-related differences might have become apparent if younger children had been included in our sample. Despite these limitations, the age-independent consistency we observed suggests that cultural frameworks for understanding artificial agents may be established early in development. This finding extends previous research on agent concepts 5 , 39 by demonstrating that the basic framework for representing AI appears to be influenced by cultural context. Our findings also provide unique insights into how individuals understand the nature of knowledge itself 19 , 20 . The Japanese participants’ tendency to represent AI as similar to divine entities suggests that they may view AI knowledge as more aligned with world models or they may not have the distinction between instrumental and worldly knowledge. In contrast, the U.S. participants’ distinct categorisation of AI might reflect a more instrumental view of AI knowledge, viewing it as a powerful but fundamentally different type of knowledge from that possessed by humans or God. This could indicate an understanding of AI as a sophisticated pattern-matching and predictive system rather than as an entity with true understanding or world knowledge. However, this study did not directly examine participants’ understanding of the qualitative nature of AI knowledge. While the representational patterns we observed are suggestive of these differing conceptualisations, future research should explicitly investigate how individuals across cultures understand the nature and quality of AI knowledge. This could involve direct comparisons of how participants characterise the knowledge possessed by different agents; for instance, whether they view AI knowledge as instrumental knowledge or world knowledge. Such research could help clarify whether the cultural differences we observed in AI representation indeed reflect deeper distinctions in how different cultures conceptualise the nature of artificial intelligence and its relationship with other forms of knowledge and understanding. This would advance our theoretical understanding of how humans represent different types of knowledgeable agents and have important implications for how AI systems are developed and integrated into different cultural contexts. Another innovation of this study is the examination of cognitive structures among agents using representational similarity analyses. Traditional methodologies that use direct questions about agent attributes, while valuable, have limitations in revealing complex interconnections and clustering relationships among different types of agents 5 , 26 . Our structural approach allowed us to visualise these relationships more comprehensively and precisely by capturing multi-dimensional patterns that questionnaire-based methods often miss. Specifically, our approach revealed non-linear relationships between different agent categories, identified clustering patterns without imposing predetermined dimensional constraints, and quantified the relative distances between all agent pairs simultaneously, rather than just measuring isolated attribute ratings. Previous methodologies often reduce complex agent perceptions to individual attribute scores, limiting their ability to detect emergent structural patterns or unexpected agent groupings. This methodological advancement builds on our previous work that examined perceptual structures in children, particularly in the domain of colour experience 36 . The successful application of this methodology to conceptual knowledge about agents demonstrates its flexibility and potential for studying various aspects of cognitive development. The present findings suggest several promising avenues for future research, while also highlighting some limitations of the current set of studies. First, regarding agent perception, while our stimulus set included various types of agents, it was not exhaustive. Future studies should employ a broader range of stimuli to provide a more comprehensive understanding of agent representation. Moreover, while we focused on biological attributes and knowledge in this study, other dimensions such as problem-solving abilities should be examined for a more complete picture of how different agent attributes are represented across cultures. Another important limitation is that our cross-cultural comparison was restricted to Japan and the United States. Since concepts of AI and divine entities are heavily influenced by cultural factors, future research should examine these representations across a broader range of cultures. Further, such broader cultural sampling could help determine the aspects of agent representation that are truly universal and those that are more culturally specific. More broadly, our structural approach can be implemented in investigations about the relationships between different cognitive domains that have traditionally been studied separately. By applying the same methodological framework across the perceptual, conceptual, and emotional domains, researchers may reveal novel connections between these areas. Such connections have been examined using behavioural data, brain data, and computational models in cognitive neuroscience research 34 , 41 . These future directions and limitations highlight both the potential of our structural approach and the need for continued research in this area. Materials and Methods Ethics Information The study protocol was approved by the Ethics Committee of the Psychological Science Unit of Kyoto University (No. 5-P-16, approved on 28th September 2023) and the University of New Hampshire (FY2023-8, approved on 18th June 2024). Informed consent was obtained from the adult participants and all the parents of the child participants when they participated in the experiment (from 30th September to 20th May 2024 in Japan and from 1st August to 12th October 2024 in USA). This study was adhered to the ethical standards established by the Helsinki Declaration of 1964 and its subsequent revisions. Participants For participant recruitment for experiments involving Japanese participants, we referred to a university database and collaborated with a research company (Cross Marketing Inc. Tokyo, Japan). For Experiment 1, we recruited 73 Japanese children (mean age in months = 68.30 (SD = 7.13), 37 girls) from nursery schools. For Experiment 2, we recruited 62 Japanese preschool children (mean age in months = 61.12 (SD = 12.77), 30 girls); 53 school-aged children (mean age in months = 113.70 (SD = 18.90), 23 girls); and 40 adults (mean age = 43.75 years (SD = 12.54), 16 women). For Experiment 3, we recruited 21 Japanese preschool children (mean age in months = 71.86 (SD = 10.03), 10 girls); 52 school-aged children (mean age in months = 109.43 (SD = 17.84), 20 girls); and 33 adults (mean age = 28.21 years (SD = 3.55), 12 women). For the U.S. population, we recruited 36 U.S. preschool children (mean age in months = 64.60 (SD = 12.40), 22 girls); 38 school-aged children (mean age in months = 106.42 (SD = 16.07), 19 girls); and 40 religious (mean age = 43.16 years (SD = 17.45), 20 women) and 38 non-religious adults (mean age = 37.43 years (SD = 11.82), 19 women) from a university database and Prolific for Experiment 1 and 2. The age in months between countries were not different in preschool and school-aged children (ps > .05). Additionally, we recruited five Japanese preschool children and eight school-aged children in Experiment 2; three Japanese preschool children, six school-aged children, and seven adults in Experiment 3; and three U.S. preschool children, eight school-aged children, and two adults in Experiment 1 and 2, but their data were excluded from analyses for not meeting the study criteria (explained further below). Preschool children in both Japan and the USA are familiar with the concept of God/gods 26 , 28 . Regarding AI, before the main experiments, we conducted preliminary studies in Japan and the United States to assess children’s understanding of the term ‘AI’. In Japan, children were presented with images of a computer and a tree and asked to indicate which was more closely related to AI. In the United States, children were directly asked whether they knew about AI. Preschool children in both countries performed at chance level (46/73 in Japan, 21/35 in the USA), indicating no cross-cultural differences in their basic comprehension of the AI concept. Although these findings suggest a limited explicit understanding of AI among preschoolers, we proceeded to include this age group in our main study. This decision was guided by our primary research objective: to investigate how young children conceptualise entities such as AI and divine entities and to examine how these conceptualisations differ from those of older children and adults, regardless of their explicit understanding of the terminology. Procedure of Experiment 1 We used pictures of a human (a female adult), a God, and an AI (computer) as agents. We asked questions about their biological and psychological properties based on previous studies 5 , 26 . Moreover, we added questions on characteristics related to omniscience and omnipotence, which we referred to as ‘extraordinary’ questions. For instance, we used the terms ‘breathe’ and ‘hungry’ as biological properties, ‘pleasure’ and ‘sad’ as psychological questions, and ‘omniscience’ and ‘omnipotence’ as extraordinary questions. The experimental procedure was a modified version of the experiment conducted by Jipson and Gelman 28 . Children were introduced to a picture of each agent in random order and then asked a series of test questions. The children were then asked questions about biological properties (‘Can X breathe?’; ‘Can X feel hungry?’) and psychological properties (‘Can X feel pleasure?’; ‘Can X feel sad?’), as well as extraordinary questions (‘Can X know everything?’; ‘Can X solve any difficult problem?’) in a random order. The experiment was conducted in-person. We used the term ‘God’ for U.S. participants and ‘kami-sama‘ (gods) for Japanese participants, adapting to each cultural context. Children’s responses were scored according to the proportion of ‘yes’ responses as a function of the question property (biological, psychological, and extraordinary) and agent type (human, AI, and God). Scores ranged from 0 to 2, with a score of 0 indicating that children did not attribute the property to the agent and a score of 2 indicating that children did attribute the property to the agent. After pre-registration ( https://osf.io/bfgpy/?view_only=6528384d20e24e969d932df57c15dc81 ), we collected the data of U.S. children. The procedures for U.S. children were the same as those of the Japanese children, except that the picture of AI included a computer, a smartphone, and a smart speaker in the U.S. population, whereas only a computer was shown in the Japanese population, based on the assumption that such devices can facilitate U.S. children’s understanding of AI. Procedure of Experiment 2 We used pictures of 10 agents, including a human adult, a God, an AI used in Experiment 1, as well as a human girl, a human baby, a chimpanzee, a frog, a robot, a ghost, and a stone. On any given trial, two stimuli were simultaneously presented. There were 55 unique agent pairs, including identical (e.g. human adult vs. human adult) and non-identical pairs (e.g. human adult vs. God), resulting in 55 trials. We only presented a unique agent pair once. After obtaining their consent, participants were shown a sequence of task instructions. Specifically, they evaluated how similar they thought the various living creatures and artefacts were to each other in terms of how much knowledge they have about many things. Next, they chose one from the following four options: ‘dissimilar’, ‘slightly dissimilar’, ‘slightly similar’, or ‘similar’. To check whether they understood the instructions, participants were given a practice trial, in which they were presented with an identical stimuli pair (human male vs. human male, not presented in the main experiment). When they chose ‘similar’, they proceeded to the main experiment. Otherwise, they were provided feedback and underwent another practice trial. Subsequently, the main experiment was conducted. In each trial, participants were presented with the stimuli pair and asked to report the perceived similarity between the agents in terms of how much knowledge they have about things. The experiment was conducted online. Data from some of the participants were excluded from the main analysis based on certain criteria. Specifically, we assessed the reliability of the participants’ responses. Since participants were required to choose ‘Similar’ responses in the identical pairs (e.g. human adult vs. human adult) in the practice trial, we calculated the percentage of ‘Similar’ responses for the identical pairs in the test trials. We excluded participants who chose ‘Similar’ responses in fewer than 7 out of 10 trials. The participants’ responses were coded on a four-point scale (1 = ‘dissimilar’, 2 = ‘slightly dissimilar’, 3 = ‘slightly similar’, and 4 = ‘similar’). After pre-registration, we collected the data of U.S. children and adults. The procedures for U.S. participants were the same as those of the Japanese participants, except two points. First, as in Experiment 1, the picture of AI included a computer, a smartphone, and a smart speaker in the U.S. population. Second, in the practice phase, participants were presented with pictures of an adult male and a bug and asked to evaluate the similarity between the two. They were required to respond ‘dissimilar’ in this practice trial. Procedure of Experiment 3 The materials and procedure were the same as in Experiment 2, except that participants evaluated how similar they thought various living creatures and artefacts were to each other in terms of animacy. Declarations Funding This study received support through the Grant-in-Aid for Transformative Research Areas (23H04830) from the Japan Society for the Promotion of Science. The funder had no role in the following: study design, data collection, analysis, decision to publish, or preparation of the manuscript. Author Contribution Y.M. developed the study concept. Y.M. and B.R.J. contributed to the study design. Y.M., R.W., and B.R.J. performed data collection. All authors contributed to data analysis and interpretation. Y.M. drafted the manuscript. R.W., Y.H., and B.R.J. revised the manuscript. All authors approved the manuscript for submission. Data Availability All the experimental data are available via (https://doi.org/10.6084/m9.figshare.28814855) References Broadbent E (2017) Interactions with robots: The truths we reveal about ourselves. Annu Rev Psychol 68:627–652 Blut M, Wang C, Wünderlich NV, Brock C (2021) Understanding anthropomorphism in service provision: a meta-analysis of physical robots, chatbots, and other AI. J Acad Mark Sci 49:632–658 Geiselmann R, Tsourgianni A, Deroy O, Harris LT (2023) Interacting with agents without a mind: the case for artificial agents. Curr Opin Behav Sci 51:101282 Andries V, Robertson J (2023) Alexa doesn’t have that many feelings: Children’s children’s understanding of AI through interactions with smart speakers in their homes. Comput Educ Artif Intell 5:100176 Jipson JL, Gelman SA (2007) Robots and rodents: children’s inferences about living and nonliving kinds. Child Dev 78:1675–1688 Gray HM, Gray K, Wegner DM (2007) Dimensions of mind perception. Science 315:619 Awad E et al (2018) The Moral Machine experiment. Nature 563:59–64 Waytz A, Gray K, Epley N, Wegner DM (2010) Causes and consequences of mind perception. Trends Cogn Sci 14:383–388 Weisman K, Dweck CS, Markman EM (2017) Rethinking people’s conceptions of mental life. Proc. Natl. Acad. Sci. U. S. A. 114, 11374–11379 Takahashi H, Ban M, Asada M (2016) Semantic differential scale method can reveal multi-dimensional aspects of mind perception. Front Psychol 7:1717 Chemero A (2023) LLMs differ from human cognition because they are not embodied. Nat Hum Behav 7:1828–1829 Kocoń J et al (2023) ChatGPT: Jack of all trades, master of none. Inf Fusion 99:101861 Hoffmann J et al (2022) Training Compute-Optimal Large Language Models. arXiv [cs.CL] Garry M, Chan WM, Foster J, Henkel LA (2024) Large language models (LLMs) and the institutionalization of misinformation. Trends Cogn Sci 28:1078–1088 Liu J (2024) ChatGPT: perspectives from human-computer interaction and psychology. Front Artif Intell 7:1418869 Kosinski M (2024) Evaluating large language models in theory of mind tasks. Proc. Natl. Acad. Sci. U. S. A. 121, e2405460121 Collins KM et al (2024) Evaluating language models for mathematics through interactions. Proc. Natl. Acad. Sci. U. S. A. 121, e2318124121 Strachan JWA et al (2024) Testing theory of mind in large language models and humans. Nat Hum Behav 8:1285–1295 Yildirim I, Paul LA (2024) From task structures to world models: What do LLMs know? Trends Cogn Sci 28:P404–415 Mahowald K et al (2024) Dissociating language and thought in large language models. Trends Cogn Sci 28:517–540 Goddu MK, Noë A, Thompson E (2024) LLMs don’t know anything: reply to Yildirim and Paul. Trends Cogn Sci 28:963–964 Barlev M, Mermelstein S, German TC (2018) Representational coexistence in the God concept: Core knowledge intuitions of God as a person are not revised by Christian theology despite lifelong experience. Psychon Bull Rev 25:2330–2338 Purzycki BG (2013) The minds of gods: a comparative study of supernatural agency. Cognition 129:163–179 Festerling J, Siraj I (2020) Alexa, what are you? Exploring primary school children’s ontological perceptions of digital voice assistants in open interactions. Hum Dev 64:26–43 Knight N, Sousa P, Barrett JL, Atran S (2004) Children’s attributions of beliefs to humans and God: cross-cultural evidence. Cogn Sci 28:117–126 Moriguchi Y, Takahashi H, Nakamata T, Todo N (2019) Mind perception of God in Japanese children. Int J Psychol 54:557–562 Barrett JL, Richert RA, Driesenga A (2001) God’s beliefs versus mother’s: the development of nonhuman agent concepts. Child Dev 72:50–65 Lane JD, Wellman HM, Evans EM (2010) Children’s understanding of ordinary and extraordinary minds: Understanding extraordinary minds. Child Dev 81:1475–1489 Heiphetz L, Lane JD, Waytz A, Young LL (2016) How children and adults represent God’s mind. Cogn Sci 40:121–144 Flanagan T, Wong G, Kushnir T (2023) The minds of machines: Children’s beliefs about the experiences, thoughts, and morals of familiar interactive technologies. Dev Psychol 59:1017–1031 Spatola N, Marchesi S, Wykowska A (2022) Different models of anthropomorphism across cultures and ontological limits in current frameworks the integrative framework of anthropomorphism. Front Robot AI 9:863319 Ikari S et al (2023) Religion-related values differently influence moral attitude for robots in the United States and Japan. J Cross Cult Psychol 54:742–759 Kou G, Zhang S (2024) The influence of culture in shaping anthropomorphic attitudes towards robots: A literature review. in HCI International 2024 Posters 357–371Springer Nature Switzerland, Cham Kriegeskorte N, Mur M, Bandettini P (2008) Representational similarity analysis - connecting the branches of systems neuroscience. Front Syst Neurosci 2:4 Zeleznikow-Johnston A, Aizawa Y, Yamada M (2023) & Tsuchiya, N. Are Color Experiences the Same across the Visual Field? J Cogn Moriguchi Y et al Comparing color qualia structures through a novel similarity task in young children versus adults. Proceedings of the National Academy of Sciences, 122(11), e2415346122. Hatano G, Inagaki K (1994) Young children’s naive theory of biology. Cognition 50:171–188 Wellman HM, Cross D, Watson J (2001) Meta-analysis of theory-of-mind development: the truth about false belief. Child Dev 72:655–684 Gelman SA, Legare CH (2011) Concepts and folk theories. Annu Rev Anthropol 40:379–398 Mara M, Appel M (2015) Science fiction reduces the eeriness of android robots: A field experiment. Comput Hum Behav 48:156–162 Spriet C, Abassi E, Hochmann J-R, Papeo L (2022) Visual object categorization in infancy. Proc. Natl. Acad. Sci. U. S. A. 119 Footnotes We used the term ‘God’ for US participants and ‘kami-sama’ (gods) for Japanese participants, adapting to each cultural context. However, in the results section, we consistently use ‘God’ to avoid redundancy while presenting findings. Additional Declarations No competing interests reported. Supplementary Files Supply.docx Cite Share Download PDF Status: Under Review Version 1 posted Editorial decision: Revision requested 06 Nov, 2025 Reviews received at journal 28 Oct, 2025 Reviews received at journal 20 Oct, 2025 Reviewers agreed at journal 20 Oct, 2025 Reviewers agreed at journal 13 Oct, 2025 Reviewers agreed at journal 05 Oct, 2025 Reviewers invited by journal 26 Sep, 2025 Editor assigned by journal 26 Sep, 2025 Editor invited by journal 15 Jul, 2025 Submission checks completed at journal 09 Jul, 2025 First submitted to journal 09 Jul, 2025 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-6824602","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":525335570,"identity":"de7dbd68-3fbf-4c70-98a1-6761f5c1475c","order_by":0,"name":"Yusuke Moriguchi","email":"data:image/png;base64,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","orcid":"","institution":"Kyoto University","correspondingAuthor":true,"prefix":"","firstName":"Yusuke","middleName":"","lastName":"Moriguchi","suffix":""},{"id":525335571,"identity":"b30d21d8-7f47-4aa4-9755-01557ed5458d","order_by":1,"name":"Ryoichi Watanabe","email":"","orcid":"","institution":"Kyoto University","correspondingAuthor":false,"prefix":"","firstName":"Ryoichi","middleName":"","lastName":"Watanabe","suffix":""},{"id":525335572,"identity":"670fd2b6-b67f-47e9-91c4-8912b1edbd29","order_by":2,"name":"Yusuke Hori","email":"","orcid":"","institution":"Kyoto University","correspondingAuthor":false,"prefix":"","firstName":"Yusuke","middleName":"","lastName":"Hori","suffix":""},{"id":525335573,"identity":"3c448e08-613d-4903-8db3-a7b3f9bddf32","order_by":3,"name":"Bolivar Reyes-Jaquez","email":"","orcid":"","institution":"University of New Hampshire","correspondingAuthor":false,"prefix":"","firstName":"Bolivar","middleName":"","lastName":"Reyes-Jaquez","suffix":""}],"badges":[],"createdAt":"2025-06-05 03:08:04","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6824602/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6824602/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":93242949,"identity":"79d18d7b-a99e-4ecd-8288-7229a2c8b097","added_by":"auto","created_at":"2025-10-10 14:55:24","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":934906,"visible":true,"origin":"","legend":"","description":"","filename":"agentexperimentv14.docx","url":"https://assets-eu.researchsquare.com/files/rs-6824602/v1/75631127dbc4a9cc654ad2fb.docx"},{"id":93244273,"identity":"dabe0229-06e3-49ae-a61d-21d3a180eeba","added_by":"auto","created_at":"2025-10-10 15:03:24","extension":"json","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":5565,"visible":true,"origin":"","legend":"","description":"","filename":"13b792a7b46448899ac519e68b0fbdfc.json","url":"https://assets-eu.researchsquare.com/files/rs-6824602/v1/94a8412eb07a4530878958a8.json"},{"id":93242956,"identity":"237a8955-256b-4e8b-b90c-587cf032aba8","added_by":"auto","created_at":"2025-10-10 14:55:24","extension":"docx","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":211592,"visible":true,"origin":"","legend":"","description":"","filename":"Supply.docx","url":"https://assets-eu.researchsquare.com/files/rs-6824602/v1/66d863b3fb96c26abf4b266e.docx"},{"id":93240751,"identity":"e7b94b0a-a7a7-4bcd-9223-f90d80f3f37e","added_by":"auto","created_at":"2025-10-10 14:47:24","extension":"xml","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":96068,"visible":true,"origin":"","legend":"","description":"","filename":"13b792a7b46448899ac519e68b0fbdfc1enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-6824602/v1/80a63242d087fd010336a34e.xml"},{"id":93240746,"identity":"61b2af29-3986-445f-9297-98e007c881fb","added_by":"auto","created_at":"2025-10-10 14:47:24","extension":"png","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":76880,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-6824602/v1/62f435925526620aac9ae439.png"},{"id":93242953,"identity":"d6b13930-76ae-412e-92ba-aba3dc3ed1b2","added_by":"auto","created_at":"2025-10-10 14:55:24","extension":"png","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":301994,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-6824602/v1/ed50c37b2649cfd699f03ffe.png"},{"id":93244272,"identity":"8404e39a-7aec-4be7-8678-dcf51bc55ace","added_by":"auto","created_at":"2025-10-10 15:03:24","extension":"png","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":157745,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-6824602/v1/127d830a502a5b0007be2193.png"},{"id":93242957,"identity":"7242e876-1148-4d97-b4c4-3a4f84cc787f","added_by":"auto","created_at":"2025-10-10 14:55:24","extension":"png","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":245485,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-6824602/v1/017d89a757e3930765ac72c6.png"},{"id":93240762,"identity":"ca5a87a7-213f-4c7b-b374-5813f659c7a0","added_by":"auto","created_at":"2025-10-10 14:47:24","extension":"png","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":21289,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-6824602/v1/7b2a57db77e0edc184654029.png"},{"id":93240753,"identity":"57ddb89c-21b4-47e9-bcdb-aa0144591bad","added_by":"auto","created_at":"2025-10-10 14:47:24","extension":"png","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":60963,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-6824602/v1/e3c35c05ba4b30ef7bbcf7f6.png"},{"id":93240757,"identity":"1155378e-90f8-4da2-bcbe-03c9c346ce2a","added_by":"auto","created_at":"2025-10-10 14:47:24","extension":"png","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":47019,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-6824602/v1/6c82fa4351027b4c323302fa.png"},{"id":93240760,"identity":"537a5c55-95f1-4728-bc17-b4679d1777cc","added_by":"auto","created_at":"2025-10-10 14:47:24","extension":"png","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":47805,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-6824602/v1/299c93cbc577b628b1f39beb.png"},{"id":93242959,"identity":"c8ea095c-a972-4488-98f7-fb6ab332885c","added_by":"auto","created_at":"2025-10-10 14:55:24","extension":"xml","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":92881,"visible":true,"origin":"","legend":"","description":"","filename":"13b792a7b46448899ac519e68b0fbdfc1structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-6824602/v1/fc0a31db1c19c794e965f3cd.xml"},{"id":93240761,"identity":"931903dd-8967-420c-bed7-107346e8e2a2","added_by":"auto","created_at":"2025-10-10 14:47:24","extension":"html","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":101682,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-6824602/v1/8fe2a6ad7c775bffee1de09a.html"},{"id":93240744,"identity":"2bcbec4a-a2d1-423a-b311-27761facdeba","added_by":"auto","created_at":"2025-10-10 14:47:24","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":53837,"visible":true,"origin":"","legend":"\u003cp\u003eMean scores of attributions of biological (Bio), psychological (Psy), and extraordinary (Extra) properties for AI, God, and a person in Japan (Left) and the USA (Right) in Experiment 1.\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-6824602/v1/e7b252dbf7fa7209ee55c968.png"},{"id":93240749,"identity":"3f328995-89da-4937-b4df-a7ae87734794","added_by":"auto","created_at":"2025-10-10 14:47:24","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":226066,"visible":true,"origin":"","legend":"\u003cp\u003eGroup mean dissimilarity matrices of knowledge for each age group in Japan (A) and the USA (B). Dark and white cells represent high similarity and dissimilarity, respectively. Blue squares represent the similarity between AI, robot and God. Two-dimensional MDS representation of knowledge in each age group in Japan (C) and the USA (D). Blue circles represent the similarity between AI, robot and God\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-6824602/v1/388855333a1edc4f926d2bd2.png"},{"id":93242950,"identity":"53d698ce-b823-411a-96cc-55ee49083113","added_by":"auto","created_at":"2025-10-10 14:55:24","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":110726,"visible":true,"origin":"","legend":"\u003cp\u003e(A) Mean similarity scores for comparison between similarity ratings of AI-God pairs and other pairs in Japan and the USA as a function of age group in Experiment 2. (B) Mean similarity scores for comparison between similarity ratings of AI-God pairs and other pairs (left) and animate pairs and other pairs (right) in Japan as a function of age group in Experiment 3.\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-6824602/v1/451d56c29baae79bc590063a.png"},{"id":93242951,"identity":"6c25ff8f-66fb-4313-b928-04c2ee8910c5","added_by":"auto","created_at":"2025-10-10 14:55:24","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":142238,"visible":true,"origin":"","legend":"\u003cp\u003e(A) Group mean dissimilarity matrices of animacy for each age group in Japan. Dark and white cells represent high similarity and dissimilarity, respectively. Red squares represent the similarity between animate agents. (B) Two-dimensional MDS representation of animacy for each age group in Japan. Red circles represent the similarity between animate agents.\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-6824602/v1/3719c1e51ac4ee1b0a05d4a9.png"},{"id":93246009,"identity":"900abfe2-ff1b-4b13-be6c-a307d0065806","added_by":"auto","created_at":"2025-10-10 15:11:25","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":968445,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6824602/v1/120185ea-75b0-47ea-99f0-0bdba59fe65c.pdf"},{"id":93242954,"identity":"6d7e1909-46a4-44cf-92a1-5cf520f82e65","added_by":"auto","created_at":"2025-10-10 14:55:24","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":211592,"visible":true,"origin":"","legend":"","description":"","filename":"Supply.docx","url":"https://assets-eu.researchsquare.com/files/rs-6824602/v1/7bc74e56ebee54f23b57e133.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Knowledge-based representations of artificial intelligence and divine agents: A developmental study across Japan and the United States","fulltext":[{"header":"Introduction","content":"\u003cp\u003eUnderstanding how humans perceive and interact with artificial intelligence (AI) and robots has become an important area of research \u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e,\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e. As AI and robots increasingly become part of our daily lives, it is essential to explore how humans conceptualise these advanced technologies. Research demonstrates that both adults and children attribute mental states\u0026mdash;including thoughts, goals, and intentions\u0026mdash;to AI systems and robots \u003csup\u003e\u003cspan additionalcitationids=\"CR4 CR5 CR6\" citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eThe literature on agent perception typically examines two primary dimensions: experience (capacity for sensations and emotions) and agency (ability for planning, thinking, and self-control) \u003csup\u003e\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e,\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e,\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u003c/sup\u003e. Previous studies have investigated how individuals attribute experience and agency to AI, robots, humans, animals, and divine entities, and shown that AI/robots and divine entities are attributed agency but perceived as lacking experience\u0026mdash;a pattern observed across diverse cultural contexts \u003csup\u003e\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e,\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e. However, the emergence of generative AI technologies such as ChatGPT and Gemini may challenge these traditional patterns of agent perception \u003csup\u003e\u003cspan additionalcitationids=\"CR12\" citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e. The ability of these AI systems with superhuman capabilities, such as super-knowledge, could blur or extend the established boundaries between humans, divine entities, and artificial minds, prompting a fundamental reconsideration of how we attribute mental states and capabilities to artificial agents \u003csup\u003e\u003cspan additionalcitationids=\"CR15 CR16 CR17\" citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eTheorists have proposed that there are fundamental differences between the knowledge of AI and humans \u003csup\u003e\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e,\u003cspan additionalcitationids=\"CR20\" citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e\u003c/sup\u003e. While LLMs acquire knowledge through next-word prediction as an instrument, humans develop structured world models that preserve representations of physical reality. In religious traditions, God is understood to possess omniscience\u0026mdash;complete and perfect knowledge that surpasses human comprehension \u003csup\u003e\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e,\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e\u003c/sup\u003e. God\u0026rsquo;s knowledge can be conceptualised as a form of worldly knowledge\u0026mdash;a complete and perfect understanding of the world\u0026rsquo;s true nature and causal relationships. This distinction between instrumental and worldly knowledge provides a novel framework for understanding how children and adults represent different types of agents.\u003c/p\u003e\u003cp\u003eDifferences in knowledge between agents raises important developmental and cultural questions. Research suggests that younger children from Western and non-Western societies initially anthropomorphise both AI and divine entities, attributing human-like knowledge limitations to them \u003csup\u003e\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e,\u003cspan additionalcitationids=\"CR25\" citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e\u003c/sup\u003e. Developmental progression reveals increasing differentiation: divine entities become viewed as possessing complete worldly knowledge, whereas AI systems are perceived as tools having extensive knowledge \u003csup\u003e\u003cspan additionalcitationids=\"CR28 CR29\" citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e\u003c/sup\u003e. However, the developmental trajectory of the conceptual relationships between AI and divine entities remains uncharacterised. Additionally, cultural factors can influence mental capacity attributions to non-human agents. Cross-cultural investigations indicate that while the basic agent structure is similar across both Western (e.g. USA) and non-Western (e.g. Japan) populations \u003csup\u003e\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e,\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e, attitudes toward and anthropomorphic tendencies regarding AI can vary depending on cultural context \u003csup\u003e\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e,\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e\u003c/sup\u003e. Moreover, cultures with animistic or polytheistic traditions (such as Japan) may foster greater acceptance of human-like machines, whereas Western cultural frameworks might engender more cautious attitudes toward artificial agents \u003csup\u003e\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e\u003c/sup\u003e. These cultural differences could influence how participants conceptualise entities with different types of knowledge capabilities, potentially affecting the perceived relationship between AI systems and divine entities.\u003c/p\u003e\u003cp\u003eThe present investigation examines whether conceptualisations of knowledge in AI and divine entities vary between Japanese and U.S. populations, and, if so, when such cultural differences emerge in human development. By focusing specifically on knowledge attribution\u0026mdash;rather than exclusively on experience or agency\u0026mdash;we address a critical aspect of AI and divine entity comparisons. To investigate these issues, we employed two distinct but complementary approaches. First, we conducted a comprehensive evaluation using traditional questioning methods to compare how children perceive divine entities, AI, and humans across multiple dimensions, including biological attributes (e.g. \u0026lsquo;Can X breathe?\u0026rsquo;), psychological capabilities (e.g. \u0026lsquo;Can X feel sad?\u0026rsquo;), and extraordinary capabilities (e.g. \u0026lsquo;Can X know everything?\u0026rsquo;) (Experiments 1) \u003csup\u003e\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u003c/sup\u003e. This holistic assessment provided insights into how these agents are conceptualised across different domains. Building on these findings, our second approach specifically focused on the dimension of knowledge, employing a structural method using representational similarity analyses, a technique that has gained prominence in cognitive neuroscience research, particularly in studies of perception and categorisation (Experiments 2 and 3) \u003csup\u003e\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e,\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e\u003c/sup\u003e. This structural approach built on our previous work, in which we developed a similarity rating task suitable for young children \u003csup\u003e\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eIn this study, we applied this methodology to analyse the patterns of similarity ratings among diverse entities, including humans, AI, and divine entities. The similarity scores were then compiled into a similarity matrix, from which we visualised the structure using dimensional reduction techniques, such as multi-dimensional scaling (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). Considering a diverse range of entities in our similarity ratings helped us gain a structural understanding of how children categorise and relate different types of agents. From developmental perspectives, we can observe how children\u0026rsquo;s initial representations of AI and divine entities develop over time, whether their early tendency to cluster these entities with humans shifts with age, and how their conceptual framework for understanding different types of agents becomes more sophisticated. While such representational structures have been examined in adults \u003csup\u003e\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u003c/sup\u003e, our study uniquely captures how these conceptual organisations emerge and transform throughout development. We first examined these aspects in the Japanese context and then conducted the same experiments in the U.S. context to assess whether the results in Japan could be generalised.\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003eData were collected in Japan first, followed by the U.S. However, to further improve the paper\u0026rsquo;s readability, findings are presented based on subtopics, rather than chronology.\u003c/p\u003e\u003cp\u003e\u003cb\u003eAttribution of Biological, Psychological, and Extraordinary Properties\u003c/b\u003e\u003c/p\u003e\u003cp\u003eIn Experiment 1, we recruited 73 Japanese preschool children and examined whether they similarly attributed biological (hungry, breathing), psychological (happy, sad), and extraordinary properties (super-knowledge, super-problem solving) to a human adult, AI, and God\u003csup\u003e[a]\u003c/sup\u003e, using a traditional questioning method \u003csup\u003e\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u003c/sup\u003e. Moreover, we preregistered this experiment and collected data from 36 U.S. preschool children. For example, children were asked questions such as, \u0026lsquo;Can a person solve any difficult problem?\u0026rsquo; and \u0026lsquo;Can a person know everything?\u0026rsquo; for extraordinary properties, and responded \u0026ldquo;yes\u0026rdquo; or \u0026ldquo;no\u0026rdquo; for each question. As shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e, both Japanese and U.S. children differentiated each agent in terms of biological and psychological properties, but equally attributed extraordinary properties to each agent. We analysed whether children\u0026rsquo;s \u0026lsquo;yes\u0026rsquo; responses differed as a function of the agent (A person, AI, and God); property (biological, psychological, and extraordinary); and culture (Japan vs. U.S.) using a mixed analysis of variance (ANOVA). We found a significant interaction between agent and property [F(4, 428)\u0026thinsp;=\u0026thinsp;26.895, p\u0026thinsp;\u0026lt;\u0026thinsp;.001, η\u003csup\u003e2\u003c/sup\u003e\u003csub\u003eG\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;0.065]. Post-hoc comparison using Shaffer\u0026rsquo;s methods revealed significant differences between agents in each property (ps\u0026thinsp;\u0026lt;\u0026thinsp;.025), but the effect size was smaller for extraordinary properties (η\u003csup\u003e2\u003c/sup\u003e\u003csub\u003eG\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;0.019) compared to biological (η\u003csup\u003e2\u003c/sup\u003e\u003csub\u003eG\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;0.288) and psychological properties (η\u003csup\u003e2\u003c/sup\u003e\u003csub\u003eG\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;0.249). Specifically, across cultures children attributed extraordinary properties to God more than AI (p\u0026thinsp;=\u0026thinsp;.019), but no other differences were found (ps\u0026thinsp;\u0026gt;\u0026thinsp;.088). No significant interactions between agent, property, and culture were observed [F (4, 428)\u0026thinsp;=\u0026thinsp;0.299, p\u0026thinsp;\u0026lt;\u0026thinsp;.001, η\u003csup\u003e2\u003c/sup\u003e\u003csub\u003eG\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;0.000]. Thus, both Japanese and U.S. children attributed the extraordinary properties to each agent in a relatively similar manner, compared to the biological and psychological properties. However, the small differences between agents in extraordinary properties do not mean that the children regarded a person, AI, and God as similar in terms of knowledge. Consequently, more direct comparisons are conducted in subsequent experiments.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003cb\u003eSimilarity Structure of Agents Based on Knowledge\u003c/b\u003e\u003c/p\u003e\u003cp\u003eNext, we directly examined whether children and adults believe that AI and God are similar in terms of knowledge using the similarity rating method. In Experiment 2, participants reported the subjective similarity between different pairs of agents. To fully understand the cognitive structure underlying children\u0026lsquo;s agent perception, we extended our analysis beyond just the target items (a human lady, AI, and God) to include non-target items, such as a human girl, a robot, and a frog, based on a previous study \u003csup\u003e\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u003c/sup\u003e (for each image, Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). Two agents (e.g. human lady and God) were presented simultaneously each trial. Participants evaluated how similar they thought various pairs of living creatures and artefacts were to each other, in terms of how much knowledge they have about things. Next, they chose one from the following four options: \u0026lsquo;Dissimilar\u0026rsquo;, \u0026lsquo;Slightly dissimilar\u0026rsquo;, \u0026lsquo;Slightly similar\u0026rsquo;, or \u0026lsquo;Similar\u0026rsquo;. In Japan, we recruited 61 preschool and 54 school-aged Japanese children and 40 adults, and administered a similarity task to investigate how participants evaluated the relationship between agents. We also recruited 32 preschool and 38 school-aged U.S. children and 80 U.S. adults (40 religious and 38 non-religious) after the pre-registration.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eTo visualise participants\u0026rsquo; responses, we first present the dissimilarity matrices in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e (panels A, B), while considering the mean for each agent comparison for a particular age group in each culture. White boxes represent the dissimilarity between agents, whereas black boxes represent the similarity across agents. For visual inspection, we focus on the blue cluster in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eA (blue square in the Japanese category). The blue cluster represents the similarity between AI, robot and God. The similarities between AI and robot were generally high, which is not surprising. Importantly, similarities between AI/robot and God differed across cultures. That is, the Japanese participants rated the AI/robot and God as similar compared to other pairs, whereas such patterns were not observed in the U.S. population. To obtain a better qualitative understanding, we visualised the geometric relationships between agents using two-dimensional multi-dimensional scaling (MDS), which maintained the relational structures of agent dissimilarities (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, panels C, D). In MDS, the nearer the agents were, the more similar. There was a striking similarity between AI/robot and God in Japanese children and adults (blue circles in the Japanese category), but not in U.S. participants. The position of God clearly differed across age and cultural factors.\u003c/p\u003e\u003cp\u003eNext, we examined whether the overall structure of the agents was similar across the developmental and cultural groups. Group-level dissimilarity matrices obtained from children and adults within/between the cultural groups appeared very comparable. Indeed, all the correlation coefficients between all pairs were \u0026gt;\u0026thinsp;0.72 (Spearman\u0026lsquo;s ρ). The highest correlation was observed between Japanese adults and Japanese school children (ρ\u0026thinsp;=\u0026thinsp;.92), while the lowest correlation coefficient emerged between Japanese adults and U.S. preschool children (ρ\u0026thinsp;=\u0026thinsp;.72). Within each culture, all the correlation coefficients between any pair of age groups were \u0026gt;\u0026thinsp;.89 in Japan and \u0026gt;\u0026thinsp;.81 in the USA. These results indicate that the overall structure of agents was similar across the groups.\u003c/p\u003e\u003cp\u003eHowever, visual inspection suggests that U.S. participants represented AI/robots differently from God, whereas Japanese participants regarded AI/robots and God as similar (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). To quantify this result, we specifically focused on the similarity between AI/robots and God and examined whether the average AI-God and robot-God similarity scores (not including AI-robot similarity due to the high similarity) were distinctly higher than the other similarity scores. Specifically, we averaged the similarity scores of all pairs except the AI-God, robot-God, AI-robot, and the identical pairs used as controls (e.g. robot-robot), which were regarded as similarity scores of other pairs. We conducted mixed ANOVAs with the agent (AI/robot-God vs. other pairs); age group (preschool, school, and adult); and culture (Japan vs. the U.S.) as predictors of similarity scores (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e, panel A). We found a significant interaction between agent and culture [F(1, 297)\u0026thinsp;=\u0026thinsp;48.013, p\u0026thinsp;\u0026lt;\u0026thinsp;.001, η\u003csup\u003e2\u003c/sup\u003e\u003csub\u003eG\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;0.041]. Post-hoc analysis using Shaffer\u0026rsquo;s method revealed that the similarity between AI/robot and God was higher than that for other pairs in the Japanese population, but the opposite pattern was observed in the U.S. population (ps\u0026thinsp;\u0026lt;\u0026thinsp;.05). The interaction between the three factors was not significant [F(2, 297)\u0026thinsp;=\u0026thinsp;0.081, p\u0026thinsp;=\u0026thinsp;.921, η\u003csup\u003e2\u003c/sup\u003e\u003csub\u003eG\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;0.000]. The results suggest that Japanese children and adults evaluated AI/robots and God as more \u003cem\u003esimilar\u003c/em\u003e than any other pairs, while U.S. children and adults evaluated AI/robots and God as more \u003cem\u003edissimilar\u003c/em\u003e than any other pairs. The analysis of age and cultural effects on each pair is reported in Supplemental analysis and Figure \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e.\u003c/p\u003e\u003cp\u003e\u003cb\u003eSimilarity Matrices of Animacy\u003c/b\u003e\u003c/p\u003e\u003cp\u003eThe analyses above demonstrate that Japanese participants, but not U.S. participants, perceived similarities between AI/robots and God in terms of knowledge. Building on these findings and the biological attribution results from Experiment 1, we conducted Experiment 3 to examine whether this perceived similarity between AI/robots and God is specific to these agents\u0026rsquo; knowledge characteristics, or extends to their biological/animate nature.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eAs shown in Experiment 1 (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e), participants attributed different biological characteristics to AI and God. This suggests that while these agents might share knowledge qualities, they may be perceived differently in terms of their animate nature. Indeed, it is well known that children have a naive biology theory (e.g. animate/inanimate) distinct from a naive psychology theory (e.g. knowledge of things) \u003csup\u003e\u003cspan additionalcitationids=\"CR38\" citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e\u003c/sup\u003e. To directly test this hypothesis, we conducted a similarity judgement task that focused specifically on animacy in Experiment 3. Since US participants did not perceive similarities between AI/robots and God in Experiment 2, we focused solely on Japanese participants in Experiment 3. We recruited 23 preschool children, 55 school-aged children, and 40 adults and administered the same similarity task using the same stimuli as in Experiment 2. However, this time, participants were explicitly asked to judge similarities based on animacy. If the similarity between AI/robots and God observed in Experiment 2 is indeed specific to knowledge, we would expect different similarity patterns in Experiment 3 that reflect the distinct biological/animate nature of these agents, as suggested by the findings of Experiment 1.\u003c/p\u003e\u003cp\u003eWe developed dissimilarity matrices and performed MDS in Experiment 3 and considered the mean for each agent comparison for a particular age group (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e, panel A). The red clusters and red circles in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e represent the similarity between animate things (e.g. a human adult and a frog). The red clusters/circles are present in school-aged children and adults but is weak in preschool children.\u003c/p\u003e\u003cp\u003eWe also conducted the same analyses in Experiment 2 and compared the similarity between AI/robots and God to the average similarity rating between other pairs in the Japanese population. Results revealed that the similarity between AI/robots and God was significantly lower than the similarity of other pairs [F(1, 113)\u0026thinsp;=\u0026thinsp;7.698, p\u0026thinsp;=\u0026thinsp;.006, η\u003csup\u003e2\u003c/sup\u003e\u003csub\u003eG\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;0.041] in Experiment 3, in contrast to Experiment 2 (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eB left). No other significant effects were observed.\u003c/p\u003e\u003cp\u003eNext, we focused on the red cluster and assessed whether the similarity between animate agents was higher than that of other pairs by comparing the average similarity between animate agents (a human lady, a girl, a baby, a chimpanzee, and a frog) to the average similarity rating between other pairs, except the identical pairs. We then conducted mixed ANOVAs with agent (animate vs. other pairs) and age group (preschool, school, and adult) as independent factors (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eB, right). We found a significant main effect of agent, showing that the similarity between animate pairs was higher than that between other pairs [F(1, 113)\u0026thinsp;=\u0026thinsp;193.732, p\u0026thinsp;\u0026lt;\u0026thinsp;.001, η\u003csup\u003e2\u003c/sup\u003e\u003csub\u003eG\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;0.041]. We also found an interaction between agent and age group [F(2, 113)\u0026thinsp;=\u0026thinsp;10.525, p\u0026thinsp;\u0026lt;\u0026thinsp;.001, η\u003csup\u003e2\u003c/sup\u003e\u003csub\u003eG\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;0.041]. Post-hoc analyses using Shaffer\u0026rsquo;s method revealed that the similarity between animate pairs was higher than that between other pairs in each age group (ps \u0026lt;. 001). However, the effect sizes in adults (Cohen\u0026rsquo;s d\u0026thinsp;=\u0026thinsp;1.826) were higher than those in preschool (Cohen\u0026rsquo;s d\u0026thinsp;=\u0026thinsp;0.699) and school-aged children (Cohen\u0026rsquo;s d\u0026thinsp;=\u0026thinsp;0.998). The results suggest that both children and adults rated the similarity of agents in multiple domains, such as knowledge and animacy, depending on the tasks, and perceived AI/robots and God as similar in terms of knowledge, but not animacy. In sum, the perceived similarity between AI/robot and God is specific to their knowledge characteristics rather than extending to their biological/animate nature.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eThis study examined how children and adults evaluate different agents in terms of knowledge. Specifically, we addressed how AI is situated in terms of knowledge among existing agents, such as divine entities, and how this position is influenced by developmental and cultural factors. First, we used a more traditional questioning method to examine the issue and found that both Japanese and U.S. children discriminate AI from divine entities and a human in terms of biological and psychological properties, but they attribute the extraordinary property to AI, divine entities, and a person in a relatively similar way. We then used a similarity rating method and found that although the overall structure of knowledge was similar across developmental and cultural groups, some differences were observed. Specifically, Japanese children and adults regarded AI/robots as similar to God, but U.S. children and adults regarded AI/robots as distinct from God. Experiment 3 confirmed that Japanese children and adults rated AI/robot and God as similar in terms of knowledge, but not animacy.\u003c/p\u003e\u003cp\u003eThis study provides novel insights into how humans represent various agents and their relationships. Previous studies of agent perception showed that participants attributed agency, but not experience, to AI/robots and divine entities similarly across diverse cultural contexts \u003csup\u003e\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e,\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e. However, while our analysis revealed that the overall structure of agent relationships showed remarkable consistency across age groups and cultures, interesting cultural differences emerged, specifically in how AI and divine entities were integrated into these otherwise stable representational structures. In the Japanese context, our analysis demonstrated that participants represented AI with notably similar patterns to their representation of divine entities, particularly in terms of the knowledge attributed. The results suggests that the Japanese participants assimilated AI into the same cognitive cluster as divine entities. This finding aligns with previous research indicating that individuals from East Asian cultures tend to understand robots through mental state attribution without using humans as a reference point, whereas those from Western cultures tend to evaluate robots based on their similarity to humans and maintain distinct categories for different types of agents \u003csup\u003e\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e,\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e\u003c/sup\u003e. In our study, Japanese participants rated AI and divine entities as similar, even though an adult lady was not necessarily regarded as similar to these agents. In contrast, U.S. participants showed a markedly different pattern, in which AI were represented as a distinct category, significantly different from both human agents and divine entities. Japanese participants may have conceptualised AI as a supernatural-like agent, whereas U.S. participants maintained a more distinct \u0026lsquo;AI as technology\u0026rsquo; category boundary. This finding enriches mind perception theories by demonstrating that a non-human, non-biological agent with superhuman capabilities can either be assimilated into the high-agency mind category (approaching God-like status) or kept separate as a distinct technological entity, depending on cultural context.\u003c/p\u003e\u003cp\u003eThe observed cross-cultural differences in the conceptual integration of AI and divine entities likely reflect broader cultural frameworks that shape how such agents are understood. In Japan, Shinto traditions that imbue non-human entities with spirit, combined with widespread media portrayals of harmonious human\u0026ndash;robot relationships (e.g., in animation), may predispose individuals to assimilate AI into spiritually or mentally agentive categories \u003csup\u003e\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e,\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e\u003c/sup\u003e. In contrast, U.S. cultural models\u0026mdash;often grounded in monotheistic religious traditions that position God as a unique, omnipotent being\u0026mdash;encourage sharper categorical boundaries between divine and technological agents. Moreover, Western narratives and media promote specific impressions about AI \u003csup\u003e\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e\u003c/sup\u003e, reinforcing the perception of AI as a powerful yet fundamentally tool-like objects. While this study did not directly assess religiosity, media exposure, or AI familiarity, future research could examine how these factors contribute to cross-cultural variation in the cognitive organisation of emerging agents such as AI. Such inquiry would help clarify whether cultural pathways operate through differential exposure, belief systems, or motivational stances toward technology and divinity.\u003c/p\u003e\u003cp\u003eInterestingly, these cultural patterns in AI representation remained relatively stable across age groups in both cultures. The results were inconsistent with previous research showing that children attribute certain mental states to AI (especially conversational agents) and moral consideration, but this tendency declines between ages 4 to 11 years \u003csup\u003e\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e\u003c/sup\u003e. The inconsistencies may be partly due to differences in the research methods (similarity judgement vs. questioning) and contents (knowledge vs. emotion/moral). Moreover, the youngest participants in our study were approximately 5\u0026ndash;6-years-old, whereas the previous studies included children as young as 3\u0026ndash;4-years-old. Thus, age-related differences might have become apparent if younger children had been included in our sample. Despite these limitations, the age-independent consistency we observed suggests that cultural frameworks for understanding artificial agents may be established early in development. This finding extends previous research on agent concepts \u003csup\u003e\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e,\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e\u003c/sup\u003e by demonstrating that the basic framework for representing AI appears to be influenced by cultural context.\u003c/p\u003e\u003cp\u003eOur findings also provide unique insights into how individuals understand the nature of knowledge itself \u003csup\u003e\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e,\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u003c/sup\u003e. The Japanese participants\u0026rsquo; tendency to represent AI as similar to divine entities suggests that they may view AI knowledge as more aligned with world models or they may not have the distinction between instrumental and worldly knowledge. In contrast, the U.S. participants\u0026rsquo; distinct categorisation of AI might reflect a more instrumental view of AI knowledge, viewing it as a powerful but fundamentally different type of knowledge from that possessed by humans or God. This could indicate an understanding of AI as a sophisticated pattern-matching and predictive system rather than as an entity with true understanding or world knowledge. However, this study did not directly examine participants\u0026rsquo; understanding of the qualitative nature of AI knowledge. While the representational patterns we observed are suggestive of these differing conceptualisations, future research should explicitly investigate how individuals across cultures understand the nature and quality of AI knowledge. This could involve direct comparisons of how participants characterise the knowledge possessed by different agents; for instance, whether they view AI knowledge as instrumental knowledge or world knowledge. Such research could help clarify whether the cultural differences we observed in AI representation indeed reflect deeper distinctions in how different cultures conceptualise the nature of artificial intelligence and its relationship with other forms of knowledge and understanding. This would advance our theoretical understanding of how humans represent different types of knowledgeable agents and have important implications for how AI systems are developed and integrated into different cultural contexts.\u003c/p\u003e\u003cp\u003eAnother innovation of this study is the examination of cognitive structures among agents using representational similarity analyses. Traditional methodologies that use direct questions about agent attributes, while valuable, have limitations in revealing complex interconnections and clustering relationships among different types of agents \u003csup\u003e\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e,\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e\u003c/sup\u003e. Our structural approach allowed us to visualise these relationships more comprehensively and precisely by capturing multi-dimensional patterns that questionnaire-based methods often miss. Specifically, our approach revealed non-linear relationships between different agent categories, identified clustering patterns without imposing predetermined dimensional constraints, and quantified the relative distances between all agent pairs simultaneously, rather than just measuring isolated attribute ratings. Previous methodologies often reduce complex agent perceptions to individual attribute scores, limiting their ability to detect emergent structural patterns or unexpected agent groupings. This methodological advancement builds on our previous work that examined perceptual structures in children, particularly in the domain of colour experience \u003csup\u003e\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e\u003c/sup\u003e. The successful application of this methodology to conceptual knowledge about agents demonstrates its flexibility and potential for studying various aspects of cognitive development.\u003c/p\u003e\u003cp\u003eThe present findings suggest several promising avenues for future research, while also highlighting some limitations of the current set of studies. First, regarding agent perception, while our stimulus set included various types of agents, it was not exhaustive. Future studies should employ a broader range of stimuli to provide a more comprehensive understanding of agent representation. Moreover, while we focused on biological attributes and knowledge in this study, other dimensions such as problem-solving abilities should be examined for a more complete picture of how different agent attributes are represented across cultures. Another important limitation is that our cross-cultural comparison was restricted to Japan and the United States. Since concepts of AI and divine entities are heavily influenced by cultural factors, future research should examine these representations across a broader range of cultures. Further, such broader cultural sampling could help determine the aspects of agent representation that are truly universal and those that are more culturally specific.\u003c/p\u003e\u003cp\u003eMore broadly, our structural approach can be implemented in investigations about the relationships between different cognitive domains that have traditionally been studied separately. By applying the same methodological framework across the perceptual, conceptual, and emotional domains, researchers may reveal novel connections between these areas. Such connections have been examined using behavioural data, brain data, and computational models in cognitive neuroscience research \u003csup\u003e\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e,\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e\u003c/sup\u003e. These future directions and limitations highlight both the potential of our structural approach and the need for continued research in this area.\u003c/p\u003e"},{"header":"Materials and Methods","content":"\u003cp\u003e\u003cb\u003eEthics Information\u003c/b\u003e\u003c/p\u003e\u003cp\u003eThe study protocol was approved by the Ethics Committee of the Psychological Science Unit of Kyoto University (No. 5-P-16, approved on 28th September 2023) and the University of New Hampshire (FY2023-8, approved on 18th June 2024). Informed consent was obtained from the adult participants and all the parents of the child participants when they participated in the experiment (from 30th September to 20th May 2024 in Japan and from 1st August to 12th October 2024 in USA). This study was adhered to the ethical standards established by the Helsinki Declaration of 1964 and its subsequent revisions.\u003c/p\u003e\u003cp\u003e\u003cb\u003eParticipants\u003c/b\u003e\u003c/p\u003e\u003cp\u003eFor participant recruitment for experiments involving Japanese participants, we referred to a university database and collaborated with a research company (Cross Marketing Inc. Tokyo, Japan). For Experiment 1, we recruited 73 Japanese children (mean age in months\u0026thinsp;=\u0026thinsp;68.30 (SD\u0026thinsp;=\u0026thinsp;7.13), 37 girls) from nursery schools. For Experiment 2, we recruited 62 Japanese preschool children (mean age in months\u0026thinsp;=\u0026thinsp;61.12 (SD\u0026thinsp;=\u0026thinsp;12.77), 30 girls); 53 school-aged children (mean age in months\u0026thinsp;=\u0026thinsp;113.70 (SD\u0026thinsp;=\u0026thinsp;18.90), 23 girls); and 40 adults (mean age\u0026thinsp;=\u0026thinsp;43.75 years (SD\u0026thinsp;=\u0026thinsp;12.54), 16 women). For Experiment 3, we recruited 21 Japanese preschool children (mean age in months\u0026thinsp;=\u0026thinsp;71.86 (SD\u0026thinsp;=\u0026thinsp;10.03), 10 girls); 52 school-aged children (mean age in months\u0026thinsp;=\u0026thinsp;109.43 (SD\u0026thinsp;=\u0026thinsp;17.84), 20 girls); and 33 adults (mean age\u0026thinsp;=\u0026thinsp;28.21 years (SD\u0026thinsp;=\u0026thinsp;3.55), 12 women). For the U.S. population, we recruited 36 U.S. preschool children (mean age in months\u0026thinsp;=\u0026thinsp;64.60 (SD\u0026thinsp;=\u0026thinsp;12.40), 22 girls); 38 school-aged children (mean age in months\u0026thinsp;=\u0026thinsp;106.42 (SD\u0026thinsp;=\u0026thinsp;16.07), 19 girls); and 40 religious (mean age\u0026thinsp;=\u0026thinsp;43.16 years (SD\u0026thinsp;=\u0026thinsp;17.45), 20 women) and 38 non-religious adults (mean age\u0026thinsp;=\u0026thinsp;37.43 years (SD\u0026thinsp;=\u0026thinsp;11.82), 19 women) from a university database and Prolific for Experiment 1 and 2. The age in months between countries were not different in preschool and school-aged children (ps\u0026thinsp;\u0026gt;\u0026thinsp;.05). Additionally, we recruited five Japanese preschool children and eight school-aged children in Experiment 2; three Japanese preschool children, six school-aged children, and seven adults in Experiment 3; and three U.S. preschool children, eight school-aged children, and two adults in Experiment 1 and 2, but their data were excluded from analyses for not meeting the study criteria (explained further below).\u003c/p\u003e\u003cp\u003ePreschool children in both Japan and the USA are familiar with the concept of God/gods \u003csup\u003e\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e,\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e\u003c/sup\u003e. Regarding AI, before the main experiments, we conducted preliminary studies in Japan and the United States to assess children\u0026rsquo;s understanding of the term \u0026lsquo;AI\u0026rsquo;. In Japan, children were presented with images of a computer and a tree and asked to indicate which was more closely related to AI. In the United States, children were directly asked whether they knew about AI. Preschool children in both countries performed at chance level (46/73 in Japan, 21/35 in the USA), indicating no cross-cultural differences in their basic comprehension of the AI concept. Although these findings suggest a limited explicit understanding of AI among preschoolers, we proceeded to include this age group in our main study. This decision was guided by our primary research objective: to investigate how young children conceptualise entities such as AI and divine entities and to examine how these conceptualisations differ from those of older children and adults, regardless of their explicit understanding of the terminology.\u003c/p\u003e\u003cp\u003e\u003cb\u003eProcedure of Experiment 1\u003c/b\u003e\u003c/p\u003e\u003cp\u003eWe used pictures of a human (a female adult), a God, and an AI (computer) as agents. We asked questions about their biological and psychological properties based on previous studies \u003csup\u003e\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e,\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e\u003c/sup\u003e. Moreover, we added questions on characteristics related to omniscience and omnipotence, which we referred to as \u0026lsquo;extraordinary\u0026rsquo; questions. For instance, we used the terms \u0026lsquo;breathe\u0026rsquo; and \u0026lsquo;hungry\u0026rsquo; as biological properties, \u0026lsquo;pleasure\u0026rsquo; and \u0026lsquo;sad\u0026rsquo; as psychological questions, and \u0026lsquo;omniscience\u0026rsquo; and \u0026lsquo;omnipotence\u0026rsquo; as extraordinary questions. The experimental procedure was a modified version of the experiment conducted by Jipson and Gelman \u003csup\u003e\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e\u003c/sup\u003e. Children were introduced to a picture of each agent in random order and then asked a series of test questions. The children were then asked questions about biological properties (\u0026lsquo;Can X breathe?\u0026rsquo;; \u0026lsquo;Can X feel hungry?\u0026rsquo;) and psychological properties (\u0026lsquo;Can X feel pleasure?\u0026rsquo;; \u0026lsquo;Can X feel sad?\u0026rsquo;), as well as extraordinary questions (\u0026lsquo;Can X know everything?\u0026rsquo;; \u0026lsquo;Can X solve any difficult problem?\u0026rsquo;) in a random order. The experiment was conducted in-person. We used the term \u0026lsquo;God\u0026rsquo; for U.S. participants and \u0026lsquo;kami-sama\u0026lsquo; (gods) for Japanese participants, adapting to each cultural context.\u003c/p\u003e\u003cp\u003eChildren\u0026rsquo;s responses were scored according to the proportion of \u0026lsquo;yes\u0026rsquo; responses as a function of the question property (biological, psychological, and extraordinary) and agent type (human, AI, and God). Scores ranged from 0 to 2, with a score of 0 indicating that children did not attribute the property to the agent and a score of 2 indicating that children did attribute the property to the agent.\u003c/p\u003e\u003cp\u003eAfter pre-registration (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://osf.io/bfgpy/?view_only=6528384d20e24e969d932df57c15dc81\u003c/span\u003e\u003cspan address=\"https://osf.io/bfgpy/?view_only=6528384d20e24e969d932df57c15dc81\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e), we collected the data of U.S. children. The procedures for U.S. children were the same as those of the Japanese children, except that the picture of AI included a computer, a smartphone, and a smart speaker in the U.S. population, whereas only a computer was shown in the Japanese population, based on the assumption that such devices can facilitate U.S. children\u0026rsquo;s understanding of AI.\u003c/p\u003e\u003cp\u003e\u003cb\u003eProcedure of Experiment 2\u003c/b\u003e\u003c/p\u003e\u003cp\u003eWe used pictures of 10 agents, including a human adult, a God, an AI used in Experiment 1, as well as a human girl, a human baby, a chimpanzee, a frog, a robot, a ghost, and a stone. On any given trial, two stimuli were simultaneously presented. There were 55 unique agent pairs, including identical (e.g. human adult vs. human adult) and non-identical pairs (e.g. human adult vs. God), resulting in 55 trials. We only presented a unique agent pair once.\u003c/p\u003e\u003cp\u003e After obtaining their consent, participants were shown a sequence of task instructions. Specifically, they evaluated how similar they thought the various living creatures and artefacts were to each other in terms of how much knowledge they have about many things. Next, they chose one from the following four options: \u0026lsquo;dissimilar\u0026rsquo;, \u0026lsquo;slightly dissimilar\u0026rsquo;, \u0026lsquo;slightly similar\u0026rsquo;, or \u0026lsquo;similar\u0026rsquo;. To check whether they understood the instructions, participants were given a practice trial, in which they were presented with an identical stimuli pair (human male vs. human male, not presented in the main experiment). When they chose \u0026lsquo;similar\u0026rsquo;, they proceeded to the main experiment. Otherwise, they were provided feedback and underwent another practice trial. Subsequently, the main experiment was conducted. In each trial, participants were presented with the stimuli pair and asked to report the perceived similarity between the agents in terms of how much knowledge they have about things. The experiment was conducted online.\u003c/p\u003e\u003cp\u003eData from some of the participants were excluded from the main analysis based on certain criteria. Specifically, we assessed the reliability of the participants\u0026rsquo; responses. Since participants were required to choose \u0026lsquo;Similar\u0026rsquo; responses in the identical pairs (e.g. human adult vs. human adult) in the practice trial, we calculated the percentage of \u0026lsquo;Similar\u0026rsquo; responses for the identical pairs in the test trials. We excluded participants who chose \u0026lsquo;Similar\u0026rsquo; responses in fewer than 7 out of 10 trials.\u003c/p\u003e\u003cp\u003eThe participants\u0026rsquo; responses were coded on a four-point scale (1 = \u0026lsquo;dissimilar\u0026rsquo;, 2 = \u0026lsquo;slightly dissimilar\u0026rsquo;, 3 = \u0026lsquo;slightly similar\u0026rsquo;, and 4 = \u0026lsquo;similar\u0026rsquo;).\u003c/p\u003e\u003cp\u003eAfter pre-registration, we collected the data of U.S. children and adults. The procedures for U.S. participants were the same as those of the Japanese participants, except two points. First, as in Experiment 1, the picture of AI included a computer, a smartphone, and a smart speaker in the U.S. population. Second, in the practice phase, participants were presented with pictures of an adult male and a bug and asked to evaluate the similarity between the two. They were required to respond \u0026lsquo;dissimilar\u0026rsquo; in this practice trial.\u003c/p\u003e\u003cp\u003e\u003cb\u003eProcedure of Experiment 3\u003c/b\u003e\u003c/p\u003e\u003cp\u003e The materials and procedure were the same as in Experiment 2, except that participants evaluated how similar they thought various living creatures and artefacts were to each other in terms of animacy.\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003eFunding\u003c/h2\u003e\u003cp\u003eThis study received support through the Grant-in-Aid for Transformative Research Areas (23H04830) from the Japan Society for the Promotion of Science. The funder had no role in the following: study design, data collection, analysis, decision to publish, or preparation of the manuscript.\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eY.M. developed the study concept. Y.M. and B.R.J. contributed to the study design. Y.M., R.W., and B.R.J. performed data collection. All authors contributed to data analysis and interpretation. Y.M. drafted the manuscript. R.W., Y.H., and B.R.J. revised the manuscript. All authors approved the manuscript for submission.\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eAll the experimental data are available via (https://doi.org/10.6084/m9.figshare.28814855)\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eBroadbent E (2017) Interactions with robots: The truths we reveal about ourselves. Annu Rev Psychol 68:627\u0026ndash;652\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBlut M, Wang C, W\u0026uuml;nderlich NV, Brock C (2021) Understanding anthropomorphism in service provision: a meta-analysis of physical robots, chatbots, and other AI. J Acad Mark Sci 49:632\u0026ndash;658\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGeiselmann R, Tsourgianni A, Deroy O, Harris LT (2023) Interacting with agents without a mind: the case for artificial agents. Curr Opin Behav Sci 51:101282\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAndries V, Robertson J (2023) Alexa doesn\u0026rsquo;t have that many feelings: Children\u0026rsquo;s children\u0026rsquo;s understanding of AI through interactions with smart speakers in their homes. Comput Educ Artif Intell 5:100176\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eJipson JL, Gelman SA (2007) Robots and rodents: children\u0026rsquo;s inferences about living and nonliving kinds. Child Dev 78:1675\u0026ndash;1688\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGray HM, Gray K, Wegner DM (2007) Dimensions of mind perception. Science 315:619\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAwad E et al (2018) The Moral Machine experiment. Nature 563:59\u0026ndash;64\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWaytz A, Gray K, Epley N, Wegner DM (2010) Causes and consequences of mind perception. Trends Cogn Sci 14:383\u0026ndash;388\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWeisman K, Dweck CS, Markman EM (2017) Rethinking people\u0026rsquo;s conceptions of mental life. \u003cem\u003eProc. Natl. Acad. Sci. U. S. A.\u003c/em\u003e 114, 11374\u0026ndash;11379\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTakahashi H, Ban M, Asada M (2016) Semantic differential scale method can reveal multi-dimensional aspects of mind perception. Front Psychol 7:1717\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eChemero A (2023) LLMs differ from human cognition because they are not embodied. Nat Hum Behav 7:1828\u0026ndash;1829\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKocoń J et al (2023) ChatGPT: Jack of all trades, master of none. Inf Fusion 99:101861\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHoffmann J et al (2022) Training Compute-Optimal Large Language Models. \u003cem\u003earXiv [cs.CL]\u003c/em\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGarry M, Chan WM, Foster J, Henkel LA (2024) Large language models (LLMs) and the institutionalization of misinformation. Trends Cogn Sci 28:1078\u0026ndash;1088\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLiu J (2024) ChatGPT: perspectives from human-computer interaction and psychology. Front Artif Intell 7:1418869\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKosinski M (2024) Evaluating large language models in theory of mind tasks. \u003cem\u003eProc. Natl. Acad. Sci. U. S. A.\u003c/em\u003e 121, e2405460121\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eCollins KM et al (2024) Evaluating language models for mathematics through interactions. \u003cem\u003eProc. Natl. Acad. Sci. U. S. A.\u003c/em\u003e 121, e2318124121\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eStrachan JWA et al (2024) Testing theory of mind in large language models and humans. Nat Hum Behav 8:1285\u0026ndash;1295\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eYildirim I, Paul LA (2024) From task structures to world models: What do LLMs know? Trends Cogn Sci 28:P404\u0026ndash;415\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMahowald K et al (2024) Dissociating language and thought in large language models. Trends Cogn Sci 28:517\u0026ndash;540\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGoddu MK, No\u0026euml; A, Thompson E (2024) LLMs don\u0026rsquo;t know anything: reply to Yildirim and Paul. Trends Cogn Sci 28:963\u0026ndash;964\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBarlev M, Mermelstein S, German TC (2018) Representational coexistence in the God concept: Core knowledge intuitions of God as a person are not revised by Christian theology despite lifelong experience. Psychon Bull Rev 25:2330\u0026ndash;2338\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePurzycki BG (2013) The minds of gods: a comparative study of supernatural agency. Cognition 129:163\u0026ndash;179\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eFesterling J, Siraj I (2020) Alexa, what are you? Exploring primary school children\u0026rsquo;s ontological perceptions of digital voice assistants in open interactions. Hum Dev 64:26\u0026ndash;43\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKnight N, Sousa P, Barrett JL, Atran S (2004) Children\u0026rsquo;s attributions of beliefs to humans and God: cross-cultural evidence. Cogn Sci 28:117\u0026ndash;126\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMoriguchi Y, Takahashi H, Nakamata T, Todo N (2019) Mind perception of God in Japanese children. Int J Psychol 54:557\u0026ndash;562\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBarrett JL, Richert RA, Driesenga A (2001) God\u0026rsquo;s beliefs versus mother\u0026rsquo;s: the development of nonhuman agent concepts. Child Dev 72:50\u0026ndash;65\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLane JD, Wellman HM, Evans EM (2010) Children\u0026rsquo;s understanding of ordinary and extraordinary minds: Understanding extraordinary minds. Child Dev 81:1475\u0026ndash;1489\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHeiphetz L, Lane JD, Waytz A, Young LL (2016) How children and adults represent God\u0026rsquo;s mind. Cogn Sci 40:121\u0026ndash;144\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eFlanagan T, Wong G, Kushnir T (2023) The minds of machines: Children\u0026rsquo;s beliefs about the experiences, thoughts, and morals of familiar interactive technologies. Dev Psychol 59:1017\u0026ndash;1031\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSpatola N, Marchesi S, Wykowska A (2022) Different models of anthropomorphism across cultures and ontological limits in current frameworks the integrative framework of anthropomorphism. Front Robot AI 9:863319\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eIkari S et al (2023) Religion-related values differently influence moral attitude for robots in the United States and Japan. J Cross Cult Psychol 54:742\u0026ndash;759\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKou G, Zhang S (2024) The influence of culture in shaping anthropomorphic attitudes towards robots: A literature review. in \u003cem\u003eHCI International 2024 Posters\u003c/em\u003e 357\u0026ndash;371Springer Nature Switzerland, Cham\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKriegeskorte N, Mur M, Bandettini P (2008) Representational similarity analysis - connecting the branches of systems neuroscience. Front Syst Neurosci 2:4\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZeleznikow-Johnston A, Aizawa Y, Yamada M (2023) \u0026amp; Tsuchiya, N. Are Color Experiences the Same across the Visual Field? J Cogn\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMoriguchi Y et al Comparing color qualia structures through a novel similarity task in young children versus adults. \u003cem\u003eProceedings of the National Academy of Sciences, 122(11), e2415346122.\u003c/em\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHatano G, Inagaki K (1994) Young children\u0026rsquo;s naive theory of biology. Cognition 50:171\u0026ndash;188\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWellman HM, Cross D, Watson J (2001) Meta-analysis of theory-of-mind development: the truth about false belief. Child Dev 72:655\u0026ndash;684\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGelman SA, Legare CH (2011) Concepts and folk theories. Annu Rev Anthropol 40:379\u0026ndash;398\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMara M, Appel M (2015) Science fiction reduces the eeriness of android robots: A field experiment. Comput Hum Behav 48:156\u0026ndash;162\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSpriet C, Abassi E, Hochmann J-R, Papeo L (2022) Visual object categorization in infancy. \u003cem\u003eProc. Natl. Acad. Sci. U. S. A.\u003c/em\u003e 119\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"},{"header":"Footnotes","content":"\u003col style=\"list-style-type: lower-alpha;\"\u003e\n \u003cli\u003e\u003cspan\u003e\u0026nbsp;We used the term \u0026lsquo;God\u0026rsquo; for US participants and \u0026lsquo;kami-sama\u0026rsquo; (gods) for Japanese participants, adapting to each cultural context. However, in the results section, we consistently use \u0026lsquo;God\u0026rsquo; to avoid redundancy while presenting findings.\u003c/span\u003e\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"humanities-and-social-sciences-communications","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"palcomms","sideBox":"Learn more about [Humanities \u0026 Social Sciences Communications](http://www.nature.com/palcomms/)","snPcode":"41599","submissionUrl":"https://submission.springernature.com/new-submission/41599/3","title":"Humanities and Social Sciences Communications","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Nature AJ","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"artificial intelligence, divine entities, children, agent perception, development","lastPublishedDoi":"10.21203/rs.3.rs-6824602/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6824602/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eAdvances in generative artificial intelligence (AI) have raised foundational questions about how we conceptualise such entities alongside traditional agents like humans and divine beings. Research demonstrates that both children and adults tend to attribute similar mental properties to AI and divine beings across cultures. However, fundamental differences exist in the nature of knowledge attributed to humans, AI and divine entities. Here we showed that Japanese children and adults across developmental stages conceptually cluster AI with divine entities in knowledge-based representational space, whereas U.S. participants maintain clear categorical boundaries between these agents using representational similarity analyses focusing on knowledge attributes. Our findings stand in contrast with previous cross-cultural investigations reporting shared basic agent structures across cultures. These results offer implications for theoretical models of agent categorisation, cross-cultural AI design, and our understanding of how humans reason about increasingly intelligent artificial minds in culturally contingent ways.\u003c/p\u003e","manuscriptTitle":"Knowledge-based representations of artificial intelligence and divine agents: A developmental study across Japan and the United States","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-10-10 14:47:20","doi":"10.21203/rs.3.rs-6824602/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-11-06T18:30:21+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-10-28T09:14:40+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-10-20T17:06:23+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"1840226690116157605623206159051802997","date":"2025-10-20T11:19:29+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"4873012501969917472877965076912861134","date":"2025-10-13T12:48:38+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"93994509867455769890930856129631190547","date":"2025-10-05T18:47:05+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-09-26T11:08:44+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-09-26T11:04:56+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2025-07-15T15:06:30+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-07-10T01:26:50+00:00","index":"","fulltext":""},{"type":"submitted","content":"Humanities and Social Sciences Communications","date":"2025-07-10T01:23:32+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"humanities-and-social-sciences-communications","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"palcomms","sideBox":"Learn more about [Humanities \u0026 Social Sciences Communications](http://www.nature.com/palcomms/)","snPcode":"41599","submissionUrl":"https://submission.springernature.com/new-submission/41599/3","title":"Humanities and Social Sciences Communications","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Nature AJ","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"ab3a50fc-8387-4419-8a62-343154fedce0","owner":[],"postedDate":"October 10th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[{"id":55828042,"name":"Social science/Cultural and media studies"},{"id":55828043,"name":"Social science/Psychology"}],"tags":[],"updatedAt":"2026-04-27T13:39:05+00:00","versionOfRecord":[],"versionCreatedAt":"2025-10-10 14:47:20","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6824602","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6824602","identity":"rs-6824602","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}