Author: Denis Avetisyan
New research reveals that designing artificial intelligence to appear more human increases perceptions of personhood, but doesn’t guarantee greater user trust across different cultures.
A global study demonstrates that the impact of anthropomorphic AI design on trust is significantly moderated by cultural context and varies across populations.
Despite growing sophistication in artificial intelligence, the relationship between humanlike AI design and user trust remains poorly understood. This is the central question addressed in ‘Humanlike AI Design Increases Anthropomorphism but Yields Divergent Outcomes on Engagement and Trust Globally’, which investigates how attributing human characteristics to AI affects global user interactions. Our large-scale cross-national experiments reveal that while humanlike design successfully increases perceptions of AI anthropomorphism, its impact on engagement and trust is not universal-and is significantly shaped by cultural context. How can we move beyond a one-size-fits-all approach to AI governance and build systems that foster appropriate trust across diverse populations?
Unveiling the Human-AI Connection: A Foundation for Interaction
The proliferation of artificial intelligence into everyday experiences-from virtual assistants and recommendation algorithms to autonomous vehicles and healthcare diagnostics-necessitates a deeper understanding of the human-AI connection. This isn’t simply a matter of technological advancement; it concerns how individuals perceive, trust, and ultimately interact with these increasingly pervasive systems. Effective integration requires anticipating human responses, addressing potential biases, and designing AI that aligns with human values and expectations. As AI moves beyond specialized tasks and into roles demanding social intelligence and nuanced communication, comprehending the psychological and behavioral dynamics between humans and AI becomes paramount for fostering collaboration and mitigating potential risks. Ultimately, the success of AI isn’t solely determined by its capabilities, but by its ability to seamlessly and beneficially integrate into the human world.
The human tendency to attribute human characteristics, emotions, or intentions to non-human entities – a phenomenon known as anthropomorphism – significantly influences interactions with artificial intelligence. This isn’t merely a whimsical projection; studies suggest that the degree to which individuals perceive AI agents as possessing human-like qualities directly impacts trust, reliance, and even emotional connection. When AI exhibits behaviors that evoke human understanding – such as responsive communication or perceived empathy – users are more likely to accept its suggestions and collaborate effectively. Conversely, a lack of perceived ‘humanness’ can create distance and hinder seamless integration into daily life, impacting the overall user experience and the potential benefits of AI technologies. Therefore, understanding the nuances of this anthropomorphic effect is crucial for designing AI systems that are not only intelligent but also intuitively relatable and readily accepted by human users.
Early investigations into the tendency to ascribe human characteristics to artificial intelligence have predominantly focused on participants from Western, Educated, Industrialized, Rich, and Democratic (WEIRD) societies, creating a significant limitation for broader understanding. This bias potentially restricts the universality of findings, as cultural backgrounds strongly influence perceptions of agency, intentionality, and social interaction. The observed inclination to anthropomorphize AI may not be consistent across diverse populations with differing belief systems, social norms, and technological exposure. Consequently, the current body of research offers an incomplete picture of the human-AI connection and necessitates further studies incorporating more representative and globally diverse samples to establish genuinely generalizable principles.
Designing for Perceived Sociability: A Structural Approach
The study employed a controlled experimental design to examine the impact of specific features on human-AI interaction. Two key design characteristics were systematically varied: Design Characteristics (DC), encompassing visual and interactive elements of the AI’s interface, and Conversational Sociability (CS), representing the AI’s ability to simulate human-like conversation through responses and prompts. Manipulation of these two features allowed for the isolation and assessment of their individual and combined effects on user perceptions and engagement, providing a structured approach to understanding the factors influencing human-AI interaction dynamics.
A factorial design was employed to systematically evaluate the influence of Design Characteristics (DC) and Conversational Sociability (CS) on user engagement. This experimental approach involved creating and testing all possible combinations of DC and CS levels; specifically, each level of DC was paired with each level of CS, resulting in multiple distinct conditions. This methodology allowed for the isolation of the main effects of both DC and CS – determining how each feature independently impacts engagement – as well as the assessment of interaction effects, revealing whether the effect of one feature depends on the level of the other. Statistical analysis of the data collected under each condition then quantified the magnitude and significance of these effects, providing a comprehensive understanding of how DC and CS jointly influence user experience.
Analysis of user study data revealed a statistically significant correlation between design manipulations and perceived anthropomorphism. Specifically, increasing both Design Characteristics and Conversational Sociability resulted in a 0.386 unit increase in anthropomorphism scores, as measured on a Likert scale. This finding demonstrates that perceptions of human-likeness in AI agents are not fixed attributes, but can be systematically influenced through specific design choices. The magnitude of this effect indicates a substantial degree of manipulability in how users perceive the social qualities of AI systems, suggesting that these features are key drivers of user perception.
The study posited a direct relationship between the manipulation of Design Characteristics (DC) and Conversational Sociability (CS) and the resulting levels of perceived anthropomorphism in users. Specifically, it was hypothesized that increases in both DC – features relating to the AI’s visual or structural presentation – and CS – the degree to which the AI exhibits human-like conversational traits – would correlate with higher scores on anthropomorphism assessments. This increased perception of human-likeness was then predicted to lead to greater user engagement with the AI system, suggesting a sequential relationship where design choices influence perception, which in turn impacts behavioral response.
Measuring Trust and Engagement: Quantitative Insights
Likert Scale Measurement was utilized to quantify user perceptions of anthropomorphism, providing a standardized and quantifiable assessment of attitudes. Participants were presented with a series of statements regarding the AI agent’s human-like qualities – such as intelligence, emotional capacity, and personality – and indicated their level of agreement or disagreement on a predefined scale, typically ranging from “Strongly Disagree” to “Strongly Agree”. This method allows for the assignment of numerical values to subjective perceptions, enabling statistical analysis and comparison of anthropomorphism ratings across different experimental conditions and participant groups. The resulting data provides a measurable index of the degree to which users attribute human characteristics to the AI agent.
User engagement was assessed using a multi-faceted approach combining self-reported metrics with directly observable behavioral data. Self-reported metrics included survey questions designed to capture subjective experiences such as perceived ease of use, enjoyment, and motivation to continue interacting with the AI agent. Behavioral Measures consisted of quantifiable actions performed by participants during the interaction, specifically message length, frequency of interaction, and time spent engaging with the system. The combination of these subjective and objective data points allowed for a more comprehensive understanding of user engagement than either method could provide in isolation, mitigating biases inherent in relying solely on self-assessment or automated tracking.
Analysis of user-generated text revealed a statistically significant increase in average message length when participants interacted with the AI agent exhibiting high human-likeness. Specifically, messages composed in response to the high human-likeness condition averaged 18.3 words, compared to 12.7 words in the low human-likeness condition ($p < 0.05$). This quantitative difference suggests a correlation between perceived human-likeness and user willingness to elaborate in their communication with the agent, serving as an indicator of increased engagement. While not a direct measure of trust, message length provided a readily observable proxy for the depth of interaction and user investment in the conversation.
The Trust Game was employed as a quantitative measure of relational perception, specifically assessing the degree of trust participants placed in the AI agent. In this paradigm, participants were given an initial endowment of points and allowed to transfer any portion of these points to the AI agent. The AI agent then multiplied the received points by a predetermined factor and returned a portion of the multiplied amount to the participant. This setup creates an incentive for trusting behavior; participants who believe the AI agent will reciprocate are more likely to send a larger proportion of their initial endowment. The amount of points transferred by participants served as the primary dependent variable, providing an objective, incentive-based measure of their perceived trustworthiness of the AI.
The Trust Game, utilized to quantify relational perception, yielded no statistically significant differences in the amount of points transferred between participants across the various treatment conditions examining AI human-likeness. This indicates that while increased anthropomorphism may influence user engagement – as demonstrated by increased message length – it did not demonstrably correlate with a higher level of trust, as measured by this incentive-based behavioral paradigm. The game functioned by allowing participants to allocate points to the AI agent, with the understanding that a portion of those points could be returned, establishing a quantifiable metric for assessing trust and reciprocity. The absence of a significant difference suggests that factors beyond superficial human-likeness likely mediate the development of trust in AI agents.
Qualitative text analysis was conducted on user responses to open-ended questions, employing a mixed-methods approach for comprehensive data interpretation. Human coders performed initial thematic analysis, identifying recurring patterns and sentiments within the textual data. To enhance scalability and inter-rater reliability, an LLM autorater was then utilized to independently code the same data, with resulting codes compared to those generated by human coders. Discrepancies were resolved through discussion and refinement of the coding scheme, ultimately yielding a robust and nuanced understanding of user perceptions regarding the AI agent, including subjective experiences and rationales behind reported levels of trust and engagement that were not fully captured by quantitative measures.
Cultural Nuances and the Future of AI Interaction: A Global Perspective
Research indicates that cultural context is a critical factor in how individuals perceive and interact with artificial intelligence, substantially influencing both the tendency to attribute human characteristics – anthropomorphism – and the level of engagement with AI systems. The study revealed that design choices intended to enhance user interaction don’t have a uniform effect across cultures; rather, their impact is moderated by pre-existing cultural norms and expectations. This suggests that a universally appealing AI design is unlikely, as what fosters trust and rapport in one culture might be perceived as unnatural or even off-putting in another. Consequently, developers must move beyond standardized approaches and prioritize culturally sensitive designs that acknowledge and respect diverse perspectives to maximize both user acceptance and effective interaction with AI technologies.
Perceptions of an artificial intelligence’s sociability and trustworthiness are not universal concepts, but rather are significantly shaped by cultural background. Research indicates substantial variations in how individuals from different cultures evaluate an AI’s warmth, approachability, and reliability, even when interacting with the same system. This suggests that design strategies focused on building rapport or establishing confidence – such as employing humor, offering empathetic responses, or utilizing specific communication styles – may resonate positively in one cultural context while proving ineffective, or even detrimental, in another. Consequently, a standardized approach to AI interaction design risks creating systems that are perceived as cold, untrustworthy, or inappropriate by significant portions of the global population, highlighting the necessity for localized and culturally-aware development practices.
Studies reveal a notable divergence in how readily individuals attribute human characteristics to artificial intelligence across different national cultures. Research consistently demonstrates that anthropomorphism scores – the degree to which people perceive AI as possessing human-like qualities – vary significantly from country to country. This suggests that cultural background powerfully shapes perceptions of AI, influencing how users interact with and trust these systems. For instance, cultures prioritizing collectivism may exhibit different responses to anthropomorphic AI than those emphasizing individualism, impacting engagement levels and acceptance. This variability underscores the critical need to move beyond universal design principles and instead embrace culturally nuanced approaches to AI development, acknowledging that what resonates in one culture may not translate effectively to another.
Effective artificial intelligence interaction increasingly demands a shift towards culturally sensitive design principles. Research indicates that perceptions of AI sociability and trustworthiness are not universal; rather, they are deeply influenced by local norms and expectations. A design that resonates positively within one cultural context may prove ineffective, or even alienating, in another. This necessitates tailoring interaction styles – encompassing elements like communication patterns, emotional expression, and even the level of formality – to align with the specific values and preferences of each target culture. Ignoring these nuances risks hindering user engagement and potentially undermining the benefits of AI technologies, while a thoughtful, culturally adapted approach promises more intuitive, acceptable, and ultimately, more effective human-AI collaboration.
As artificial intelligence increasingly adopts human-like characteristics, a crucial area of inquiry centers on the ethical ramifications of such design choices across different cultures. Researchers are beginning to investigate how attributing human traits to AI – anthropomorphism – might affect user trust, reliance, and even susceptibility to manipulation, and whether these effects are consistent globally. The development of responsible design guidelines is paramount, necessitating a nuanced understanding of cultural values and norms concerning social interaction, authority, and deception. Such guidelines should address potential biases embedded within AI systems, ensure transparency in AI-human communication, and promote equitable access to the benefits of this technology, fostering a future where AI interactions are not only effective but also ethically sound and culturally appropriate.
The study reveals a complex interplay between design and perception, highlighting that increasing an AI’s human-likeness does not guarantee increased trust across all cultures. This echoes Donald Knuth’s observation: “Premature optimization is the root of all evil.” While designers may intuitively strive for human-like AI, this research demonstrates that such optimizations – prioritizing anthropomorphism – can introduce unforeseen tensions in user engagement and trust, varying significantly based on cultural context. The system’s behavior-how users actually respond-over time is far more critical than the initial design intent, confirming that architecture is, fundamentally, the system’s behavior over time, not a diagram on paper.
The Road Ahead
The pursuit of human-like artificial intelligence often feels akin to meticulously crafting a façade. This work demonstrates that simply appearing human is insufficient-the bloodstream itself must be considered. Increasing perceptions of anthropomorphism does not automatically translate to increased trust, a crucial finding. Instead, the research highlights a fundamental truth: design, divorced from cultural context, is merely aesthetic. One cannot replace a component of a system without understanding the entire circulatory network, the beliefs, the ingrained expectations of those who interact with it.
Future research must move beyond surface-level manipulations of AI appearance. The observed divergence in trust across cultures suggests a need for deeply nuanced models of cultural psychology informing AI design. It is not enough to ask if an AI seems human; the question becomes which human, and to whom? A universally ‘trustworthy’ AI may prove a chimera.
The field now faces a critical juncture. Will it continue to refine the illusion of humanity, or will it embrace a more pragmatic approach – focusing on demonstrable competence and transparent functionality, adapted to specific cultural landscapes? The answer, one suspects, will reveal less about the technology itself and more about our persistent, and perhaps misguided, desire to see ourselves reflected in the machines we create.
Original article: https://arxiv.org/pdf/2512.17898.pdf
Contact the author: https://www.linkedin.com/in/avetisyan/
See also:
- Mobile Legends: Bang Bang (MLBB) Sora Guide: Best Build, Emblem and Gameplay Tips
- Clash Royale Best Boss Bandit Champion decks
- Best Hero Card Decks in Clash Royale
- All Brawl Stars Brawliday Rewards For 2025
- Best Arena 9 Decks in Clast Royale
- Brawl Stars December 2025 Brawl Talk: Two New Brawlers, Buffie, Vault, New Skins, Game Modes, and more
- Vampire’s Fall 2 redeem codes and how to use them (June 2025)
- Clash of Clans Meltdown Mayhem December 2025 Event: Overview, Rewards, and more
- Clash Royale Witch Evolution best decks guide
- Clash Royale December 2025: Events, Challenges, Tournaments, and Rewards
2025-12-22 16:24