Cute Names, Babyfaces and AI Avatars Bias Who We Expect to Sacrifice a Life

A name like Liu Guoguo, a face with a large forehead and small chin, or a deliberately cute artificial intelligence avatar can quietly change how people expect a stranger to behave in a life-and-death moral dilemma. In a series of three experiments with 858 Chinese university students, researchers report that decision-makers associated with infant-like cues were consistently judged less likely to make the utilitarian choice in a trolley scenario, the choice to actively divert a runaway trolley so that one person dies instead of five. The bias appeared whether the cue was linguistic, facial, or embedded in the design of a non-human agent, and it persisted even though none of the cues carried any genuine information about the target’s moral character.

The study, published in Frontiers in Psychology by Zhicong Zhang and Xiuge Liu of Quanzhou Normal University, addresses a question that is distinct from the classic focus of moral psychology. Rather than asking how people resolve sacrificial dilemmas themselves, the researchers examined how observers predict the choices of others. Such predictions are far from idle: they inform whom people trust, hire, cooperate with, or delegate responsibility to, and they increasingly shape public attitudes toward artificial intelligence systems entrusted with morally significant decisions, from autonomous vehicles to clinical decision support tools. Previous work had shown that evaluations of moral decision-makers are shaped by characteristics such as gender, race, socioeconomic status, and cultural background, but relatively little was known about whether superficial cues like a name or a face could bias expectations about someone’s future moral choices.

The theoretical anchor of the research is the baby schema, or Kindchenschema, a cluster of infantile physical characteristics first described by Konrad Lorenz in 1943: a large head, round face, large eyes, and chubby cheeks. These features are widely regarded as an innate releasing mechanism that triggers stereotypical perceptual and behavioral reactions, most notably perceived cuteness, an affective response associated with warmth, tenderness, and protective motivation. The authors are careful to distinguish the feature configuration itself from the observer’s response, treating cuteness as the proximal affective reaction elicited by the features rather than a property of the stimuli. Crucially, they treat repeated two-syllable names, babyfaced appearance, and cute avatar designs not as interchangeable measures of a single latent construct, but as convergent operationalizations of cues that share the function of evoking infant-like associations across very different surface forms.

Each of these cue types has an established history of shaping social perception. Babyfaced adults are perceived as friendlier, more trustworthy, and less threatening, but also as less competent and less physically capable. These impressions carry real consequences: babyfaced individuals are judged less harshly in legal contexts, viewed as less suitable for leadership positions, seen as less competent in political settings, and receive more empathy and assistance when in need. In applied settings, babyfaced physicians generate higher patient satisfaction but are also more likely to be perceived as deceptive in malpractice cases. Linguistic cues operate similarly, with repeated two-syllable names such as Huahua Zhang or Bingbing Wang, common in Chinese naming practice, evoking perceptions of cuteness and leading observers to infer greater warmth but lower competence. Cute design features in non-human agents likewise increase perceived warmth, trust, emotional attachment, and forgiveness after service failures.

Experiment 1 tested the purely linguistic cue. A total of 215 students were recruited, with ten excluded for excessively long completion times or failed attention checks, leaving a final sample of 205 participants randomly assigned to see either a repeated two-syllable name, Liu Guoguo, or a non-repeated two-syllable name, Liu Guo. Participants first read the standard trolley dilemma and reported their own behavioral inclination, then read a modified version in which a track worker in a control room could switch the track, sacrificing one person to save five. A manipulation check confirmed the names worked as intended: repeated names were rated as significantly more baby-like, with a large effect size of d = 1.01, while name familiarity did not differ between conditions. Participants then estimated the percentage likelihood that the named track worker would make the utilitarian choice.

The effect was clear. Because participants’ own choices were strongly related to their predictions, with 65.9 percent choosing the utilitarian option themselves, the primary analysis was a two-way analysis of variance controlling for participants’ own moral choice. The main effect of name condition was significant: the decision-maker with the repeated name received an adjusted mean likelihood of 59.55 percent, compared with 69.32 percent for the non-repeated name. A supplementary comparison without the covariate yielded a smaller but still reliable effect, d = 0.38, and the pattern held when gender and age were added as covariates. Participants who had chosen the utilitarian option themselves were also more likely to expect the decision-maker to do the same, indicating a projection component in moral prediction, but this did not interact with the name manipulation.

Experiment 2 replicated the effect with visual stimuli. From 333 recruited students, the final sample was 307 participants, randomly assigned to view either a babyfaced or a mature-faced photograph of the decision-maker. The manipulation check confirmed that the babyfaced target was perceived as more baby-faced, d = 0.99, and also as warmer and more empathetic, though the two faces did not differ significantly in perceived trustworthiness or expertise. The main effect on moral prediction was substantial: the babyfaced decision-maker received an adjusted mean likelihood of 55.05 percent versus 67.92 percent for the mature-faced target, an unadjusted effect of d = 0.62. But the mediation analysis produced a surprise. Perceived warmth was positively, not negatively, associated with the expected likelihood of utilitarian choice, and perceived competence did not differ between conditions and did not mediate the effect. The warmth-based account the researchers had initially considered was therefore not supported, and no other measured perception, including trust and empathy, provided a consistent explanation.

Experiment 3 extended the question to non-human agents, framed explicitly as an exploratory extension rather than a confirmatory replication. From 375 recruited students, the final sample was 346 participants assigned to view either a high-cuteness or a low-cuteness static AI avatar, drawn from stimuli used in prior research, with no human operator mentioned or depicted. The high-cuteness avatar was rated as cuter and warmer, while the low-cuteness avatar was rated as more serious and colder; the two did not differ in perceived expertise or competence. The pattern of moral expectations matched the human experiments: the high-cuteness avatar received an adjusted mean likelihood of 62.81 percent versus 71.80 percent for the low-cuteness avatar, with an unadjusted effect of d = 0.39 that held under Welch’s correction for unequal variances. Again, perceived warmth was positively associated with expected utilitarian action and could not explain the bias. An exploratory analysis hinted at a possible alternative: cute avatars were perceived as less serious, and seriousness was positively associated with expected utilitarian action, but this indirect effect was marginal and the authors urge caution.

The broader picture that emerges is one of a reliable cue-based bias in moral inference whose mechanism remains unresolved. The authors note that because utilitarian action in the trolley dilemma requires active intervention, expectations that baby-schema targets will refrain from switching the track may partly reflect assumptions about passivity or low assertiveness rather than genuine deontological commitment, a distinction future studies could disentangle using process dissociation procedures or the CNI model, which separates consequences, norms, and inaction. Several further limitations temper the findings: the studies used a single impersonal dilemma, relied on a small number of facial stimuli, did not measure perceived age, and did not verify whether participants in Experiment 3 treated the avatar as an autonomous AI decision-maker or as a representation of a human operator. The repeated two-syllable names are meaningful in Chinese naming practice and may not carry the same childlike connotations in other languages. Even so, the practical implications are striking. AI systems in autonomous driving, healthcare, and customer service increasingly use anthropomorphic designs to encourage acceptance, and these results suggest that visual appearance shapes users’ moral expectations as much as system functionality does. Designers may intentionally or unintentionally steer moral perception through avatar aesthetics, raising ethical concerns about the deliberate use of appearance-based cues to influence trust, particularly when those cues are unrelated to the system’s actual decision-making processes.

Subject of Research: The influence of baby-schema-related cues in names, faces, and AI avatar designs on observers' moral expectations about decision-makers in sacrificial dilemmas

Article Title: Baby-schema-related cues shape moral expectations about decision-makers: evidence from names, faces, and AI avatars

Article References: Zhang, Z., & Liu, X. (2026). Baby-schema-related cues shape moral expectations about decision-makers: evidence from names, faces, and AI avatars. Frontiers in Psychology, 17, Article 1958973. https://doi.org/10.3389/fpsyg.2026.1958973

Image Credits: AI Generated

DOI: 10.3389/fpsyg.2026.1958973

Keywords: baby schema, moral psychology, trolley dilemma, utilitarian judgment, person perception, cuteness, AI avatars, social cognition, warmth and competence, moral prediction, face perception, human-AI interaction