We investigated children's spontaneous impressions of faces; a question critical for understanding the developmental trajectory of facial stereotypes. Adults and children aged 4 to 10 from the UK (children: n = 59, adults: n = 61) and the US (children: n = 189, adults: n = 180) described what they thought when viewing each of the four child's faces. Natural language processing was used to classify free-response descriptors into categories related to traits, emotions, social groups, and appearance. This approach captured over 90% of children's and adults' impressions. The vocabulary and the prevalence of descriptors related to each of the four categories were comparable across two samples that differed in participant and face diversity. Across childhood, trait descriptors increased, with 14 impressions emerging as the top trait words used across both samples. Notably, children as young as four spontaneously formed trait impressions, suggesting an early emergence of facial stereotyping. SUMMARY: Children and adults spontaneously mention personality traits when viewing unfamiliar child faces. Children as young as four mentioned personality traits, and the frequency of trait references increased across childhood. The results converged across two samples that differed in terms of participant and face diversity and the use of real versus AI-generated faces. These findings demonstrate the relevance of trait impressions in early childhood and underscore the importance of increasing diversity in face perception research.

Humans rapidly form prosocial evaluations from physical appearance, and these impressions profoundly shape social interactions. Yet, little is known about the developmental origins of mental representations of “nice” individuals and if these representations reflect visual stereotypes, or appearance cues that shape first impressions. Computational models of young children’s mental representations are essential for addressing these questions, but existing methods are largely limited to older children and adults because they require many trials to generate stable models. In the present research, we adapted a data-driven Generative Reverse Correlation (GRC) method and introduced a child-friendly version capable of visualizing children’s individual- and group-level mental representations in just 40 trials, with children as young as three. In Study 1, children 3-7 (N = 160) categorized 40 synthetic faces as nice or mean to generate mental models. Ratings revealed that the majority of children—across gender and age groups—associate "niceness" with women and individuals high in perceived trustworthiness, happiness, and femininity, highlighting the early emergence of appearance-based stereotyping. In Study 2, children 5-9 (N = 165) categorized faces as nice or smart to compare non-oppositional traits. Findings replicated associations from Study 1, demonstrating that visual stereotypes of niceness emerge across different comparison traits. Together, these results suggest that impressions of niceness are already structured by stable visual biases in the preschool years. Our approach provides a powerful tool for making young children’s mental representations directly observable and offers new opportunities to track the developmental origins of biased impression formation.

The influence of visual appearance on pet adoption is poorly understood, partially due to inconsistent findings. We examined visual biases in the appearance of dogs and cats using a novel data-driven method that leverages generative AI to visualize mental prototypes at both the individual and group levels. The results demonstrated both strong idiosyncratic preferences and shared, group-level consistencies, offering an explanation for previously reported inconsistencies. Importantly, we found that greater similarity between images of real shelter animals and the group prototypes of “adoptable” dogs and cats, as well as “friendly” dogs, was associated with a shorter time in the shelter before adoption. This study provides novel evidence for the interplay between idiosyncratic variation and shared biases in shaping visual stereotypes of companion animals.

First impressions play a powerful role in shaping judgments and decisions, yet little is known about what makes someone appear brilliant. To examine visual stereotypes, or physical appearance attributes that drive impressions of brilliance, we applied a data-driven reverse correlation method that uses generative AI to visualize mental representations of brilliant and smart individuals. In Study 1, adult participants categorized 300 synthetic faces. Based on these categorizations, we generated individual and group-level visual models of “brilliance” and “smartness.” These models allow for parametric manipulations of perceived brilliance and smartness and revealed a high degree of heterogeneity. However, for the majority of participants, brilliance was more masculinized than smartness, and perceived masculinity predicted visual models of brilliance even when controlling for other theoretically relevant attributes (i.e., perceived dominance, privilege, affect). Study 2 replicated this masculinity bias using only female-appearing faces, demonstrating a broader visual stereotype based on physical features rather than perceived gender. Study 3 validated participants’ visual models in a job hiring context, where faces manipulated to look more brilliant were judged as more fit for a job requiring genius-level but not average-level intelligence. Together, these findings reveal a robust “masculine = brilliant” visual stereotype that emerges despite individual variability in mental representations, demonstrate that impressions of brilliance are visually distinct from smartness, and identify how these visual biases can shape consequential real-world outcomes.

Decades of research have shown that androcentric bias (i.e., assuming that men represent the default) is prevalent across text, images, and social interactions. Despite this evidence, recent research on the mental imagery of faces shows an opposite of the expected androcentric bias: Mental representations of "typical" faces appear more like women than men. In the present research, we first aim to reconcile this apparent discrepancy by examining the mental imagery associated with typical persons using generative reverse correlation-a data-driven method that leverages artificial intelligence to construct photo-realistic imagery of visual stereotypes for both groups and individuals. Second, we explore potential reasons for the observed reversal in mental imagery by examining individual differences in typicality judgments. Across two studies, our results show (1) individuals tend to equate "typical persons" with "men" in their mental representations at the group-level; but (2) there is large individual-level variability in both "typical person" mental imagery and explicit "typicality" judgments that complicates conclusions drawn from aggregate-level data.

Across two studies (N = 4,526), we characterize a taxonomy of spontaneous face impressions by applying artificial intelligence text analyses to thousands of free-response descriptions of computer-generated faces. The taxonomy codes almost 100% of the impressions into Appearance (including Beauty), Sociability, Morality, Ability, Assertiveness, Emotion, Social Group, socioeconomic Status, Uniqueness, Family, Health, Occupation, Geographic origin, and political-religious Beliefs content. Results suggest that dimensions from low-dimensional models (e.g., Communion, Agency facets) are highly prevalent, but that alternative dimensions such as Uniqueness and Health are also prevalent. Most dimensions show high (positive) directions, and their correlational structure supports the clustering of low-dimensional models as separate from the expanded taxonomy dimensions. Finally, the taxonomy improves predictions of general evaluations of faces (how positive/negative the face is evaluated overall) and decision making in hypothetical scenarios (e.g., how much to prioritize a target for health care access or antidiscrimination protections).

Neurons in the human amygdala and hippocampus are classically thought to encode a person's identity invariant to visual features. However, it remains largely unknown how visual information from higher visual cortical areas is translated into such a semantic representation of an individual person. Here, across four experiments (3,581 neurons from 19 neurosurgical patients over 111 sessions), we demonstrate a region-based feature code for faces, where neurons encode faces on the basis of shared visual features rather than associations of known concepts, contrary to prevailing views. Feature neurons encode groups of faces regardless of their identity, broad semantic categories or familiarity; and the coding regions (that is, receptive fields) predict feature neurons' response to new face stimuli. Together, our results reveal a new class of neurons that bridge perception-driven representation of facial features with mnemonic semantic representations, which may form the basis for declarative memory.

A considerable body of research has explored how first impressions from voices influence social decisions, demonstrating the significance of perceived emotional and social traits alongside acoustic parameters across various contexts, including elections, legal decisions, and economic or mating-related choices. In a novel ecological context, specifically healthcare emergency dispatch, where trained nurses respond to critical health situations, we investigated whether caller voice characteristics (e.g., acoustic parameters, gender, emotional valence and arousal, social trait attributions) is related to healthcare professionals' prioritization decisions beyond available clinical information. We selected cases when the call-taker deviated from an algorithm, changing the proposed priority, to understand which factors may bias their decision. We dissociated acoustic-phonetic features from semantic content by manipulating stimuli across different experimental samples, comprised of Italian and American participants. Judgments of emotional traits (valence and arousal) and social traits (perceived trustworthiness, dominance, familiarity, and attractiveness) were collected from all participants while listening brief excerpts (mean audio length of 3 s) from original call recordings. Critically, we found that social attributions of dominance and attractiveness, arousal and valence ratings, as well as control variables such as voice gender, pitch, harmonic-to-noise ratio (HNR), and COVID-19 related pathologies, were significantly associated with instances where call-takers made wrong decisions by either overestimating or underestimating care priority.

Children’s faces are underrepresented in face databases, and existing databases that do focus on children tend to have limitations in terms of the number of faces available and the diversity of ages and ethnicities represented. To improve the availability of children’s faces for experimental research purposes, we created a novel face database that contains 500 artificial images of children that are diverse in terms of both age (ages 3 to 10) and ethnicity (representing 15 different racial or ethnic groups). Using deep neural networks, we produced a large collection of synthetic photographs that look like naturalistic, realistic faces of children. To assess the representativeness of the dataset, adult participants (N = 585) judged the age, gender, ethnicity, and emotion of artificial faces selected from the set of 500 images. The images present a diverse array of artificial children’s faces, offering a valuable resource for research requiring children’s faces. The images and ratings are publicly available to researchers on Open Science Framework (https://osf.io/m78r4/).

We report on five experiments studying people's (n > 12,000) responses to a prototypical random process: predicting the outcomes in a sequence of five fair coin tosses. "Success" rates in making predictions followed the binomial distribution, and randomly assigned participants to zero to five success experiences, capturing the entire distribution of possible performance outcomes without deception. We found that more successful predictions led to more optimistic expectations of future performance and an increased propensity for risk-taking behaviors, whereas more unsuccessful predictions led to more pessimistic expectations and risk-averse behavior, demonstrating the tendency to believe that there is a signal in performance predicting random sequences of events. Inference from performance was stronger for participants who changed their predictions more often, suggesting that it is more likely to emerge when participants detect a spurious correlation between their behavior and the experienced outcomes. The findings could not be explained by distorted beliefs about the nature of the outcome-generating process, poor knowledge of probability, or risk attitudes, and were unaffected by the presence of performance-related rewards.

Complex evaluative judgments from facial appearance are made efficiently and are consequential. We review some of the most important findings and methods over the last two decades of research on face evaluation. Such evaluative judgments emerge early in development and show a surprising consistency over time and across cultures. Judgments of trustworthiness, in particular, are closely associated with general valence evaluation of faces and are grounded in resemblance to emotional expressions, signaling approach versus avoidance behaviors. Data-driven computational models have been critical for the discovery of the configurations of features, including resemblance to emotional expressions, driving specific judgments. However, almost all models are based on judgments aggregated across individuals, essentially masking idiosyncratic differences in judgments. Yet, recent research shows that most of the meaningful variance of complex judgments such as trustworthiness is idiosyncratic: explained not by stimulus features, but by participants and participants by stimuli interactions. Hence, to understand complex judgments, we need to develop methods for building models of judgments of individual participants. We describe one such method, combining the strengths of well-established methods with recent developments in machine learning.

How individuals view the world is critical to understanding human behavior. Yet, almost all research within perception and judgment has drawn inferences from group-level behavior, with little work focused on understanding how the individual perceives their world. However, for complex judgments (e.g., trustworthiness), most of the meaningful variance is due to factors specific to the individual. Here we showcase a data-driven reverse correlation method for visualizing any perceptually-derived stereotype at the individual level. We show that our method 1) produces photorealistic and reliable results related to a broad range of judgments, 2) produces valid, psychologically-aligned representations of what individuals are imagining “in their mind’s eye”, and 3) is capable of capturing visual representations sensitive enough to examine context-dependent categories (e.g., a trustworthy individual to babysit your children vs. to fix your car). Across all studies, we highlight the theoretical implications and utility of developing idiosyncratic models of visual perception.

Social stereotypes are prevalent and consequential, yet sometimes inaccurate. How do people learn these inaccurate beliefs in the first place and why do these beliefs persist in the face of counter evidence? Building on past research on cognitive limitations and environmental sample biases, we propose an integrative perspective: Insufficient statistical learning (Insta-learn). Instalearn posits that humans are active learners of the environment. Starting from a small sample, people are able to extract statistical patterns within the sample accurately and quickly. However, people do not continue sampling sufficiently. If they decide not to collect more samples once they are (prematurely) satisfied, inaccurate stereotypes can emerge even when more data would show otherwise. We investigated this hypothesis across six online experiments (N = 1565), using novel pairs of computer-generated faces and social behaviors. Fixing the population level statistics of face-behavior associations to zero and varying the initial sample statistics, we found that participants quickly learned the initial sample statistics (from as few as three examples) and persisted in using such spurious associations in their final decisions. Granting the sampling power to participants — samples were endogenously generated by participants and not defined by the experimenters — we found insufficient sampling caused spurious associations to persist. Insta-learn provides a domain-general framework for a mechanistic explanation of the emergence and persistence of social stereotypes.

U.S. immigration discourse has spurred interest in characterizing who illegalized immigrants are or perceived to be. What are the associated visual representations of migrant illegality? Across two studies with undergraduate and online samples (N=686), we used face-based reverse correlation and similarity sorting to capture and compare mental representations of illegalized immigrants, native-born U.S. citizens, and documented immigrants. Documentation statuses evoked racialized imagery. Immigrant representations were dark-skinned and perceived as non- white, while citizen representations were light-skinned, evaluated positively, and perceived as white. Legality further differentiated immigrant representations: documentation conjured trustworthy representations, illegality conjured threatening representations. Participants spontaneously sorted unlabeled faces by documentation status in a spatial arrangement task. Faces’ spatial similarity correlated with their similarity in pixel luminance and “American” ratings, confirming racialized distinctions. Representations of illegalized immigrants were uniquely racialized as dark-skinned un-American threats, reflecting how U.S. imperialism and colorism set conditions of possibility for existing representations of migrant illegalization.

Social impressions from faces have been studied in psychology for over 100 years. These impressions are rapid, efficient, and consequential. Yet, disagreements about the content of face impressions remain. Across two studies (N = 4,526), we develop a taxonomy of spontaneous face impressions content by applying novel interdisciplinary methods from Artificial Intelligence text analysis to thousands of free-response descriptions of computer-generated faces. We identify a taxonomy of face impression dimensions, and describe their coverage, prevalence, directionality, and correlational structure. We characterize a diverse and nuanced taxonomy of content that, when compared to just the content that dominant low-dimensional models focus on, increases the coverage of spontaneous responses from about 50% to almost 100%. Our results describe general patterns of prevalence, indicating that dimensions from low-dimensional models (e.g., Sociability, Morality, Assertiveness) are highly prevalent, but that alternative dimensions such as Uniqueness and Health, among others, are also significantly prevalent in face impressions of naturalistic face photographs. Most dimensions show a positivity bias, and the correlational structure of the dimensions further supports the clustering of low-dimensional model’s content as separate from the expanded taxonomy dimensions. Finally, this expanded taxonomy improves predictions of general evaluations and decision making in various real-world relevant contexts. The derived taxonomy of spontaneous face impressions content is a foundation for further theoretical development and practical applications in an area central to human behavior.

Recent work has shown that the idiosyncrasies of the observer can contribute more to the variance of social judgments of faces than the features of the faces. However, it is unclear what conditions determine the relative contributions of idiosyncratic and shared variance. Here, we examine three conditions: type of judgment, response scales, and diversity of face stimuli. First, we show that for simpler, directly observable judgments (e.g., masculinity) shared exceeds idiosyncratic variance, whereas for more complex, less directly observable judgments (e.g., trustworthiness) idiosyncratic exceeds shared variance. Second, dichotomous forced-choice responses (i.e., “yes”/”no”) resulted in greater shared variance compared to multi-point Likert- type responses. Third, we show that judgments of more diverse face images increase the amount of shared variance. Finally, using machine learning methods, we examine how stimulus (e.g., emotion resemblance, skin luminosity) and observer variables (e.g., race, age) contribute to shared and idiosyncratic variance of judgments. Overall, our results indicate that an observer's age is the most consistent and best predictor of idiosyncratic variance contributions to face judgments measured in the current research.

Trustworthy-looking faces are also perceived as more attractive, but are there other meaningful cues that contribute to perceived trustworthiness? Using data-driven models, we identify these cues after removing attractiveness cues. In Experiment 1, we show that both judgments of trustworthiness and attractiveness of faces manipulated by a model of perceived trustworthiness change in the same direction. To control for the effect of attractiveness, we build two new models of perceived trustworthiness: a subtraction model, which forces the perceived attractiveness and trustworthiness to be negatively correlated (Experiment 2), and an orthogonal model, which reduces their correlation (Experiment 3). In both experiments, faces manipulated to appear more trustworthy were indeed perceived to be more trustworthy, but not more attractive. Importantly, in both experiments, these faces were also perceived as more approachable and with more positive expressions, as indicated by both judgments and machine learning algorithms. The current studies show that the visual cues used for trustworthiness and attractiveness judgments can be separated, and that apparent approachability and facial emotion are driving trustworthiness judgments and possibly general valence evaluation.

When we look at someone's face, we rapidly and automatically form robust impressions of how trustworthy they appear. Yet while people's impressions of trustworthiness show a high degree of reliability and agreement with one another, evidence for the accuracy of these impressions is weak. How do such appearance-based biases survive in the face of weak evidence? We explored this question using an iterated learning paradigm, in which memories relating (perceived) facial and behavioral trustworthiness were passed through many generations of participants. Stimuli consisted of pairs of computer-generated people's faces and exact dollar amounts that those fictional people shared with partners in a trust game. Importantly, the faces were designed to vary considerably along a dimension of perceived facial trustworthiness. Each participant learned (and then reproduced from memory) some mapping between the faces and the dollar amounts shared (i.e., between perceived facial and behavioral trustworthiness). Much like in the game of 'telephone', their reproductions then became the training stimuli initially presented to the next participant, and so on for each transmission chain. Critically, the first participant in each chain observed some mapping between perceived facial and behavioral trustworthiness, including positive linear, negative linear, nonlinear, and completely random relationships. Strikingly, participants' reproductions of these relationships showed a pattern of convergence in which more trustworthy looks were associated with more trustworthy behavior - even when there was no relationship between looks and behavior at the start of the chain. These results demonstrate the power of facial stereotypes, and the ease with which they can be propagated to others, even in the absence of any reliable origin of these stereotypes.