AI Hyperrealism: Why AI Faces Are Perceived as More Real Than Human Ones.

Recent evidence shows that AI-generated faces are now indistinguishable from human faces. However, algorithms are trained disproportionately on White faces, and thus White AI faces may appear especially realistic. In Experiment 1 (N = 124 adults), alongside our reanalysis of previously published data, we showed that White AI faces are judged as human more often than actual human faces-a phenomenon we term AI hyperrealism. Paradoxically, people who made the most errors in this task were the most confident (a Dunning-Kruger effect). In Experiment 2 (N = 610 adults), we used face-space theory and participant qualitative reports to identify key facial attributes that distinguish AI from human faces but were misinterpreted by participants, leading to AI hyperrealism. However, the attributes permitted high accuracy using machine learning. These findings illustrate how psychological theory can inform understanding of AI outputs and provide direction for debiasing AI algorithms, thereby promoting the ethical use of AI.

Three's a crowd: Fast ensemble perception of first impressions of trustworthiness.

Trustworthiness impressions are fundamental social judgements with far-reaching consequences in many aspects of society, including criminal justice, leadership selection and partner preferences. Thus far, most research has focused on facial characteristics that make a face individually appear more or less trustworthy. However, in everyday life, faces are not always perceived in isolation but are often encountered in crowds. It has been proposed that we deal with the large amount of facial information in a group by extracting summary statistics of the crowd, a phenomenon called ensemble perception. Prior research showed that ensemble perception occurs for various facial features, such as emotional expression, facial identity, and attractiveness. Here, we investigated whether observers can integrate the level of trustworthiness from multiple faces to extract an average impression of the crowd. Across four studies, participants were presented with crowds of faces and were asked to report their average level of trustworthiness with an adjustment (Experiment 1) and a rating task (Experiments 2 and 3). Participants were able to extract an ensemble perception of trustworthiness impressions from multiple faces. Moreover, observers were able to form a summary statistic of trustworthiness impressions from a group of faces as quickly as 250 ms (Experiment 4). Taken together, these results demonstrate that ensemble perception can occur at the level of impressions of trustworthiness. Thus, these critical social judgements not only occur for individual faces but are also integrated into a unique ensemble impression of crowds. Our findings contribute to the development of a more ecological approach to the study of trust impressions, since they provide an understanding of trustworthiness judgements not only on an individual level, but on a much broader social group level. Furthermore, our results drive forward new theory because they demonstrate for the first time that ensemble representations cover a broad range of phenomena than previously recognized, including complex high-level facial trait judgements such as trustworthiness impressions.

A first impression of the future.

We offer a response to six commentaries on our target article 'Understanding trait impressions from faces'. A broad consensus emerged with authors emphasizing the importance of increasing the diversity of faces and participants, integrating research on impressions beyond the face, and continuing to develop methods needed for data-driven approaches. We propose future directions for the field based on these themes.

Children show neural sensitivity to facial trustworthiness as measured by fast periodic visual stimulation.

Adults exhibit neural responses over the visual occipito-temporal area in response to faces that vary in how trustworthy they appear. However, it is not yet known when a mature pattern of neural sensitivity can be seen in children. Using a fast periodic visual stimulation (FPVS) paradigm, face images were presented to 8-to-9-year-old children (an age group which shows development of trust impressions; N = 31) and adult (N = 33) participants at a rate of 6 Hz (6 face images per second). Within this sequence, an 'oddball' face differing in the level of facial trustworthiness compared to the other faces, was presented at a rate of 1 Hz (once per second). Children were sensitive to variations in facial trustworthiness, showing reliable and significant neural responses at 1 Hz in the absence of instructions to respond to facial trustworthiness. Additionally, the magnitude of children's and adults' neural responses was similar, with strong Bayesian evidence that implicit neural responses to facial trustworthiness did not differ across the groups, and therefore, that visual sensitivity to differences in facial trustworthiness can show mature patterns by this age. Thus, nine or less years of social experience, perceptual and/or cognitive development may be sufficient for adult-like neural sensitivity to facial trustworthiness to emerge. We also validate the use of the FPVS methodology to examine children's implicit face-based trust processing for the first time, which is especially valuable in developmental research because this paradigm requires no explicit instructions or responses from participants.

Trustworthiness perception is mandatory: Task instructions do not modulate fast periodic visual stimulation trustworthiness responses.

Although it is often assumed that humans spontaneously respond to the trustworthiness of others' faces, it is still unclear whether responses to facial trust are mandatory or can be modulated by instructions. Considerable scientific interest lies in understanding whether trust processing is mandatory, given the societal consequences of biased trusting behavior. We tested whether neural responses indexing trustworthiness discrimination depended on whether the task involved focusing on facial trustworthiness or not, using a fast periodic visual stimulation electroencephalography oddball paradigm with a neural marker of trustworthiness discrimination at 1 Hz. Participants judged faces on size without any reference to trust, explicitly formed impressions of facial trust, or were given a financial lending context that primed trust, without explicit trust judgement instructions. Significant trustworthiness discrimination responses at 1 Hz were found in all three conditions, demonstrating the robust nature of trustworthiness discrimination at the neural level. Moreover, no effect of task instruction was observed, with Bayesian analyses providing moderate to decisive evidence that task instruction did not affect trustworthiness discrimination. Our finding that visual trustworthiness discrimination is mandatory points to the remarkable spontaneity of trustworthiness processing, providing clues regarding why these often unreliable impressions are ubiquitous.

Children's dynamic use of face- and behavior-based cues in an economic trust game.

Who do children trust? We investigated the extent to which children use face-based versus behavior-based cues when deciding whom to trust in a multiturn economic trust game. Children's (N = 42; aged 8 to 10 years; 31 females; predominantly White) trust decisions were informed by an interaction between face-based and behavior-based cues to trustworthiness, similarly to those of adults (N = 41; aged 17 to 48 years; 23 females; predominantly White). Facial trustworthiness guided children's investment decisions initially, such that they invested highly with trustworthy-looking partners and less with untrustworthy-looking partners. However, by the end of the trust game, after children had experienced game partners' fair or unfair return behavior, they overcame this bias and instead used partners' previous behavior to guide their trust decisions. Using partners' return behavior to guide decisions was the most rational strategy, because partners' facial trustworthiness was not an accurate cue to their actual trustworthiness. This dynamic use of different cues to trustworthiness suggests sophisticated levels of social cognition in children, which may reflect the social importance of trust impressions. (PsycInfo Database Record (c) 2022 APA, all rights reserved).

Understanding trait impressions from faces.

Impressions from faces are made remarkably quickly and they can underpin behaviour in a wide variety of social contexts. Over the last decade many studies have sought to trace the links between facial cues and social perception and behaviour. One such body of work has shown clear overlap between the fields of face perception and social stereotyping by demonstrating a role for conceptual stereotypes in impression formation from faces. We integrate these results involving conceptual influences on impressions with another substantial body of research in visual cognition which demonstrates that much of the variance in impressions can be predicted from perceptual, data-driven models using physical cues in face images. We relate this discussion to the phylogenetic, cultural, individual and developmental origins of facial impressions and define priority research questions for the field including investigating non-WEIRD cultures, tracking the developmental trajectory of impressions and determining the malleability of impression formation.

Evidence for a kernel of truth in children's facial impressions of children's niceness, but not shyness.

Adults teach children not to "judge a book by its cover." However, adults make rapid judgments of character from a glance at a child's face. These impressions can be modestly accurate, suggesting that adults may be sensitive to valid signals of character in children's faces. However, it is not clear whether such sensitivity requires decades of social experience, in line with the development of other face-processing abilities (e.g., facial emotion recognition), or whether this sensitivity emerges relatively early, in childhood. An important theoretical question therefore, is whether or not children's impressions are at all accurate. Here, we examined the accuracy in children's impressions of niceness and shyness from children's faces. Children (aged 7-12 years, ∼90% Caucasian) and adults rated 84 unfamiliar children's faces (aged 4-11 years, 48 female, ∼80% Caucasian) for niceness (Study 1) or shyness (Study 2). To measure accuracy, we correlated facial impressions with parental responses to well-established questionnaires about the actual niceness/shyness of those children in the images. Overall, children and adults formed highly similar niceness (r = .94) and shyness (r = .84) impressions. Children also showed mature impression accuracy: Children and adults formed modestly accurate niceness impressions, across different images of the same child's face. Neither children nor adults showed evidence for accurate shyness impressions. Together, these results suggest that children's impressions are relatively mature by middle childhood. Furthermore, these results demonstrate that any mechanisms driving accurate niceness impressions are in place by 7 years, and potentially before. (PsycInfo Database Record (c) 2021 APA, all rights reserved).

Individual differences in trust evaluations are shaped mostly by environments, not genes.

People evaluate a stranger's trustworthiness from their facial features in a fraction of a second, despite common advice "not to judge a book by its cover." Evaluations of trustworthiness have critical and widespread social impact, predicting financial lending, mate selection, and even criminal justice outcomes. Consequently, understanding how people perceive trustworthiness from faces has been a major focus of scientific inquiry, and detailed models explain how consensus impressions of trustworthiness are driven by facial attributes. However, facial impression models do not consider variation between observers. Here, we develop a sensitive test of trustworthiness evaluation and use it to document substantial, stable individual differences in trustworthiness impressions. Via a twin study, we show that these individual differences are largely shaped by variation in personal experience, rather than genes or shared environments. Finally, using multivariate twin modeling, we show that variation in trustworthiness evaluation is specific, dissociating from other key facial evaluations of dominance and attractiveness. Our finding that variation in facial trustworthiness evaluation is driven mostly by personal experience represents a rare example of a core social perceptual capacity being predominantly shaped by a person's unique environment. Notably, it stands in sharp contrast to variation in facial recognition ability, which is driven mostly by genes. Our study provides insights into the development of the social brain, offers a different perspective on disagreement in trust in wider society, and motivates new research into the origins and potential malleability of face evaluation, a critical aspect of human social cognition.

An objective and reliable electrophysiological marker for implicit trustworthiness perception.

Trustworthiness is assumed to be processed implicitly from faces, despite the fact that the overwhelming majority of research has only involved explicit trustworthiness judgements. To answer the question whether or not trustworthiness processing can be implicit, we apply an electroencephalography fast periodic visual stimulation (FPVS) paradigm, where electrophysiological cortical activity is triggered in synchrony with facial trustworthiness cues, without explicit judgements. Face images were presented at 6 Hz, with facial trustworthiness varying at 1 Hz. Significant responses at 1 Hz were observed, indicating that differences in the trustworthiness of the faces were reflected in the neural signature. These responses were significantly reduced for inverted faces, suggesting that the results are associated with higher order face processing. The neural responses were reliable, and correlated with explicit trustworthiness judgements, suggesting that the technique is capable of picking up on stable individual differences in trustworthiness processing. By demonstrating neural activity associated with implicit trustworthiness judgements, our results contribute to resolving a key theoretical debate. Moreover, our data show that FPVS is a valuable tool to examine face processing at the individual level, with potential application in pre-verbal and clinical populations who struggle with verbalization, understanding or memory.

Appearance-based trust processing in schizophrenia.

OBJECTIVES: Schizophrenia is characterized by impaired social interactions and altered trust. In the general population, trust is often based on facial appearance, with limited validity but enormous social consequences. The aim was to examine trust processing in schizophrenia and specifically to examine how people with schizophrenia use facial appearance as well as actual partner fairness to guide trusting decisions. DESIGN: An experimental economic game study. METHODS: Here, we tested how patients with schizophrenia and control participants (each N = 24) use facial trustworthiness appearance and partner fairness behaviour to guide decisions in a multi-round Trust Game. In the Trust Game, participants lent money to 'partners' whose facial appearance was either untrustworthy or trustworthy, and who either played fairly or unfairly. Clinical symptoms were measured as well as explicit trustworthiness impressions. RESULTS: Overall, the patients with schizophrenia showed unimpaired explicit facial trustworthiness impressions and unimpaired facial appearance biases in the Trust Game. Crucially, patients and controls significantly differed so that the patients with schizophrenia did not learn to discriminate in the Trust Game based on actual partner fairness, unlike control participants. CONCLUSION: A failure to discriminate trust has important implications for everyday functioning in schizophrenia, as forming accurate trustworthiness beliefs is an essential social skill. Critically, without relying on more valid trust cues, people with schizophrenia may be especially susceptible to the misleading effect of appearance when making trusting decisions. PRACTITIONER POINTS: Findings People with schizophrenia made very similar facial trustworthiness impressions to healthy controls and also used facial appearance to guide trust decisions similarly to controls. However, the patient group were less able to explicitly distinguish between fair and unfair partners based on their behaviour compared with the control group. Moreover, people with schizophrenia failed to use actual partner fairness to guide their financial decisions in the Trust Game, unlike controls, and this impairment was specific to a social task. People with schizophrenia may be particularly reliant on facial appearance when trusting others, as they may struggle to incorporate more valid trustworthiness information in their decision-making, such as actual partner fairness.

Do facial first impressions reflect a shared social reality?

Influential facial impression models have repeatedly shown that trustworthiness, youthful-attractiveness, and dominance dimensions subserve a wide variety of first impressions formed from strangers' faces, suggestive of a shared social reality. However, these models are built from impressions aggregated across observers. Critically, recent work has now shown substantial inter-observer differences in facial impressions, raising the important question of whether these dimensional models based on aggregated group data are meaningful at the individual observer level. We addressed this question with a novel case series approach, using factor analyses of ratings of twelve different traits to build individual models of facial impressions for different observers. Strikingly, three dimensions of trustworthiness, youthful/attractiveness, and competence/dominance appeared across the majority of these individual observer models, demonstrating that the dimensional approach is indeed meaningful at the individual level. Nonetheless, we also found differences in the stability of the competence/dominance dimension across observers. Taken together, results suggest that individual differences in impressions arise in the context of a largely common structure that supports a shared social reality.

Best-worst scaling improves measurement of first impressions.

A common goal in psychological research is the measurement of subjective impressions, such as first impressions of faces. These impressions are commonly measured using Likert ratings. Although these ratings are simple to administer, they are associated with response issues that can limit reliability. Here we examine best-worst scaling (BWS), a forced-choice method, as a potential alternative to Likert ratings for measuring participants' facial first impressions. We find that at the group level, BWS scores correlated almost perfectly with Likert scores, indicating that the two methods measure the same impressions. However, at the individual participant level BWS outperforms Likert ratings, both in terms of ability to predict preferences in a third task, and in terms of test-retest reliability. These benefits highlight the power of BWS, particularly for use in individual differences research.

Sexual unfaithfulness can be judged with some accuracy from men's but not women's faces.

We routinely make judgements of trustworthiness from the faces of others. However, the accuracy of such judgements remains contentious. An important context for trustworthiness judgements is sexual unfaithfulness. Accuracy in sexual unfaithfulness judgements may be adaptive for avoiding reproductive costs associated with having an unfaithful partner. Indeed, emerging studies suggest that women, and to a lesser degree, men, show above-chance accuracy in judging sexual unfaithfulness from opposite-sex faces. In the context of mate guarding, it is important not only to assess the likelihood of a partner defecting, but also to detect same-sex poachers. Therefore, here, we examine whether individuals can also judge sexual unfaithfulness (self-reported cheating and poaching behaviour) from same-sex faces. We found above-chance accuracy in judgements of unfaithfulness from same-sex faces in men but not women. Conversely, we found above-chance accuracy for opposite-sex faces in women but not men. Therefore, both men and women showed above-chance accuracy, but only for men's, and not women's, faces. Raters were making accurate (above-chance) judgements of unfaithfulness from men's faces using facial masculinity, a well-established signal of propensity to adopt short-term mating strategies. In summary, we found above-chance accuracy in impressions of unfaithfulness from men's faces. Although very modest, the level of accuracy could nevertheless have biological significance as an evolved adaptation for identifying potential cheaters/poachers.

Testing the functional basis of first impressions: Dimensions for children's faces are not the same as for adults' faces.

Despite warnings not to "judge a book by its cover," people rapidly form facial impressions. In Oosterhof and Todorov's (2008) two-dimensional model of facial impressions, trustworthiness, and dominance underlie impressions and primarily function to signal the potential threat of others. Here, we test a key assumption of these models, namely that these dimensions are functional, by evaluating whether the adult-face dimensions apply to young children's faces. Although it may be functional for adults to judge adult faces on dimensions that signal threat, adults associate different social goals with children, and these goals are likely to impact the impressions adults make of such faces. Thus, a functional approach would predict that the dimensions for children's faces are not threat focused. In Studies 1 and 2, we build a data-driven model of Caucasian adults' impressions of Caucasian children's faces, finding evidence for two dimensions. The first dimension, niceness, is similar (although not identical) to the adult dimension of trustworthiness. However, we find a second dimension, shyness, that is clearly dissociable from dominance (Study 3), and critically, is not focused on threat. We demonstrate that adults are sensitive to subtle facial manipulations of these dimensions (Studies 4 and 5) and that these impressions impact adults' behavioral expectations of children (Study 6). Finally, we show that niceness and shyness dimensions generalize to an independent sample of ambient images, demonstrating their robustness (Study 7). Our results suggest that social goals have the power to drive functional impressions and highlight the flexibility of our visual system when forming such inferences. (PsycINFO Database Record (c) 2019 APA, all rights reserved).

Children show adult-like facial appearance biases when trusting others.

A large research literature details the powerful behavioral consequences that a trustworthy appearance can have on adult behavior. Surprisingly, few studies have investigated how these biases operate among children, despite the theoretical importance of understanding when these biases emerge in development. Here, we used an economic trust game to systematically investigate trust behavior in young children (5-8 years), older children (9-12 years), and adults. Participants played the game with child and adult "partners" that varied in emotional expression (mild displays of happiness and anger, and a neutral baseline), which is known to modulate perceived trustworthiness. Strikingly, both groups of children showed adult-like facial appearance biases when trusting others, with no "own-age bias." There were no developmental differences in the magnitude of this effect, which supports adult-like overgeneralization of these transient emotion cues into enduring trait impressions that guide interpersonal behavior from as early as 5 years of age. Irrespective of whether or not they were explicitly directed to do so, all participants modulated their behavior in line with the emotion cues: more generous and trusting with happy partners, followed by neutral, and then angry. These findings speak to the impressive sophistication of children's early social cognition and provide key insights into the causal mechanisms driving trait impressions, suggesting they are not necessarily contingent upon protracted social experience. (PsycINFO Database Record (c) 2019 APA, all rights reserved).

Should I trust you? Autistic traits predict reduced appearance-based trust decisions.

Facial impressions of trustworthiness guide social decisions in the general population, as shown by financial lending in economic Trust Games. As an exception, autistic boys fail to use facial impressions to guide trust decisions, despite forming typical facial trustworthiness impressions (Autism, 19, 2015a, 1002). Here, we tested whether this dissociation between forming and using facial impressions of trustworthiness extends to neurotypical men with high levels of autistic traits. Forty-six Caucasian men completed a multi-turn Trust Game, a facial trustworthiness impressions task, the Autism-Spectrum Quotient, and two Theory of Mind tasks. As hypothesized, participants' levels of autistic traits had no observed effect on the impressions formed, but negatively predicted the use of those impressions in trust decisions. Thus, the dissociation between forming and using facial impressions of trustworthiness extends to the broader autism phenotype. More broadly, our results identify autistic traits as an important source of individual variation in the use of facial impressions to guide behaviour. Interestingly, failure to use these impressions could potentially represent rational behaviour, given their limited validity.

Impressions of sexual unfaithfulness and their accuracy show a degree of universality.

Forming accurate impressions of others' trustworthiness is a critical social skill, with faithfulness representing a key aspect of trust in sexual relationships. Interestingly, there is evidence for a small degree of accuracy in facial impressions of sexual unfaithfulness. Theoretical accounts suggest that these impressions may function to help with partner selection, and may be universal. If so, impressions should be similar for perceivers from different cultures and accuracy should not be limited to own-race faces. We tested these predictions by asking Caucasian and Asian women to judge the likelihood of unfaithfulness from the faces of Caucasian males whose past sexual history was known. In two studies we found high cross-cultural agreement in these impressions, consistent with universality in the impressions themselves. In Study 1, we found an other-race effect in impression accuracy, with significantly less accurate cross-race impressions by Asian women than own-race impressions by Caucasian women. Asian women showed no accuracy. Interestingly, in Study 2, Asian women who had grown up in the West showed small but significant accuracy in their impressions, with no other-race effect. Results are consistent with a degree of universality in the accuracy of this important aspect of social perception, provided that perceivers have experience with the faces being assessed.