Familiar faces rendered strange: Why inconsistent realism drives characters into the uncanny valley

Discrepancies in perceived humanness between spatially filtered and unfiltered faces and their associations with uncanny feelings

Human and artificial features that coexist in certain types of human-like robots create a discrepancy in perceived humanness and evoke uncanny feelings in human observers. However, whether this perceptual mismatch in humanness occurs for all faces, and whether it is related to the uncanny feelings toward them, is unknown. We investigated this by examining perceived humanness for a variety of natural images of robot and human faces with different spatial frequency (SF) information: that is, faces with only low SF, middle SF, and high SF information, and intact (spatially unfiltered) faces. Uncanny feelings elicited by these faces were also measured. The results showed perceptual mismatches that LSF, MSF, and HSF faces were perceived as more human than intact faces. This was particularly true for intact robot faces that looked slightly human, which tended to evoke strong uncanny feelings. Importantly, the mismatch in perceived humanness between the intact and spatially filtered faces was positively correlated with uncanny feelings toward intact faces. Given that the human visual system performs SF analysis when processing faces, the perceptual mismatches observed in this study likely occur in real life for all faces, and as such might be a ubiquitous source of uncanny feelings in real-life situations.

A stimulus exposure of 50 ms elicits the uncanny valley effect

The uncanny valley (UV) effect captures the observation that artificial entities with near-human appearances tend to create feelings of eeriness. Researchers have proposed many hypotheses to explain the UV effect, but the visual processing mechanisms of the UV have yet to be fully understood. In the present study, we examined if the UV effect is as accessible in brief stimulus exposures compared to long stimulus exposures (Experiment 1). Forty-one participants, aged 21-31, rated each human-robot face presented for either a brief (50 ms) or long duration (3 s) in terms of attractiveness, eeriness, and humanness (UV indices) in a 7-point Likert scale. We found that brief and long exposures to stimuli generated a similar UV effect. This suggests that the UV effect is accessible at early visual processing. We then examined the effect of exposure duration on the categorisation of visual stimuli in Experiment 2. Thirty-three participants, aged 21-31, categorised faces as either human or robot in a two-alternative forced choice task. Their response accuracy and variance were recorded. We found that brief stimulus exposures generated significantly higher response variation and errors than the long exposure condition. This indicated that participants were more uncertain in categorising faces in the brief exposure condition due to insufficient time. Further comparisons between Experiment 1 and 2 revealed that the eeriest faces were not the hardest to categorise. Overall, these findings indicate (1) that both the UV effect and categorical uncertainty can be elicited through brief stimulus exposure, but (2) that categorical uncertainty is unlikely to cause the UV effect. These findings provide insights towards the perception of robotic faces and implications for the design of robots, androids, avatars, and artificial intelligence agents.

Asynchrony enhances uncanniness in human, android, and virtual dynamic facial expressions

OBJECTIVE

Uncanniness plays a vital role in interactions with humans and artificial agents. Previous studies have shown that uncanniness is caused by a higher sensitivity to deviation or atypicality in specialized categories, such as faces or facial expressions, marked by configural processing. We hypothesized that asynchrony, understood as a temporal deviation in facial expression, could cause uncanniness in the facial expression. We also hypothesized that the effect of asynchrony could be disrupted through inversion.

RESULTS

Sixty-four participants rated the uncanniness of synchronous or asynchronous dynamic face emotion expressions of human, android, or computer-generated (CG) actors, presented either upright or inverted. Asynchrony vs. synchrony expressions increased uncanniness for all upright expressions except for CG angry expressions. Inverted compared with upright presentations produced less evident asynchrony effects for human angry and android happy expressions. These results suggest that asynchrony can cause dynamic expressions to appear uncanny, which is related to configural processing but different across agents.

The inversion effect on the cubic humanness-uncanniness relation in humanlike agents

The uncanny valley describes the typically nonlinear relation between the esthetic appeal of artificial entities and their human likeness. The effect has been attributed to specialized (configural) processing that increases sensitivity to deviations from human norms. We investigate this effect in computer-generated, humanlike android and human faces using dynamic facial expressions. Angry and happy expressions with varying degrees of synchrony were presented upright and inverted and rated on their eeriness, strangeness, and human likeness. A sigmoidal function of human likeness and uncanniness ("uncanny slope") was found for upright expressions and a linear relation for inverted faces. While the function is not indicative of an uncanny valley, the results support the view that configural processing moderates the effect of human likeness on uncanniness and extend its role to dynamic facial expressions.

A database of heterogeneous faces for studying naturalistic expressions

Facial expressions are thought to be complex visual signals, critical for communication between social agents. Most prior work aimed at understanding how facial expressions are recognized has relied on stimulus databases featuring posed facial expressions, designed to represent putative emotional categories (such as 'happy' and 'angry'). Here we use an alternative selection strategy to develop the Wild Faces Database (WFD); a set of one thousand images capturing a diverse range of ambient facial behaviors from outside of the laboratory. We characterized the perceived emotional content in these images using a standard categorization task in which participants were asked to classify the apparent facial expression in each image. In addition, participants were asked to indicate the intensity and genuineness of each expression. While modal scores indicate that the WFD captures a range of different emotional expressions, in comparing the WFD to images taken from other, more conventional databases, we found that participants responded more variably and less specifically to the wild-type faces, perhaps indicating that natural expressions are more multiplexed than a categorical model would predict. We argue that this variability can be employed to explore latent dimensions in our mental representation of facial expressions. Further, images in the WFD were rated as less intense and more genuine than images taken from other databases, suggesting a greater degree of authenticity among WFD images. The strong positive correlation between intensity and genuineness scores demonstrating that even the high arousal states captured in the WFD were perceived as authentic. Collectively, these findings highlight the potential utility of the WFD as a new resource for bridging the gap between the laboratory and real world in studies of expression recognition.

The deviation-from-familiarity effect: Expertise increases uncanniness of deviating exemplars

Humanlike entities deviating from the norm of human appearance are perceived as strange or uncanny. Explanations for the eeriness of deviating humanlike entities include ideas specific to human or animal stimuli like mate selection, avoidance of threat or disease, or dehumanization; however, deviation from highly familiar categories may provide a better explanation. Here it is tested whether experts and novices in a novel (greeble) category show different patterns of abnormality, attractiveness, and uncanniness responses to distorted and averaged greebles. Greeble-trained participants assessed the abnormality, attractiveness, uncanniness of normal, averaged, and distorted greebles and their responses were compared to participants who had not previously seen greebles. The data show that distorted greebles were more uncanny than normal greebles only in the training condition, and distorted greebles were more uncanny in the training compared to the control condition. In addition, averaged greebles were not more attractive than normal greebles regardless of condition. The results suggest uncanniness is elicited by deviations from stimulus categories of expertise rather than being a purely biological human- or animal-specific response.

Monkey visual attention does not fall into the uncanny valley

Very humanlike artificial agents can induce feelings of uneasiness in human perceivers. Stimuli that generate this response are said to occupy "the uncanny valley". Given inconsistent findings in the literature, whether or not nonhuman animals experience the uncanny valley is unclear. Here, we recorded the visual attention of eleven male rhesus monkeys as they viewed faces varying in realness across five levels, with visual attention measured by both number and duration of visual fixations on faces as a whole and on areas of interest within the faces (e.g., eyes, mouth). Face stimuli varied in terms of the realism of the image and behavior depicted by the face (lipsmack, threat, bared teeth, and neutral). We largely found no support that rhesus monkeys perceive an uncanny valley when viewing our stimuli; however, monkeys did generally pay more attention to eyes and less attention to mouths in real images compared to less realistic images. Across all stimuli, monkeys' visual attention was drawn to the mouths of images when teeth were visible. These findings suggest that rhesus monkeys in our study did not display an uncanny valley effect when viewing realistic stimuli but did percieve affective information depicted by faces regardless of how real those faces appear.

Familiarity, orientation, and realism increase face uncanniness by sensitizing to facial distortions

The uncanny valley predicts aversive reactions toward near-humanlike entities. Greater uncanniness is elicited by distortions in realistic than unrealistic faces, possibly due to familiarity. Experiment 1 investigated how familiarity and inversion affect uncanniness of facial distortions and the ability to detect differences between the distorted variants of the same face (distortion sensitivity). Familiar or unfamiliar celebrity faces were incrementally distorted and presented either upright or inverted. Uncanniness ratings increased across the distortion levels, and were stronger for familiar and upright faces. Distortion sensitivity increased with increasing distortion difference levels, again stronger for familiar and upright faces. Experiment 2 investigated how face realism, familiarity, and face orientation interacted for the increase of uncanniness across distortions. Realism increased the increase of uncanniness across the distortion levels, further enhanced by upright orientation and familiarity. The findings show that familiarity, upright orientation, and high face realism increase the sensitivity of uncanniness, likely by increasing distortion sensitivity. Finally, a moderated linear function of face realism and deviation level could explain the uncanniness of stimuli better than a quadratic function. A re-interpretation of the uncanny valley as sensitivity toward deviations from familiarized patterns is discussed.

Aspects of visual avatar appearance: self-representation, display type, and uncanny valley

The visual representation of human-like entities in virtual worlds is becoming a very important aspect as virtual reality becomes more and more "social". The visual representation of a character's resemblance to a real person and the emotional response to it, as well as the expectations raised, have been a topic of discussion for several decades and have been debated by scientists from different disciplines. But as with any new technology, the findings may need to be reevaluated and adapted to new modalities. In this context, we make two contributions which may have implications for how avatars should be represented in social virtual reality applications. First, we determine how default and customized characters of current social virtual reality platforms appear in terms of human likeness, eeriness, and likability, and whether there is a clear resemblance to a given person. It can be concluded that the investigated platforms vary strongly in their representation of avatars. Common to all is that a clear resemblance does not exist. Second, we show that the uncanny valley effect is also present in head-mounted displays, but-compared to 2D monitors-even more pronounced.

Creepy cats and strange high houses: Support for configural processing in testing predictions of nine uncanny valley theories

In 1970, Masahiro Mori proposed the uncanny valley (UV), a region in a human-likeness continuum where an entity risks eliciting a cold, eerie, repellent feeling. Recent studies have shown that this feeling can be elicited by entities modeled not only on humans but also nonhuman animals. The perceptual and cognitive mechanisms underlying the UV effect are not well understood, although many theories have been proposed to explain them. To test the predictions of nine classes of theories, a within-subjects experiment was conducted with 136 participants. The theories' predictions were compared with ratings of 10 classes of stimuli on eeriness and coldness indices. One type of theory, configural processing, predicted eight out of nine significant effects. Atypicality, in its extended form, in which the uncanny valley effect is amplified by the stimulus appearing more human, also predicted eight. Threat avoidance predicted seven; atypicality, perceptual mismatch, and mismatch+ predicted six; category+, novelty avoidance, mate selection, and psychopathy avoidance predicted five; and category uncertainty predicted three. Empathy's main prediction was not supported. Given that the number of significant effects predicted depends partly on our choice of hypotheses, a detailed consideration of each result is advised. We do, however, note the methodological value of examining many competing theories in the same experiment.

Virtual Reality Immersion Rescales Regulation of Interpersonal Distance in Controls but not in Autism Spectrum Disorder

Interpersonal distance (IPD) is a simple social regulation metric which is altered in autism. We performed a stop-distance paradigm to evaluate IPD regulation in autism spectrum disorder (ASD) and control groups in a real versus a virtual environment mimicking in detail the real one. We found a bimodal pattern of IPDs only in ASD. Both groups showed high IPD correlations between real and virtual environments, but the significantly larger slope in the control group suggests rescaling, which was absent in ASD. We argue that loss of nuances like non-verbal communication, such as perception of subtle body gestures in the virtual environment, lead to changed regulation of IPD in controls, whilst ASD participants show similar deficits in perceiving such subtle cues in both environments.

When Agents Become Partners: A Review of the Role the Implicit Plays in the Interaction with Artificial Social Agents

The way we interact with computers has significantly changed over recent decades. However, interaction with computers still falls behind human to human interaction in terms of seamlessness, effortlessness, and satisfaction. We argue that simultaneously using verbal, nonverbal, explicit, implicit, intentional, and unintentional communication channels addresses these three aspects of the interaction process. To better understand what has been done in the field of Human Computer Interaction (HCI) in terms of incorporating the type channels mentioned above, we reviewed the literature on implicit nonverbal interaction with a specific emphasis on the interaction between humans on the one side, and robot and virtual humans on the other side. These Artificial Social Agents (ASA) are increasingly used as advanced tools for solving not only physical but also social tasks. In the literature review, we identify domains of interaction between humans and artificial social agents that have shown exponential growth over the years. The review highlights the value of incorporating implicit interaction capabilities in Human Agent Interaction (HAI) which we believe will lead to satisfying human and artificial social agent team performance. We conclude the article by presenting a case study of a system that harnesses subtle nonverbal, implicit interaction to increase the state of relaxation in users. This “Virtual Human Breathing Relaxation System” works on the principle of physiological synchronisation between a human and a virtual, computer-generated human. The active entrainment concept behind the relaxation system is generic and can be applied to other human agent interaction domains of implicit physiology-based interaction.

Computer-Controlled Virtual Humans in Patient-Facing Systems: Systematic Review and Meta-Analysis

BACKGROUND

Virtual humans (VH) are computer-generated characters that appear humanlike and simulate face-to-face conversations using verbal and nonverbal cues. Unlike formless conversational agents, like smart speakers or chatbots, VH bring together the capabilities of both a conversational agent and an interactive avatar (computer-represented digital characters). Although their use in patient-facing systems has garnered substantial interest, it is unknown to what extent VH are effective in health applications.

OBJECTIVE

The purpose of this review was to examine the effectiveness of VH in patient-facing systems. The design and implementation characteristics of these systems were also examined.

METHODS

Electronic bibliographic databases were searched for peer-reviewed articles with relevant key terms. Studies were included in the systematic review if they designed or evaluated VH in patient-facing systems. Of the included studies, studies that used a randomized controlled trial to evaluate VH were included in the meta-analysis; they were then summarized using the PICOTS framework (population, intervention, comparison group, outcomes, time frame, setting). Summary effect sizes, using random-effects models, were calculated, and the risk of bias was assessed.

RESULTS

Among the 8,125 unique records identified, 53 articles describing 33 unique systems, were qualitatively, systematically reviewed. Two distinct design categories emerged - simple VH and VH augmented with health sensors and trackers. Of the 53 articles, 16 (26 studies) with 44 primary and 22 secondary outcomes were included in the meta-analysis. Meta-analysis of the 44 primary outcome measures revealed a significant difference between intervention and control conditions, favoring the VH intervention (SMD = .166, 95% CI .039-.292, P=.012), but with evidence of some heterogeneity, I2=49.3%. There were more cross-sectional (k=15) than longitudinal studies (k=11). The intervention was delivered using a personal computer in most studies (k=18), followed by a tablet (k=4), mobile kiosk (k=2), head-mounted display (k=1), and a desktop computer in a community center (k=1).

CONCLUSIONS

We offer evidence for the efficacy of VH in patient-facing systems. Considering that studies included different population and outcome types, more focused analysis is needed in the future. Future studies also need to identify what features of virtual human interventions contribute toward their effectiveness.

Swiping right: face perception in the age of Tinder

With an estimated 50 million or more users worldwide, Tinder has become one of the most popular mobile dating applications. Although judgments of physical attractiveness are assumed to drive the "swiping" decisions that lead to matches, we propose that there is an additional evaluative dimension driving behind these decisions: judgments of moral character. With the aim of adding empirical support for this proposition, we critically review the most striking findings about first impressions extracted from faces, moral character in person perception, creepiness, and the uncanny valley, as they apply to Tinder behavior. Drawing on this research and the evolutionary theory of biological markets, we formulate several hypotheses that offer directions for future studies of Tinder and other dating apps. We conclude that research on face perception of novel targets supports the plausibility of moral character as a potential factor affecting the swiping decisions and subsequent behavior of Tinder users.

Evaluation of the uncanny valley hypothesis based on declared emotional response and psychophysiological reaction

The uncanny valley (UV) hypothesis suggests that the observation of almost human-like characters causes an increase of discomfort. We conducted a study using self-report questionnaire, response time measurement, and electrodermal activity (EDA) evaluation. In the study, 12 computer-generated characters (robots, androids, animated, and human characters) were presented to 33 people (17 women) to (1) test the effect of a background context on the perception of characters, (2) establish whether there is a relation between declared feelings and physiological arousal, and (3) detect the valley of the presented stimuli. The findings provide support for reverse relation between human-likeness and the arousal (EDA). Furthermore, a positive correlation between EDA and human-likeness appraisal reaction time upholds one of the most common explanations of the UV – the categorization ambiguity. The absence of the significant relationship between declared comfort and EDA advocates the necessity of physiological measures for UV studies.

The doctor's digital double: how warmth, competence, and animation promote adherence intention

BACKGROUND

Each year, patient nonadherence to treatment advice costs the US healthcare system more than $300 billion and results in 250,000 deaths. Developing virtual consultations to promote adherence could improve public health while cutting healthcare costs and usage. However, inconsistencies in the realism of computer-animated humans may cause them to appear eerie, a phenomenon termed the uncanny valley. Eeriness could reduce a virtual doctor's credibility and patients' adherence.

METHODS

In a 2 × 2 × 2 between-groups posttest-only experiment, 738 participants played the role of a patient in a hypothetical virtual consultation with a doctor. The consultation varied in the doctor's Character (good or poor bedside manner), Outcome (received a fellowship or sued for malpractice), and Depiction (a recorded video of a real human actor or of his 3D computer-animated double). Character, Outcome, and Depiction were designed to manipulate the doctor's level of warmth, competence, and realism, respectively.

RESULTS

Warmth and competence increased adherence intention and consultation enjoyment, but realism did not. On the contrary, the computer-animated doctor increased adherence intention and consultation enjoyment significantly more than the doctor portrayed by a human actor. We propose that enjoyment of the animated consultation caused the doctor to appear warmer and more real, compensating for his realism inconsistency. Expressed as a path model, this explanation fit the data.

DISCUSSION

The acceptance and effectiveness of the animation should encourage the development of virtual consultations, which have advantages over creating content with human actors including ease of scenario revision, internationalization, localization, personalization, and web distribution.

The uncanny valley effect in typically developing children and its absence in children with autism spectrum disorders

Robots and virtual reality are gaining popularity in the intervention of children with autism spectrum disorder (ASD). To shed light on children’s attitudes towards robots and characters in virtual reality, this study aims to examine whether children with ASD show the uncanny valley effect. We varied the realism of facial appearance by morphing a cartoon face into a human face, and induced perceptual mismatch by enlarging the eyes, which has previously been shown as an effective method to induce the uncanny valley effect in adults. Children with ASD and typically developing (TD) children participated in a two-alternative forced choice task that asked them to choose one they liked more from the two images presented on the screen. We found that TD children showed the effect, i.e., the enlargement of eye size and the approaching realism reduced their preference. In contrast, children with ASD did not show the uncanny valley effect. Our findings in TD children help resolve the controversy in the literature about the existence of the uncanny valley effect among young children. Meanwhile, the absence of the uncanny valley effect in children with ASD might be attributed to their reduced sensitivity to subtle changes of face features and their limited visual experience to faces caused by diminished social motivation. Last, our findings provide practical implications for designing robots and virtual characters for the intervention of children with ASD.

Understanding the Uncanny: Both Atypical Features and Category Ambiguity Provoke Aversion toward Humanlike Robots

Robots intended for social contexts are often designed with explicit humanlike attributes in order to facilitate their reception by (and communication with) people. However, observation of an “uncanny valley”—a phenomenon in which highly humanlike entities provoke aversion in human observers—has lead some to caution against this practice. Both of these contrasting perspectives on the anthropomorphic design of social robots find some support in empirical investigations to date. Yet, owing to outstanding empirical limitations and theoretical disputes, the uncanny valley and its implications for human-robot interaction remains poorly understood. We thus explored the relationship between human similarity and people's aversion toward humanlike robots via manipulation of the agents' appearances. To that end, we employed a picture-viewing task (N_agents = 60) to conduct an experimental test (N_participants = 72) of the uncanny valley's existence and the visual features that cause certain humanlike robots to be unnerving. Across the levels of human similarity, we further manipulated agent appearance on two dimensions, typicality (prototypic, atypical, and ambiguous) and agent identity (robot, person), and measured participants' aversion using both subjective and behavioral indices. Our findings were as follows: (1) Further substantiating its existence, the data show a clear and consistent uncanny valley in the current design space of humanoid robots. (2) Both category ambiguity, and more so, atypicalities provoke aversive responding, thus shedding light on the visual factors that drive people's discomfort. (3) Use of the Negative Attitudes toward Robots Scale did not reveal any significant relationships between people's pre-existing attitudes toward humanlike robots and their aversive responding—suggesting positive exposure and/or additional experience with robots is unlikely to affect the occurrence of an uncanny valley effect in humanoid robotics. This work furthers our understanding of both the uncanny valley, as well as the visual factors that contribute to an agent's uncanniness.

Categorization-based stranger avoidance does not explain the uncanny valley effect

The uncanny valley hypothesis predicts that an entity appearing almost human risks eliciting cold, eerie feelings in viewers. Categorization-based stranger avoidance theory identifies the cause of this feeling as categorizing the entity into a novel category. This explanation is doubtful because stranger is not a novel category in adults; infants do not avoid strangers while the category stranger remains novel; infants old enough to fear strangers prefer photographs of strangers to those more closely resembling a familiar person; and the uncanny valley's characteristic eeriness is seldom felt when meeting strangers. We repeated our original experiment with a more realistic 3D computer model and found no support for categorization-based stranger avoidance theory. By contrast, realism inconsistency theory explains cold, eerie feelings elicited by transitions between instances of two different, mutually exclusive categories, given that at least one category is anthropomorphic: Cold, eerie feelings are caused by prediction error from perceiving some features as features of the first category and other features as features of the second category. In principle, realism inconsistency theory can explain not only negative evaluations of transitions between real and computer modeled humans but also between different vertebrate species.