The dot-perspective paradigm was created by Samson et al. (2010) as an experimental paradigm for measuring L1VPT processing. The essential idea behind the task is that if adults can implicitly track another person's visual perspective, then participants should do so even when they do not need to. In the paradigm, disc(s) were presented on the left- or right-side wall of a room with a computer-generated human avatar standing in the center of the room and facing to one side of the walls. Two kinds of visual perspectives—“You see N” on “Self-perspective” trials (“N” ranges from 0 to 3 dots), “S/he sees N” on “Other-perspective” trials—were presented before the scene. The participants were required to judge whether the picture matched the given perspective, leading to matching and mismatching trials. On half of the trials (Congruent condition), the avatar and the participant could see the same disc(s). On the remaining trials (Incongruent condition), the participant could see the disc(s) that were invisible to the avatar.

Consistent with previous studies of egocentric biases (i.e., strong biases toward the participants' own perspectives on other-perspective judgements) (e.g., Birch and Bloom, 2007), findings indicated egocentric intrusions when the participants were instructed to take the avatar's visual perspective. For instance, Samson et al. (2010) found slower response times and lower accuracy in incongruent condition compared to congruent condition (i.e., Consistency effect) on other-perspective trials. Additionally, the novel and key finding of the study was that the participants made more errors and responded more slowly in inconsistent condition compared to consistent ones when making self-perspective judgements. The finding suggested that adults could rapidly and effortlessly take the avatar's visual perspective even when not required to do so. The researchers explained the result as an effect of altercentric interference/intrusion, namely, the participant's own visual perspective was interfered by the implicit computation of the avatar's visual perspective that was task-irrelevant (Samson et al., 2010).

The consistency effect on self-perspective trials, considered as an effect of altercentric intrusion in L1VPT processing, has been replicated and extended. Further, the effect has even persisted under secondary task conditions where cognitive-resource tasks are added. In the original study, Samson et al. (2010) firstly observed that adults performed more slowly and made more errors on inconsistent trials compared to consistent trials when they were asked to take their own perspectives (i.e., the consistency effect on self-perspective trials). Later, a line of L1VPT-related experiments using the dot-perspective task has generalized the consistency effect under self-judgement circumstances. In addition to Surtees et al.'s (2016) replication of Samson et al.'s (2010) findings, Surtees and Apperly (2012) extended the altercentric intrusion effect from adults to 6–10-years-old children. More importantly, the effect persisted when considering its relationship with cognitive load. The consistency effects remained when participants judged their own perspectives regardless of time pressure for responses (i.e., shorter-deadline of 600 ms compared with a long-deadline of 1,200 ms) (Todd et al., 2017), and also persisted when the dot-perspective task was performed together with a secondary task requiring executive-function resources (Qureshi et al., 2010). The two studies, then, demonstrated that tasks requiring cognitive resources did not influence the elicitation of efficient computation of others' visual perspectives on self-perspective judgements.

Recent work has suggested that the processing of others' minds depends on two cognitive systems. One is a flexible system that enables us to explicitly reason about how others' mental states (beliefs, intentions, emotions, etc.) influence their behaviors. The other is an efficient system that allows us to automatically track others' mental states (Kovács et al., 2010; Low et al., 2016; Schneider et al., 2014). Samson et al. (2010) suggested that the efficient system enables us to automatically track what someone else sees (i.e., L1VPT). However, there is obvious debate over whether and to what extent the efficient system is specialized (implicit and automatic mentalizing account: e.g., Apperly, 2010) or domain-general (submentalizing account, e.g., Heyes, 2014). Specifically, the debate was sparked by the adults' performance in the dot-perspective task of L1VPT. That is, implicit mentalizing account claims that the consistency effect on self-perspective trials is elicited by implicit and effortless computation of the avatar's visual perspective via connecting her/his line-of-sight with the discs. The submentalizing account, alternatively, holds the view that the effect is evoked by merely attentional orienting produced by the directional but not agentive features of the avatar (e.g., head and/or body directions).

Researchers supporting the implicit mentalizing account claim that the consistency effect on the self-perspective judgement is invoked by implicit computation of others' mental state of seeing. Specifically, if participants could easily understand others' visual information, then they would rapidly and efficiently track the avatar's visual perspective in the dot-perspective task even without the explicit judgements of others' perspectives.

With respect to the implicit mentalizing account, researchers regard eye gaze as the key factor, which can be supported by the evidence showing the important role of eye gaze in mentalizing-related processes. For example, Baron-Cohen et al. (1995) observed that 3 and 4-year-old children can infer a person's mental state of wanting a chocolate bar (i.e., desires) by tracking others' eye-gaze direction to the target. More importantly, some researchers have found a strong relationship between the eye gaze and L1VPT processing (Sodian et al., 2007). They reported that 14-month-old infants can make expectations about an agent's goal-directed action based on understanding whether or not the line of sight between the agent's eyes and an object is physically unblocked (i.e., L1VPT ability). Based on these studies, eye gaze can convey information about others' visual perspectives. Therefore, manipulation of visual access has been created to measure L1VPT processing tapped into the mentalizing-related process.

Several studies have manipulated the gazer's line-of-sight as a novel way to clarify the mentalizing vs. submentalizing debate. Among those studies, only one has found evidence for implicit mentalizing (Furlanetto et al., 2016). Specifically, Furlanetto et al. (2016) manipulated the avatar's visibility to explore the debate by adopting transparent goggles (i.e., visible condition) and opaque goggles (i.e., invisible condition). The participants were checked to be able to associate different colored goggles with the corresponding avatar's ability to see (i.e., in the seeing condition, the red goggles worn by the avatar were transparent, whereas in the non-seeing condition, the orange goggles worn by the avatar were opaque). The participants were instructed to judge their own perspectives or perspectives of the avatar wearing the different colored goggles. The authors found that participants judged their own perspectives more slowly and less accurately in the inconsistent condition compared with the consistent condition, but the consistency effect was present in the seeing but not non-seeing conditions. The explanation of the discrepancy was that participants had different beliefs of the avatar's epistemic state of seeing via understanding the transparent and opaque features of the goggles, which then lead to the connection of the gazer's line-of-sight with the disc(s) on the wall(s) in the seeing condition but the disconnection in the non-seeing condition.

The findings of this study cast doubt on the submentalizing account claiming the role of directional information as it would predict the consistency effects in both visible and invisible conditions due to the identical directional features. Thus, participants can implicitly and efficiently compute the visual perspective of the avatar wearing transparent goggles even when they were not required to do so, which lent support to the implicit mentalizing account. However, the study is limited on its own. Specifically, the study cannot rule out the carry-over effect between self- and other-perspective conditions as the two conditions presented in the intermixed block. Therefore, the consistency effect on self-perspective judgements may be contaminated by explicit judgements about others' perspectives. To explore whether L1VPT processing is implicit mentalizing, it would be better to separate seeing condition from non-seeing condition.

Researchers supporting the submentalizing account have claimed that the consistency effect is elicited by domain-general mechanisms that are not specialized for processing of others' minds (e.g., attentional orienting, Heyes, 2014). Specifically, it is the directional features of the avatar that modulate participants' attentional shifts toward the number of dots on one side of the room. Therefore, on consistent trials of the dot-perspective task, the directional property of the centrally presented avatar oriented participants' attention toward the dot(s) on the target wall; whereas on inconsistent trials of the dot-perspective task, the directional property of the avatar oriented participants' attention neither to the dot(s) on the target wall nor to all the dots on both targeted walls. Then, the directional property of the centrally presented avatar may trigger more errors and slower response times in the inconsistent condition compared to the consistent condition. Related studies that tried to cast light on the debate by adding arrows as control stimuli relative to avatars and by manipulating the agent's line of sight via opaque barriers will be described in the next two subsections.

Some researchers have attempted to dissociate the mentalizing account from the submentalizing account by contrasting relevant paradigms. Gardner et al. (2018) attempted to dissociate the competing accounts by contrasting effects in modified dot-perspective task with the Posner task. The Posner task has been widely used to measure attentional orienting, in which a cue is first presented, followed by a target with a stimulus onset asynchrony (SOA), and then, participants are required to detect the target location. In Experiment 1, Gardner et al. examined if reflexive attention orienting can sufficiently induce the self-perspective-related consistency effect that would be considered as automatic visual perspective-taking in the dot-perspective task. They adopted a revised dot-perspective task by eliminating the “YOU perspective” instruction from the original Samson et al. (2010) study. The novel dot-perspective task made the participants unaware that they were completing a perspective-taking task. Thus, in Experiment 1, removal of the “YOU perspective” instruction resulted in a non-significant consistency effect, demonstrating that the effect cannot be evoked merely by reflexive attention orienting. In Experiment 2, they used the Posner task with dot-perspective-task's stimuli to investigate whether the attentional orienting property of the avatar contributed to the consistency effect that was previously induced in the dot-perspective task. They found the cue-validity effect only for longer SOAs. Specifically, adults were faster to detect a target when the avatar was directed to the target (valid trials) compared to when the avatar was directed away from the target (invalid trials) when SOA was 600 ms but not 100 ms or 300 ms. The findings demonstrated that voluntary rather than reflexive attention shift contributed to the consistency effect in the dot-perspective task. Taken together, the consistency effect in the classic dot-perspective task might be less automatic than first reported. Nonetheless, the discrepancy between visual-perspective-taking and attention-shift processes cannot be directly distinguished.

Previous findings revealed that compared to stance-maintained avatars (i.e., avatar's head and torso faced to the same wall), stance-averted avatars (i.e., avatar's head was oriented to one wall whereas the torso faced to the participant) induced an increased attentional orienting effect (e.g., Hietanen, 2002). Accordingly, Gardner et al. (2018) hypothesized that avatar-stance may modulate attentional orienting but not visual perspective-taking. They attempted to distinguish the implicit mentalizing from the submentalizing accounts by manipulating avatar stance (stance-averted vs. stance-maintained). Specifically, they explored whether avatar-stance could differently modulate the effect from visual-perspective-taking tasks (i.e., consistency effect in the dot-perspective task) and from attentional-orienting tasks (i.e., cue-validity effect in Posner task). Experiment 1 used the Posner cueing task to examine the cue-validity effect, finding that the target was more slowly to be detected in the invalid condition compared to the valid condition. The attentional orienting effect was modulated by avatar stance, which is reflected by the significant effect for stance-averted rather than for stance-maintained avatars. Experiment 2 adopted the dot-perspective task to replicate the classic consistency effect. More importantly, avatar-stance did not moderate the magnitude of the consistency effect in the classic visual-perspective-taking task. Accordingly, the dissociation between attentional orienting and visual-perspective-taking processes casts doubt on the submentalizing hypothesis regarding the role of the directional cue but supports the implicit mentalizing hypothesis.

The implicit mentalizing vs. submentalizing debate has important methodological and theoretical implications. In the methodological aspect, the debate challenges the effectiveness of the dot-perspective paradigm as a measure of L1VPT ability. It is important to find a universally-recognized way to measure and clarify L1VPT processing, which can lay a foundation for connecting L1VPT with the later more complex ToM processes and understanding the related social communications. Theoretically, the debate raises the question about the efficient part of ToM processing system (i.e., whether people can effortlessly track others' mental states, e.g., Meert et al., 2017). Practically, resolution of the debate is important because it is beneficial to further, understand related dysfunction in social behaviors in atypical individuals. For example, psychopathic patients have been found to have deficits in L1VPT ability, and their dysfunction in effortlessly taking others' visual perspective have been demonstrated to be correlated with their callous and criminal behaviors in real-world (Drayton et al., 2018; Baskin-Sommers and Brazil, 2022). Furthermore, a recent review (Capozzi and Ristic, 2020) has demonstrated social orienting attribute to the integration of the attribution of mental states and the manipulation of domain-general attentional processes. However, whether it is the case needs further investigations.

In addition to Mori's original hypotheses about the UV effect, there are several potential explanations for the UV effect from a cognitive mechanism perspective. One view suggests that category uncertainty during identification results in the UV effect (Jentsch, 1906/1997; MacDorman and Ishiguro, 2006; Pollick, 2010). The idea is that when presented with stimuli sharing human-like features, participants will attempt to classify the stimuli into human and non-human, or inanimate and animate categories; and observers' lack of certainty about what an entity is (e.g., entities that are positioned at the category boundary), is speculated to give rise to stronger negative affective response (MacDorman and Ishiguro, 2006; Pollick, 2010; Kätsyri et al., 2015). Moreover, based on Cheetham et al.'s (2014) claims, Kätsyri et al. (2015) proposed that within a categorical perception framework, the UV hypotheses suggested that increased perceptual discrimination challenges for neighboring character pairings would be linked to greater negative emotional responses. However, increasing an entity's overall category uncertainty does not heighten cold, eerie feelings. In fact, MacDorman and Chattopadhyay (2016) found that in a categorization task on animacy (living vs. inanimate) and realism (computer animated vs. real) indices, the eeriest and coldest stimuli were those categorized with the most certainty. Despite that the fact category uncertainty theory has been criticized, there are a wide range of theories proposed for the UV, many of which have not received as much negative criticism as category uncertainty.

According to mind perception hypothesis, humanoid robots can be eerie as their realism leads people to ascribe to them the abilities of feelings and sensations. Yet, these abilities are unlikely to develop in robots (Gray and Wegner, 2012). Expectation violation theory expands the mind perception hypothesis by positing that people expect humanoid robots with human-like appearances to behave like humans (i.e., human-directed expectations). However, the humanoid robots frequently violate the above expectations by, for instance, moving mechanically (Wang et al., 2015), resulting in negative emotional responses and feelings of eeriness and coldness (Broadbent, 2017). Furthermore, Saygin et al. (2012) interpreted expectation violation theory as prediction errors for viewed actions in predictive coding that may be triggered by a real human moving like robots (e.g., a performance artist painted himself/herself gold, stood in front of a cathedral and moved like a robot) (Saygin et al., 2012).

Apart from some cognitive theories regarding the UV effect, several theories can explain the UV effect from a perceptual viewpoint (i.e., the UV effect is regarded as a stimulus-driven effect). Moore (2012) explains the perceptual mismatch hypothesis from a mathematical perspective, namely, perceptual distortion caused by conflicting cues supplied by a Bayesian model may result in negative, fearful or even violent responses (Moore, 2012). Moreover, an explanation based on realism inconsistency between the human-likeness levels of specific sensory cues would suggest that, high skin realism, for instance, elicits neurocognitive expectancies of high eye realism (but not vice versa), and inconsistencies in realism could violate neurocognitive expectancies, leading to large feedback error signals (MacDorman and Chattopadhyay, 2016; Saygin et al., 2012). More studies are reporting that uncanniness can be explained by realism inconsistency theory resulting from mismatches in perceptual processing of the human likeness of an entity's features at different levels (e.g., when human face and skin paired with computer-generated eyes; human voice with robot head) (e.g., Chattopadhyay and MacDorman, 2016; Di Natale et al., 2023; MacDorman and Chattopadhyay, 2017; Zhang et al., 2020). Research overall suggests that realism inconsistency theory in contrast to category uncertainty theory is a better explanation for the UV phenomenon. Apart from realism inconsistency theory, which is a form of perceptual mismatch hypothesis, the perceptual mismatch hypotheses also involve sensitivity to atypical features (e.g., human-like entities with atypical features will generate a stronger negative reaction when compared to artificial entities with atypical features, as well as human-like and artificial entities with typical features; Kätsyri et al., 2015), as evidenced by findings indicating that the most negative affinity was triggered when artificially enlarged eyes were matched with real human faces (Seyama and Nagayama, 2007).

Configural processing theories assume that deviations in the configural pattern of specific and familiar stimuli cause the UV effect (Kätsyri, 2018; Kätsyri et al., 2019). The sensitivity to facial proportions heightens as human likeness increases, and a feeling of eeriness is generated by deviation from the ideal proportions of more attractive faces (Green et al., 2008). Furthermore, the Thatcher illusion also supports configural processing theories' explanations of the UV effect. Specifically, a face with inverted eyes and mouth can make the appearance grotesque, which indicates that recognizing deviations in configural processing can elicit uncomfortable reactions (Diel and MacDorman, 2021). Recently, Diel and Lewis's (2022) configural processing account interprets the UV effect by a moderated linear function for which perceptual specialization enhances the sensitivity to deviating stimuli (e.g., voice distortions), which gives rise to an increase in uncanniness (Diel and Lewis, 2024).

Mori (1970) initially proposed the importance of the feeling of uncanniness in human self-preservation. Diseases and death have been two primary threats to human evolution. Avoidance of threats (e.g., pathogens or unit mates, Moosa and Ud-Dean, 2010) can explain the UV effect. One viewpoint, known as the Pathogen Avoidance hypothesis, holds that the UV effect is associated with an evolutionary mechanism for pathogen avoidance. Specifically, people connect the humanoid robots' (with great human-likeness) flaws with indicators of transmissible diseases, causing disgust (Wang et al., 2015). Even though previous work provides indirect evidence for the Pathogen Avoidance hypothesis by showing the correlation between disgust and UV (MacDorman and Entezari, 2015; Villacampa et al., 2019), direct support for this hypothesis is lacking. Another explanation regarding threat avoidance is the Mortality Salience hypothesis, which posits that the defects of human-like entities can remind people of death and the entities may be regarded as dead persons who come alive; that the feeling of uncanniness is elicited since defense systems are activated to deal with the fear of death and anxiety for mortality (MacDorman and Ishiguro, 2006). Moreover, the Mortality Salience hypothesis is supported by the finding of the association between the physical body's vulnerability and impermanence sensitivity and the android's eerie evaluations (MacDorman and Entezari, 2015).

Psychopathy avoidance Theory links uncanniness with psychopathic personality traits. Tinwell et al. (2013) found that virtual characters lacking upper facial expressions were perceived to be most uncanny, and this uncanniess could be strongly predicted by psychopathy assessments (Tinwell et al., 2013). The researchers believe that the virtual characters without upper facial expressions may be uncoordinated with other facial movements, failing to express their emotions as well as potentially implying an effort to conceal their negative personality traits. Additionally, the personality traits may be connected with psychopathy with other aspects (e.g., jerky movement, MacDorman et al., 2010; lip-sync error, Tinwell et al., 2010), resulting in the feeling of uncanniness. Therefore, the uncanniness may be a trigger of avoidance response.