Do Beliefs About Whether Others Can See Modulate Social Seeking in Autism?

Current models of gaze processing suggest two possible neurocognitive responses to the gaze of another person—a basic alerting mechanism and a more elaborate perceptual mentalizing process. The basic alerting mechanism is believed to be implemented by a sub-cortical pathway, whereby an image of a pair of eyes activates the superior colliculus, pulvinar and amygdala (Senju and Johnson 2009b), and directs attention towards the eyes (Senju and Hasegawa 2005). It is thought that this pathway has an evolutionary meaning, since in many non-human species direct gaze signals threat from a predator (Emery 2000). In humans, neuroimaging studies indicate that the subcortical regions modulate, in turn, the recruitment of the social brain network (e.g. superior temporal sulcus, medial prefrontal cortex), which modulates the processing of contextual social information (Senju and Johnson 2009b). Overall, the subcortical mechanism operates rapidly and seems to function in the same way for all stimuli which match the basic features of a pair of eyes. Thus, we could describe this attentional gaze processing as being driven by an ‘eye-shape’, regardless of whether that is a photo, a schematic drawing or a live person.

More recently, research has also begun to uncover how people process gaze as a form of perceptual mentalizing. It is important to distinguish perceptual mentalizing from the audience effect. Perceptual mentalizing is the attribution of perceptual states to other people about whether they can or cannot see something (Teufel et al. 2010a). Instead, the audience effect refers to changes in behaviour that happen under the belief that other people can specifically see me (Hamilton and Lind 2016). Substantial research has investigated the audience effect, showing that the belief in being seen modulates emotional arousal (Myllyneva and Hietanen 2015), eye gaze directed at the observer (Gobel et al. 2015; Laidlaw et al. 2011), and prosocial behaviour (Izuma et al. 2009). However, only few studies have looked at how perceptual mentalizing influences social information processing.

For instance, Teufel et al. (2010b) found that gaze cueing effects are driven by perceptual mentalizing. Participants saw videos of an actor wearing obscured goggles with different coloured frames, and were told that one pair of goggles was transparent whereas the other was opaque; participants were also told that the videos were live streamed from an adjacent room. Results showed that the gaze cueing effect was greater in valid trials (i.e. trials where actor’s gaze and target are on the same side) only for the transparent goggles. This suggests that the gaze cueing effect was strongly influenced by the attribution of meaningful mental states to the confederate, and that this only happened when the confederate could see through the goggles (see Nuku and Bekkering 2008 for a similar study).

In another study, Furlanetto et al. (2016) showed that perceptual mentalizing also influences visual perspective taking. In each trial participants first saw the words ‘YOU’ or ‘SHE/HE’ on the screen, then a digit, and finally an avatar looking towards the right or left wall of a room (created with a virtual reality software). Each wall had a number of dots that ranged from 0 to 2. Participants had to judge whether the digit was consistent with the number of dots that themselves (‘YOU’) or the avatar (‘SHE/HE’) could see, respectively. Crucially, this task was combined with the goggles paradigm designed by Teufel et al. (2010b): the avatar was wearing obscured goggles that participants believed to be either transparent or opaque. They found that reaction times were slower when the avatar was wearing transparent goggles, suggesting that participants took the perspective of the avatar only when they knew the avatar could see the dots through the goggles.

Thus, these studies show that perceptual mentalizing itself modulates how we process social information from eye gaze. Here, we aim to test the relationship between perceptual mentalizing and social motivation, and whether this relationship is different in autism.

People with the Autism Spectrum Condition (ASC) show persistent difficulties in social interactions (DSM-5; Diagnostic and Statistical Manual of Mental Disorders 5th Ed. 2013) including the processing of eye gaze. Eye tracking studies suggest autistic people make fewer fixations on the eyes of actors in a movie (Klin et al. 2002), though this effect is reduced for static photos of faces (Chevallier et al. 2015). They have also difficulties in understanding the social meaning of gaze duration and gaze direction, and show abnormal recruitment of brain areas that process this information (Georgescu et al. 2013; Senju et al. 2005). At the same time, there is evidence suggesting that autistic people might be less susceptible to the belief in being seen (audience effect): for instance, they do not seem to increase prosocial behaviour when other people are observing them (Cage et al. 2013; Izuma et al. 2011). However, no previous research has yet addressed perceptual mentalizing in autism. Thus, it is not yet clear if differences in responsiveness to gaze in autism are due to changes in the subcortical gaze-attention mechanism, or due to differences in perceptual mentalizing about whether others can see something. The present paper investigates this issue, in relation to the measurement of motivation.

Recently, it has been suggested that typical adults find social stimuli and social interactions inherently rewarding, possessing an inbuilt social motivation, but that this motivation might be reduced in autism (Chevallier et al. 2012). For instance, autistic participants self-report diminished enjoyment in social situations when compared to typical ones (Chevallier et al. 2011). At the neural level, they activate reward-related brain areas (e.g. ventral striatum) significantly less than typical people, specifically in response to social reward (Scott-Van Zeeland et al. 2010; see also; Delmonte et al. 2012 and; Kohls et al. 2013). Chevallier et al. (2012) distinguish three levels of social motivation. First, social orienting happens when we give attentional priority to social stimuli rather than non-social; second, there is social seeking when we make an effort to get social stimuli because we like them (i.e. we want them because we find them inherently rewarding); finally, social maintaining refers to the development of strategies to enhance relationships with others (e.g. be viewed as likeable, cooperative, etc.). For the scope of the present study we will focus on social seeking.

A number of studies have found ways to measure social seeking behaviour. For example, people will sacrifice a small monetary reward to view a genuine smile (Shore and Heerey 2011), and will press a key repeatedly to see attractive people (Hayden et al. 2007). A new measure of social seeking behaviour has been recently developed, which attempts to quantify how much effort people will make to seek out social stimuli. In this Choose-A-Movie (CAM) paradigm, participants are presented on each trial with two coloured boxes and know that (for example) the orange box contains movies of people, while the green box contains movies of household objects. They can chose which box to open and thus which movie category to watch on each trial. To encourage careful decision making, on each trial there are a number of locks on each box (one, two or three locks), and to open a box with three locks participants must press a key and wait for that lock to open three times. This imposes a small but noticeable effort cost on that box, compared to a box with one lock. Thus, participants must trade-off their preference to view a particular movie against the effort involved in viewing it. Results show that they do just that (Dubey et al. 2015). More critically, the data show that typical adults prefer to view direct gaze video-clips to averted gaze video-clips and object video-clips. Conversely, autistic individuals prefer object clips to direct or averted gaze, and have a weak preference for averted gaze over direct gaze.

However, it is not yet clear why autistic people show differences in their motivation to view direct gaze stimuli. Many differences in gaze processing have been reported in autism, such as gaze comprehension (Baron-Cohen et al. 1997) and attentional cueing from gaze (Senju and Johnson 2009a). These effects could be explained by differences in a basic sub-cortical mechanism sensitive to eye-shapes. Autistic individuals also show differences when attributing mental states to others (Baron-Cohen et al. 1985; Senju et al. 2009), and seem to be less sensitive to the presence of other people observing them (Cage et al. 2013; Izuma et al. 2011). This suggests that differences in response to direct gaze might reflect differences in perceptual mentalizing capacities. In the present study, we contrast these two hypotheses in the context of social motivation. Our task also allows us to distinguish whether differences in social motivation are due to active avoidance of eye gaze and social information or passive omission of these cues (Corden et al. 2008; Senju and Johnson 2009a).

The present study aimed to dissociate responses to eye-shapes, presumably processed by a subcortical mechanism (Senju and Johnson 2009a, b), and responses to the belief that another person can see something, presumably related to perceptual mentalizing (Teufel et al. 2010a). To test responses to eye-shapes, we contrasted videos of an actress with empty sunglasses (no lenses) to videos of the actress wearing normal sunglasses. To test responses to the belief that others can see, we adopted the manipulation of Teufel et al. (2010b), where participants believed the actress can see through one set of sunglasses (normal sunglasses, e.g. those with blue frames) but cannot see through the other (opaque sunglasses, e.g. those with red frames). We counterbalanced which frame colour was linked to each belief (can see or cannot see) across participants. To complete our 2 × 2 factorial design, we created videos of the actress wearing sunglasses with paper eyes glued over the lenses—typical eye-shapes are present in this manipulation, but the actress cannot see through. This set of stimuli gave a complete 2 × 2 factorial design with levels Belief in seeing (Belief, B+ and B−) and Eye-shape (Eyes, E+ and E−); sample stimuli from each cell are shown in Table 1.

Taking these four social video-clips, we used the Chose-A-Movie task to measure which movie participants preferred to view. Expanding on the design by Dubey et al. (2015), we gave participants a choice of all four movies on each trial, with variable numbers of locks to manipulate effort. Our primary outcome measure was the percentage of trials on which participants chose each movie. Note that we did not attempt to persuade participants that these videos represented a live-feed of another person from another room. We took this decision for three reasons. First, we wanted to specifically test the effect of perceptual mentalizing (belief that someone can see something) on social motivation. Second, we wanted to remain close to the procedure of Dubey et al. (2015), where there was no live-feed manipulation. Third, it is not very plausible to have a live-feed of the same person wearing multiple different sets of sunglasses at very short notice. Thus, our experimental design addresses perceptual mentalizing in a non-interactive way: participants know that they are not in an online interaction with the actress.

Based on our hypotheses, we can make several predictions. First, if typical participants prefer attributing perceptual mental states because they have no difficulties in further social information processing, they should choose to view stimuli with empty and normal sunglasses (B+ conditions, actress can see through) over stimuli with paper-eyes and opaque sunglasses (B− conditions, actress cannot see through). Second, if autistic people show reduced social motivation because they are not sensible to eye-shapes, they will show no difference in preference for E+ and E− conditions. However, if autistic individuals actively avoid eye-shapes, they should prefer videos with normal and opaque sunglasses (E− conditions, eyes not visible) to videos with empty and paper-eyes sunglasses (E+ conditions, eyes visible). Third, if autistic people show reduced social motivation due to difficulties in attributing a perceptual mental state to the actress, they should show no difference in preference for B+ and B− conditions. If they attribute perceptual mental states but are less motivated to engage because of difficulties in further processing of this or other social information, they should prefer videos with opaque and paper-eyes sunglasses (B− conditions) to videos of empty and normal sunglasses (B+ conditions). Systematic group differences should be reflected in Group × belief and group × eyes interactions. Because the paper-eyes sunglasses are a slightly odd stimulus with low ecological validity, the strongest test of the claim that beliefs about whether someone can see impact on social motivation comes from examining the conditions where participants cannot see the eyes of the actress but their belief about whether or not she can see through is manipulated. Thus, the Group X Belief interaction was also examined including only the E− conditions, that is, normal sunglasses (B+E−) and opaque sunglasses (B−E−).

A group of 25 typical adults and 27 adults with ASC participated in this experiment. Previous studies (Dubey et al. 2015) show that this sample size is suitable to detect differences between two groups. Table 2 shows that the participants included in the analyses (i.e. after removal of outliers; sample of 25 typical and 24 ASC participants) were matched on age, gender, handedness and Intelligence Quotient (IQ; Wechsler Adult Intelligence Scale, WAIS-III UK, Wechsler 1999a, b; or Wechsler Abbreviated Scale of Intelligence, WASI-II); note that the ASC group was high functioning, which means that their IQ is higher than 80: since both groups were matched, the typical group also had high IQ on average. The two groups differed on Autism Quotient (AQ; Baron-Cohen et al. 2001). All participants were recruited using an autism database at the author’s institution. Recruitment of ASC participants was based on diagnosis from an independent clinician. Routine diagnostic procedures include the Autism Diagnostic Observation Schedule (ADOS; Lord et al. 2000) and the Autism Diagnostic Interview-Revised (ADI-R; Lord et al. 1994), among others. ASC participants included in the analyses were diagnosed as either Asperger’s Syndrome (17), Autism (6) or Autism Spectrum Disorder (1). They were also tested on module 4 of the ADOS by a trained researcher: seven participants met the ADOS classification for Autism; ten for Autism Spectrum; seven did not meet the classification for any of them, but all seven participants had a clear diagnosis from an independent clinician, and reached the cut-off for autism spectrum on either the communication or reciprocal social interaction subscale. They gave written informed consent before doing the experiment and were compensated for their time and travel expenses. All procedures were approved by the local Research Ethics Committee and were in accordance with the Declaration of Helsinki and APA ethical standards.

Four different types of video-clips were created. These video-clips corresponded to the four categories of a 2-by-2 factorial design with factors Belief (B) and Eyes (E) (see Table 1), which were determined by the type of sunglasses the actress on the video-clip was wearing. The four categories were the following: B+E+ (belief in seeing through and eyes visible: empty sunglasses), B−E+ (no belief in seeing through but eyes visible: paper-eyes sunglasses), B+E− (belief in seeing through but eyes not visible: normal sunglasses) and B−E− (no belief in seeing through and eyes not visible: opaque sunglasses). Empty sunglasses had black frames and no lenses; paper-eyes sunglasses had black frames and a picture of an eye glued over each lens; normal and opaque sunglasses had either blue or red frames (counterbalanced across participants) and mirrored lenses (so that participants could not see the eyes of the actress); opaque sunglasses also had a black paper glued behind each lens.

During the filming session, we initially recorded eight different actors (four female, four male). They sat in front of a blue background and the camera was prepared to depict a medium close shot. They were instructed to first look down and, upon hearing the experimenter call their name, look up directly to the camera and smile as if they were greeting a friend. They were recorded while wearing empty sunglasses, paper-eyes sunglasses, and normal sunglasses (with both red and blue frames). Since the lenses of the sunglasses were mirrored, the videos with normal sunglasses were also used for the opaque sunglasses condition: this ensured that there were no perceptual differences in the face between these two conditions. The final video-clips had a duration of 3 s, and were saved at 600 × 450 pixels.

All 32 videos (eight possible actors for each of the four video categories) were rated on trustworthiness, friendliness and genuineness of the greeting (see Supplementary Materials S1). We found that one of the female actresses received higher ratings than the other actors, for all four video categories. Her performance matched our requirements for these stimuli, since we needed to ensure that social seeking was not affected by the actress being too neutral on the facial expression. Thus, we decided to only use this actress in the stimuli.

In addition to the video-clips, four abstract coloured patterns were used as cues for the four video categories (Fig. 1a). Patterns associated with empty (green pattern) and paper-eyes (yellow pattern) sunglasses were the same for all participants. Patterns associated with normal and opaque sunglasses were counterbalanced according to the counterbalancing of the frame colour of the sunglasses (red and blue patterns).

The choose-a-movie task design was coded with MATLAB (8.5, MathWorks 2015) and Cogent Graphics. Similar to Dubey et al. (2015), participants first learnt that each video-clip category was linked to one particular patterned box (see Fig. 1a). Then, on each trial (see Fig. 1b for a sample trial), participants saw four patterned boxes on the screen, and each box could have 1, 2 or 3 locks. All possible combinations of locks (3) and video categories (4) appeared across trials, and each video category randomly appeared in one of the four positions on the screen. Participants unlocked the boxes by pressing a specific key on the numeric keypad in front of them. The key they were required to press was linked to the position of the box (one for bottom–left, three for bottom–right, seven for top–left, nine for top–right), and they had to press the key as many times as number of locks the box had (e.g. if the box had three locks, they would press the key three times). Thus, on each trial participants had to choose their preferred video category in trade-off with effort (i.e. number of locks or keyhits). When a box was unlocked, the video-clip with the actress wearing the corresponding type of glasses was displayed as a reward.

Participants came to the lab to perform the experiment as part of a Research Day. They sat in a chair approximately 1 m from the projector screen, and between them and the screen there was a table with the numeric keypad. The experimenter first showed the four types of sunglasses to the participants in order to manipulate their beliefs. For each type of sunglasses, the experimenter first named the sunglasses and then described whether a person would be able to see through them or not. The sunglasses were always presented in the same order: empty sunglasses, paper-eyes sunglasses, normal sunglasses and opaque sunglasses. However, for half of the participants the normal sunglasses had blue frames and the opaque sunglasses had red frames, whereas for the other half it was the opposite. Participants tried all the sunglasses and left them on the table. In front of each type of sunglasses the experimenter placed a label with the name of the sunglasses. Participants then started the learning phase, during which they learnt that each video category was linked to one particular patterned box. In case they forgot the link between patterns and video category, the background of the label in front of each type of glasses matched the corresponding pattern for that category. During the learning phase participants completed eight trials in order to practise how to unlock boxes. Then participants performed the test phase, during which they completed 81 trials (i.e. all possible combinations of three locks and four video categories) across three blocks (27 trials in each block). Because participants originally performed a mimicry task interspersed between the CAM task blocks, the overall duration of the experiment was approximately 45 min, from which 20 min where devoted to the CAM task (the results from the mimicry task will be reported elsewhere). In the end of the task participants completed a post-test questionnaire in order to check that they understood the meaning of each type of glasses (see Supplementary Materials S2). Only two typical participants and three ASC participants gave a wrong answer in one of the questions. The two typical participants responded wrongly the questions about normal (red in the questionnaire) and opaque (blue in the questionnaire) sunglasses, respectively. The three ASC participants responded wrongly the questions about paper-eyes (fake-eyes in the questionnaire), normal (blue in the questionnaire) and opaque (red in the questionnaire) sunglasses, respectively. Removal of these participants did not affect the results so they were all included in the statistical analysis. Data on ADOS scores, IQ scores and AQ scores for all participants was available from previous testing sessions at the author’s institution.

The primary outcome measure for each participant was the percentage of trials on which each video was chosen. There were four possible video-clip categories, so 25% was considered chance level. Three outliers within the autistic sample were removed, since their mean percentage of choices for one of the conditions was above 3 SDs from the overall mean. In particular, each of these three participants had chosen the same condition in more than 95% of the trials (these conditions were empty sunglasses, paper-eyes sunglasses and normal sunglasses), which suggests that their choice may based on a particular preference for the colour or pattern associated with that video category. Statistical analyses were performed for both the full sample and the final sample excluding outliers (25 typical and 24 autistic participants; see Table 2). A three-way repeated measures ANOVA was performed. Belief in seeing (B+ and B−) and Eye-shape (E+ and E−) were used as within-subjects factors, Group (typical and autism) as between-subjects factor, and the measure percentage of choices as dependent variable. However, to test and interpret our hypotheses we focused on two-way interactions between Belief X Eyes, Belief X Group and Eyes X Group. The Belief X Group interaction was also examined including only the E− conditions, that is, normal sunglasses (B+E−) and opaque sunglasses (B−E−). Post-hoc pairwise comparisons using Bonferroni’s adjustment were also computed. Since the analyses for the full and final sample showed the same pattern of results, here we only report the results for the final sample.

When analysing choice preferences it is important to take into account how the choice of a particular video category was affected by the relative effort required to watch that video. The algorithm implemented in Dubey et al. (2015) can only be used in designs with two options, but ours has four options. Instead, we developed an algorithm that expanded upon this approach, and that is suitable for designs with four options. For each participant this algorithm computes a set of values (one for each video category) that best explains the choices, taking into account the number of locks present on each trial (1, 2 or 3). We call these values ‘choice scores’. A more detailed description of how the choice score is computed, together with the Matlab script that we used can be found in Supplementary Materials (S3.1). Data inspection using the choice score revealed that the model failed to optimise the values for seven participants, which were excluded from the statistical analysis (two outliers in the typical group and five outliers in the ASC group; but only three outliers in the ASC group were found when using the percentage of choices). The pattern of results using the choice score and the percentage of choices was the same. This indicates that preference for a particular video category, rather than the number of locks on the box, was driving the choices of participants. Since in the analysis with the percentage of choices we can include four extra participants, and the pattern of results is similar in both analysis, here we only report the results for the percentage of choices. The full analysis using choice scores can be found in the Supplementary Materials (S3.2).

Finally, we also tested group differences in the number of locks that were opened on average. We performed a two-way ANOVA with number of locks (1, 2, 3) as within-subjects factor, Group (typical and autism) as between-subjects factor, and proportion of choice of each number of locks as dependent variable (using a sample of 25 typical and 24 autistic participants). Given that the analyses using the percentage of choices (for video categories) and choice score yielded the same pattern of results, we expected no differences between groups regarding the number of locks opened throughout the task. Results showed that this was the case: both groups chose to open one lock more often than two and three locks (see Supplementary Materials, S4). This indicates that both groups put same overall amount of effort to see the videos.