A motion aftereffect from viewing other people’s gaze

Recent work suggests that our brains may generate subtle, false motion signals streaming from other people to the objects of their attention, aiding social cognition. For instance, brief exposure to static images depicting other people gazing at objects made subjects slower at detecting subsequent motion in the direction of gaze, suggesting that looking at someone else’s gaze caused a directional motion adaptation. Here we confirm, using a more stringent method, that viewing static images of another person gazing in a particular direction, at an object, produced motion aftereffects in the opposite direction. The aftereffect was manifested as a change in perceptual decision threshold for detecting left versus right motion. The effect disappeared when the person was looking away from the object. These findings suggest that the attentive gaze of others is encoded as an implied agent-to-object motion that is sufficiently robust to cause genuine motion aftereffects, though subtle enough to remain subthreshold.

Graziano, 2020b). In this study, we made use of a visual phenomenon called the motion 14 aftereffect to test a prediction of this proposed model: viewing static images depicting other 15 people gazing in a particular direction, at an object, should lead to an illusory subsequent 16 motion in the opposite direction. 17 The motion aftereffect is a classic phenomenon of a false motion signal in the visual 18 image caused by prior exposure to motion in the opposite direction (Anstis et al., 1998;19 probes (Glasser et al., 2011;Levinson and Sekuler, 1974). A genuine motion aftereffect is 1 associated with slower reaction times and decreased accuracy for motion test probes of the 2 same directionality as the adapting stimulus, reflecting direction-specific neuronal fatigue 3 affecting motion processing time and perceptual decision-making. In a series of seven 4 behavioral experiments , we previously showed that brief 5 exposure to static images depicting a person gazing in a particular direction, at an object, 6 made subjects significantly slower at detecting subsequent motion in the direction of gaze, 7 which is compatible with a motion aftereffect caused by gaze encoded as implied motion. 8 The effect disappeared when the depicted person was blindfolded or looked away from the 9 object, and control experiments excluded differences in eye movements or asymmetric 10 allocation of covert attention as possible drivers of the effect. However, because the 11 paradigm in ) was primarily designed for analysis of 12 reaction time rather than accuracy, the task was made easy and accuracy was close to 13 ceiling (mean accuracy across experiments = 91%). Thus, that experiment showed only 14 reaction time effects, and failed to reveal any meaningful accuracy effects. The goal of the 15 present study was to examine if seeing someone else's gaze direction caused enough of a 16 motion aftereffect to shift subjects' perceptual decisions about subsequent motion. The 17 present experiment is therefore a conceptual replication of the previous studies, but using 18 a different measure of the motion aftereffect to test whether the discovery is reliable and 19 robust across methods. 20 To achieve this goal, we modified the motion adaptation paradigm described in 21 , which was based on a random-dot motion direction 22 discrimination task, to maximize the likelihood of detecting meaningful differences in 23 . CC-BY 4.0 International license available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint this version posted November 9, 2020. ; https://doi.org/10.1101/2020.11.08.373308 doi: bioRxiv preprint accuracy. Subjects were tested using an online, remote platform (Prolific) (Palan and 1 Schitter, 2018) due to restrictions on research imposed by the coronavirus epidemic. (See 2 Materials and Methods for details of sample sizes and exclusion criteria). Just as in 3 , in each trial, subjects were first exposed to an image 4 depicting a face on one side of the screen, gazing at a neutral object, a tree, on the other 5 side (Fig 1A). After 1.5 s, the face-and-tree image disappeared, and subjects saw a random 6 dot motion stimulus in the space interposed between where the head and the tree had been. 7 The stimulus was shown for 1.0 s. The proportion of dots moving coherently in one 8 direction (dot coherence) varied across seven different levels. The coherence ranged from 9 30% of the dots moving left (and 70% moving randomly) to 30% moving right in 10 increments of 10% (thus, the middle condition of 0% coherence had 100% of the dots 11 moving randomly). After the dots disappeared, subjects made a forced-choice left-or-right 12 judgement of the global direction of the moving-dots stimulus. 13 This approach allowed us to calculate, at each level of coherence and on a subject-14 by-subject basis, the frequency of responses that were spatially congruent with the gaze 15 direction in the preceding face-and-tree image (i.e., the direction toward the location of the 16 tree). By fitting this data to a sigmoid function and extracting the sigmoid central point, we 17 estimated the perceived null motion, that is, the amount of motion coherence for which 18 subjects were equally likely to respond that the motion direction of the test probe was 19 congruent or incongruent with the preceding gaze direction. We found that viewing 20 another's gaze significantly shifted the perceived null motion, as if that gaze caused an 21 illusory motion aftereffect in the opposite direction (experiment 1). The effect disappeared 22 when the face in the display was looking away from the object (experiment 2; Fig 1B), 23 . CC-BY 4.0 International license available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint this version posted November 9, 2020. ; https://doi.org/10.1101/2020.11.08.373308 doi: bioRxiv preprint suggesting that the perception of the other actively gazing at the object was the key factor. 1 These findings extend previous results by demonstrating that viewing other people's gaze 2 is associated with a false motion signal, below the level of explicit detection but still 3 capable of generating a motion aftereffect that influences not only perceptual processing 4 time, but also perceptual decision thresholds about subsequent motion.

7
In both experiment 1 (face looking toward the tree) and experiment 2 (face looking 8 away from the tree), the appearance of the face on the left and tree on the right, or the face 9 on the right and tree on the left, were balanced and presented in a random order. The 10 subsequent dot-stimulus could move either leftward or rightward with 10%, 20% or 30% 11 coherence, or be completely random (0% coherence). For analysis, the trial types were 12 collapsed into seven conditions: -30%, -20%, -10%, 0%, +10%, +20%, and +30%, where 13 motion toward the location of the (preceding) tree were arbitrarily coded as positive 14 coherence, and motion away from the tree as negative coherence. Thus, the predicted 15 motion aftereffect from viewing the face actively gazing in the direction toward the tree 16 (in experiment 1) should produce a positive shift (>0%) of the perceived null motion. 17 Subjects performed 70 trials in seven blocks of 10 trials each, thus 10 trials per condition. prediction that implied motion streaming from the eyes toward the tree causes a motion 23 . CC-BY 4.0 International license available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint this version posted November 9, 2020. ; https://doi.org/10.1101/2020.11.08.373308 doi: bioRxiv preprint aftereffect in the opposite direction . In other words, 1 immediately after subjects saw a face gazing one direction, the amount of real motion 2 needed to make subjects think a test stimulus was randomly balanced between left and right 3 movement was 1.18% coherence in the direction that the face had been gazing. 4 In experiment 2 (n=64), where the face was looking away from the tree, the central 5 point of the sigmoid function was not significantly different from 0 (M = -0.47%, S.E.M. 6 = 0.47%; t63 = -1.01, p = 0.3165; Fig 2B). In a between-groups comparison, we found that  After all trials were completed, subjects in experiments 1 and 2 were asked what they 10 thought the purpose of the experiment might be, and whether they were explicitly aware of 11 any influence of the head-and-tree stimulus on their ability to respond to the dot motion 12 stimulus. Though subjects offered guesses about the purpose of the experiment, none 13 indicated anything close to a correct understanding. All subjects also insisted that, as far as 14 they were aware, the head-and-tree stimulus had no impact on their response to the second 15 stimulus. These questionnaire results suggest that any motion aftereffects observed here 16 probably occurred at an implicit level. 17

19
These results strongly support the notion that when people view a face looking at an 20 object, the brain treats that gaze as though a movement were present, passing from the face 21 to the object. The motion test probes were more likely to be judged as moving in the 22 direction opposite the gaze direction depicted in the previous adapting image than to be 23 . CC-BY 4.0 International license available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint this version posted November 9, 2020. ; https://doi.org/10.1101/2020.11.08.373308 doi: bioRxiv preprint moving in the same direction, but only when the agent in the image was actively gazing at 1 the object. This work extends previous results that focused on reaction times (Guterstam 2 and Graziano, 2020a). Here, perception of other people's gaze significantly biased 3 perceptual decisions about subsequent motion, which is a hallmark of the motion 4 aftereffect. We propose that this hidden motion signal, associated with gaze, is part of an 5 implicit 'fluid-flow' model of other people's attention, that assists in human social 6 cognition. 7 The null result of experiment 2 suggest that spatial priming, i.e., subjects simply 8 being more prone to choose the direction that the face was looking, is an unlikely 9 explanation to the findings of experiment 1. Had spatial priming been the driving factor, 10 one would expect a significant negative shift of the central point when the face was looking 11 away from the object (experiment 2), of the same magnitude as the observed positive shift 12 in experiment 1 where the face was gazing at the object. The absence of a motion aftereffect 13 in experiment 2 also suggests that the presence of a face on one side of the screen drawing 14 subjects' attention or gaze more to that side, cannot easily explain effect observed in 15 The present set of results adds an important piece of evidence to a growing body of 17 research on how people model the attention of others to support social cognition. The brain 18 seems to represent other people's attention as an implied, beam-like motion travelling from 19 an agent to the attended object. This motion signal may be detected using sensitive 20 behavioral motion adaptation paradigms, such as in the present study or in (Guterstam and 21 Graziano, 2020a). It can also be quantified using a tube-tilting task, in which subjects' 22 angular judgements of the tipping point of a paper tube were implicitly biased by the 23 . CC-BY 4.0 International license available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint this version posted November 9, 2020. ; https://doi.org/10.1101/2020.11.08.373308 doi: bioRxiv preprint presence of the gazing face, as if beams of force-carrying energy emanated from eyes, 1 gently pushing on the paper tube (Guterstam et al., 2019). The motion signal is also 2 detectable in brain activity patterns in the human motion-sensitive MT complex and in the 3 temporo-parietal junction, which responded to the gaze of others, and to visual flow, in a 4 similar manner (Guterstam et al., 2020a). Finally, by contaminating a subject's visual 5 world with a subthreshold motion that streams from another person toward an object, we 6 could manipulate the subject's perception of that other person's attention, suggesting that 7 subthreshold motion plays a functional role in social cognition (Guterstam and Graziano, 8 2020b). 9 Together, these present and previous findings suggest that the visual motion system 10 is used to facilitate social brain mechanism for tracking the attention of others. We 11 speculate that this implicit social-cognitive model, borrowing low-level perceptual 12 mechanisms that evolved to process physical events in the real world, may help to explain 13 the extraordinary cultural persistence of the belief in extramission, the myth that vision is

Participants 20
For each experiment, participants were recruited through the online behavioral 21 testing platform Prolific (Palan and Schitter, 2018). Using the tools available on the Prolific 22 platform, we restricted participation such that no subject could take part in more than one 23 . CC-BY 4.0 International license available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint this version posted November 9, 2020. ; https://doi.org/10.1101/2020.11.08.373308 doi: bioRxiv preprint experiment. Thus, all subjects were naïve to the paradigm when tested. All participants 1 indicated normal or corrected-to-normal vision, English as a first language, and no history 2 of mental illness or cognitive impairment. All experimental methods and procedures were 3 approved by the Princeton University Institutional Review Board, and all participants 4 confirmed that they had read and understood a consent form outlining their risks, benefits, 5 compensation, and confidentiality, and that they agreed to participate in the experiment. 6 Each subject completed a single experiment in a 6-8 min session in exchange for monetary 7 compensation. As is standard for online experiments, because of greater expected variation 8 than for in-lab experiments, relatively large numbers of subjects were tested. A target 9 sample size of 100 subjects per experiment was chosen arbitrarily before data collection 10 began. Because of stringent criteria for eliminating those who did not follow all instructions 11 or showed poor task performance (see below), initial, total sample sizes were larger than 12 100 and final sample sizes for those included in the analysis varied between experiments 13 (experiment 1, ntotal=115, nincluded=59, 17  (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint this version posted November 9, 2020. ; https://doi.org/10.1101/2020.11.08.373308 doi: bioRxiv preprint coherence levels, we excluded all subjects whose accuracy was less than 80% when 30% 1 of the dots moved either right or left, in accordance with the exclusion criterium used in 2 (Guterstam and Graziano, 2020a). The relatively high rate of exclusion due to poor 3 performance here (35% on average) was expected given that the average exclusion rate 4 was 19% in a previous study (Guterstam and Graziano, 2020a) using the same dot motion 5 direction discrimination task but with a fixed 40% coherence level, which is easier to 6 detect. Moreover, participants in (Guterstam and Graziano, 2020a) underwent up to four 7 sets of 10 practice trials, with feedback, before commencing the main experiment, since 8 reaction times (RTs), and not accuracy, was the outcome of interest in that study. In the 9 present study, subjects did not undergo any practice sessions, because accuracy was our 10 primary outcome. It therefore seems probable that the absence of practice trials and lower 11 dot coherence levels in the present study fully explain the higher exclusion rates reported 12 here compared to in .  (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint this version posted November 9, 2020. to demonstrate that you have read these instructions carefully, please ignore the 'Continue' 1 button below, and click on the 'x' to start the practice session." Two buttons were 2 presented at the bottom of the screen, "Continue" and "x", and clicking on "Continue" 3 resulted in a failed IMC. (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint this version posted November 9, 2020. ; https://doi.org/10.1101/2020.11.08.373308 doi: bioRxiv preprint looking in this spot throughout the trial", (2) Image, (

3) Dot motion, and (4) Response]. 1
After viewing the dots for 1 sec, indicate the direction of the motion using the left or right 2 arrow keys. There is always one correct answer. However, the motion direction will be 3 obvious in some trials, and very subtle in other trials. The preceding face-and-tree image 4 is irrelevant to the task and does not predict the motion direction in any way. square. Dot speed was 2 px/frame. Dot lifetime was 12 frames, which corresponds to 17 200ms on a standard 60 Hz monitor. There were seven different types of dot motion 18 coherence levels: either 30%, 20%, or 10% of the dots moved leftward (while the 70%, 19 80%, or 90% of the dots moved in random directions), or 30%, 20%, or 10% of the dots 20 moved rightward. In one condition, zero % of the dots moved coherently (i.e., 100% 21 random directions). After 1.0 s, the dots disappeared, and the subjects were prompted about 22 the direction of the motion ("Left or Right?). They indicated their response by pressing the 23 . CC-BY 4.0 International license available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint this version posted November 9, 2020. ; https://doi.org/10.1101/2020.11.08.373308 doi: bioRxiv preprint left or right arrow key on their keyboard, after which the trial terminated and the next trial 1

began. 2
For the statistical analysis, the trial types were collapsed into seven conditions: -30%, 3 -20%, -10%, 0%, +10%, +20%, and +30%. Motion toward the location of the (preceding) 4 tree were arbitrarily coded as positive coherence, and motion away from the tree as 5 negative coherence. On a subject-per-subject basis, for each condition, we calculated the 6 proportion of responses that was spatially congruent with the direction away from the face 7 and toward the tree (which, in experiment 1, corresponded to the gaze direction of the face). After the experiment, the subjects completed a survey. They were first asked an open-20 ended question: "What do you think the hypothesis of the experiment was?". They were 21 then given the binary yes-or-no question: "Did the head-and-tree image influence your 22 responses to the dots?" Subjects who responded "yes" were asked, "Please describe in 23 . CC-BY 4.0 International license available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint this version posted November 9, 2020. ; https://doi.org/10.1101/2020.11.08.373308 doi: bioRxiv preprint what way the head-and-tree image influenced your responses to the dots." Subjects who 1 in any way indicated that they had figured out the purpose of the experiment were excluded 2 from the analysis. 3 4

Experiment 2 5
The design, procedures, and statistical analysis of experiment 2 were identical to 6 those of experiment 1, with one exception: the face was turned away from the tree. This 7 control condition should eliminate any gaze-induced effect on motion judgments 8 . We therefore predicted that the mean central point in 9 experiment 2 would not significantly differ from to 0 (two-tailed one-sample t-test), and 10 that it would be significantly smaller than the mean central point among participants in 11 experiment 1 (two-tailed two-sample t-test). (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint this version posted November 9, 2020. ; https://doi.org/10.1101/2020.11.08.373308 doi: bioRxiv preprint   . CC-BY 4.0 International license available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint this version posted November 9, 2020. ; https://doi.org/10.1101/2020.11.08.373308 doi: bioRxiv preprint 1 Figure 2. Results. A. For experiment 1 ("Face looking toward tree"; solid line) and 2 experiment 2 ("Face looking away from tree"; dotted line), the frequency of dot-motion 3 responses that were spatially congruent with the direction toward the tree is plotted as a 4 function of motion coherence relative to the location of the tree. (We collapsed trials in 5 which the tree appeared on the left or right side, and motion toward the tree is arbitrarily 6 coded as positive coherence, and motion away from the tree as negative coherence.) Error 7 bars indicate SEM. The sigmoid functions shown here were fitted to the group-mean values 8 across coherence levels for the two experiments, for display purposes. B. Mean sigmoid 9 central point estimates in experiment 1 and 2, based on subject-level fitting. The central 10 point reflects the perceived null motion, that is, the amount of motion coherence for which 11 subjects were equally likely to respond that the motion is "going toward the tree" as "going 12 away from the tree". When the face was looking at the tree (experiment 1), the central point 13 . CC-BY 4.0 International license available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint this version posted November 9, 2020. ; https://doi.org/10.1101/2020.11.08.373308 doi: bioRxiv preprint