Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.
Advertisement
Communications Psychology volume 2, Article number: 117 (2024)
15k
7
219
Metrics details
Paranoia (believing others intend harm) and excess teleological thinking (ascribing too much purpose) are non-consensual beliefs about agents. Human vision rapidly detects agents and their intentions. Might paranoia and teleology have roots in visual perception? Using displays that evoke the impression that one disc (‘wolf’) is chasing another (‘sheep’), we find that paranoia and teleology involve perceiving chasing when there is none (studies 1 and 2) — errors we characterize as social hallucinations. When asked to identify the wolf or the sheep (studies 3, 4a, and 4b), we find high-paranoia participants struggled to identify sheep, while high-teleology participants were impaired at identifying wolves — both despite high-confidence. Both types of errors correlated with hallucinatory percepts in the real world. Although paranoia and teleology both involve excess perception of agency, the current results collectively suggest a perceptual distinction between the two, perhaps with clinical import.
People perceive intentions not just when interacting with others, but even when viewing simple geometric shapes which move in certain ways1,2. These social impressions arise from low-level visual primitives, such as relative distance3, facingness4, contingency5, and asymmetrical relatedness6. These experiences may also be reflected directly in neural data—with lower-level primary visual areas, temporoparietal junction7, and posterior superior temporal sulcus (STS)8,9 comprising a ‘who’ pathway dedicated to rapid social information processing10.
When the attribution of intention goes awry, clinical symptoms such as paranoia or excess teleological thinking have been observed11. Paranoia is a belief in the malicious intentions of others12, while excess teleological thinking evokes purposes to explain unintentional events13. Though seemingly distinct, both seem to involve Theory-of-Mind (ToM)—the attribution of mental states to others. A ToM deficit in patients with paranoia may stem from difficulties reasoning about hypothetical mental states14,15, and teleological intuitions may also entail ToM, insofar as purposive behaviours often imply agents with minds and intentions16. Here, we ask whether high-level beliefs like paranoia and teleology may, in part, be rooted in perceptual mechanisms to a greater degree than previously thought11,16.
We suggest that one route through which high-level beliefs like paranoia and teleology might manifest in perception is through social hallucinations. Hallucinations have been argued to reflect the influence of expectations on perception17, and recent work has operationalized hallucinations as high-confidence false-alarms—believing that one has perceived, with high conviction, in the absence of a signal17,18,19,20. Some have argued that high-confidence false-alarms are caused by excess top-down influence17,20,21, while others suggest that they represent a dearth of expectation and are instead responses to bottom-up noise18. Critically, hallucinations often also co-occur with ‘high-level’ symptoms such as persecutory delusions in schizophrenia22, and both have distinctly social contents23. It remains unclear exactly how hallucinations relate to higher-level beliefs about the intentions of others. And thus far, distinctly social visual stimuli have not been examined within this framework of hallucinations, even though clinical hallucinations often involve the sensed presence of another agent23. We thus sought to establish whether paranoia and/or teleology—which both seem to involve ToM processes—may involve high-confidence errors in a distinctly social form of visual perception.
To investigate how people varying in paranoia and teleological beliefs perceive (and hallucinate) the intentions of agents, we employed a type of perceived animacy involving the detection of chasing3,24,25 (Fig. 1). In the chasing paradigm, participants viewed multiple discs moving around a screen3,24,26, and their task was to report whether they perceived one disc (the ‘wolf’) chasing another (the ‘sheep’)3. Of course, a critical factor for the perception of animacy in such displays—and for chasing success in general27—is the noise of the wolf’s trajectory with respect to the sheep. In this paradigm, such noise is operationalized as chasing subtlety3—the angular extent across which the ‘wolf’ is displaced on every frame of motion, relative to the ‘sheep’ (with 0° of chasing subtlety giving rise to perfect heat-seeking, and 180° of chasing subtlety giving rise to purely random motion). In the current studies, we used a modest 30° of chasing subtlety that has been found to result in middling chasing detection3, and participants had to discriminate on each trial whether a chase was present or not. On chasing-present trials, one disc (the wolf) pursued a randomly-chosen and randomly moving distractor (the sheep), without ever catching it (see Fig. 1A; Supporting Animation S1). Critically, to dissociate the perception of chasing from mere correlated motion, chasing-absent trials involved a ‘mirror’ manipulation—in which the wolf instead chased the (invisible!) mirror image of the sheep, as reflected through the centre of the display (see Fig. 1B; Supporting Animation S2)5,28. See more details of the task and different studies in Fig. 2. Reports of chasing on chasing-absent trials are thus false-alarms.
A is a graphical example of a chase-present trial where the wolf pursues the sheep with a chasing subtlety of 30°. The sheep moves randomly across the display, as do the distractor discs. B is a graphical display of a chase-absent control condition, where the wolf pursues an invisible sheep which mirrors the moves of another randomly moving disc.
A The temporal structure of a test trial (practice trials did not have a confidence screen, but they provided correct/incorrect feedback) for the 3 versions of the task. Study 1, chase detection; Study 2, chase detection and confidence; Study 3, identify either wolf or sheep, depending on the group; and Study 4 (both a and b, its replication), identify both wolf and sheep by participant. B Five frames as an example for one chase-present trial, with the wolf (black/dark maroon) chasing the sheep (beige). (Colours are only displayed for this diagram and were not shown to participants (see two actual displays in https://github.com/santiagocdo/socialHallucinations/tree/main/supporting_animation). Colour palette by Daniel Barreto (dbarreto.com).
We chose this paradigm, in particular, instead of other social displays (such as those in classic Heider and Simmel1 animations) because we wanted to focus on percepts relevant to paranoia, such as perceived threat. And this particular phenomenon of ‘chasing’ has been shown to operate in the domain of perception2,3,5,25,29,30. In particular, previous work has suggested that a hallmark of perception (as opposed to deliberation) is a strict dependence on subtle visual display details29: percepts seem to be irresistibly controlled by the nuances of the visual input (including aspects of the input that subjects do not consciously notice), and this has been observed for chasing detection3,26. Observers cannot simply decide to interpret any set of parameters as chasing; instead, this ability seems confined to only a relatively narrow range of temporal and spatial parameters. (It would have been helpful in these earlier studies if subjects could have ‘decided’ to treat 90° of chassing subtlety as indicating chasing (as, in fact, it did in terms of the underlying motion patterns); but even when observers have full knowledge of such manipulations and have every incentive to discount them, they cannot do so. These sorts of results suggest that the apprehension of chasing in such displays reflects some properties and constraints of automatic perceptual processing, rather than flexible higher-level decisions that subjects are overtly making about the contents of the displays29. (Of course, this does not rule out the possibility that subjective judgments may supervene the visual perceptual processes. Indeed, this is likely when we request that participants also express their confidence in their choices.)
Although not as direct as behavioural data, neural activation in visual areas associated with social perception, such as the extrastriate body area, temporoparietal junction7, and posterior STS4,9,31, supports the perceptual nature of these displays. These areas are more aligned with ‘lower’ rather than ‘higher’ visual processing. Others have argued that coalitional cognition and social status are fundamental for delusions and paranoia32, but the animacy task we used here targets rather ‘lower’ areas of social cognition33, and the invoked regions do not appear to implement reputation processing or coalitional threat detection4, which instead engage orbitofrontal cortex34. In sum, we are confident that the current stimuli are indeed capable of producing social percepts in vision.
In Studies 1 and 2, participants were required to detect if there was a chase or not, and we hypothesized that people with paranoia and excessive teleological thinking would be more likely to perceive chasing even when there were none (i.e., false alarms; Studies 1 and 2). In Study 3, we wondered whether particular aspects of the displays would be more or less difficult for people with differing beliefs about agency and intentions (paranoia and teleology). Thus, in a between-subjects design, we asked participants to identify either the wolf or the sheep (also measuring confidence but without detection)—and we hypothesized that paranoia and excessive teleology would result in impaired wolf and sheep identification. Finally, in Studies 4a and 4b, we conceptually replicated Study 3 (asking participants to identify both the wolf and sheep and to report their confidence) in two pre-registered within-subjects designs (see Fig. 2). Details about the samples from each individual experiment are available in Table 1, and the preregistrations for Studies 4a and 4b are available at https://aspredicted.org/wr7hb.pdf (preregistered on 17/10/2023) and https://aspredicted.org/4tz4d.pdf (preregistered on 22/10/2023). Studies 1–3 were not preregistered.
After consenting, participants were redirected to the behavioural task website, where they first completed practice trials with feedback and then the rest of the test trials without feedback. Following the task, participants were redirected to Qualtrics® to respond to questionnaires; first sociodemographic questions (age: “What is your age in years?”; and sex “What is your sex assigned at birth?”) and then the psychometric scales (described below). Participants completed the study in full-screen mode on either a laptop or desktop computer. Participants were paid $6.00 USD for their participation in each of the studies. We used the displays created for another study24, consisting of 600 4-s animations, half chasing-present and half chasing-absent. Studies 1 and 4 were run in CloudResearch® and Amazon-Web-Services®, Study 3 was run in Prolific®, and Study 2 was run in Connect®. This research was conducted under Helsinki Declaration35 and was approved by the Institutional Review Board (IRB) from Yale University.
We conducted a series of four cross-sectional studies, with the last study (4a) and its replication (4b) being preregistered. We next describe the differences between studies and participants.
Participants (Table 1) saw 10 chase-present and 10 chase-absent displays in fully randomized order as practice trials where feedback (correct/incorrect) was provided. Then, 90 chasing-present and 90 chasing-absent trials were used for testing. On each trial, participants reported (as fast as possible, by pressing one of two keys) whether one disc was ‘chasing’ another or not. If participants responded before the end of a display, it ended and the next trial immediately began; otherwise, participants saw the question “Chase or No Chase?” on the screen until they responded. In Study 1, participants only self-reported their paranoia12. Study 2 was similar, except that we also measured confidence at the end of each trial (assessed via keys 1–5) and participants also reported their teleological thinking36. The exclusion criteria for these studies were: (i) failing to achieve at least 55% accuracy across all trials; (ii) completing the task in under 7 min (which would indicate anomalously fast and uncareful responses); or (iii) at least with 50 valid trials (25% of the total), we defined “valid” those trials where the decision time was >500 ms and <8000 ms. One-hundred fifty participants finished Studies 1 and 2. We excluded 30 in Study 1 and 36 in Study 2.
Two-hundred participants were tested in two groups, each with a unique task: wolf-identification or sheep-identification. Before the test trials, participants were exposed to 10 practice trials where feedback (correct/incorrect) was provided. Then, at the test, they saw 50 chasing-present and 50 chasing-absent trials, and for each, they were asked to select (by clicking on a disc with their mouse) which disc was the wolf (for the wolf-identification group) or sheep (for the sheep-identification group). No detection response was required in Studies 3 and 4 (4a and 4b). In chase-absent trials, participants could not withhold their response, such that a correct response in a chase-absent trial was coincidental. This is why correct performance is at random levels (1/8 discs) in Fig. 3C and D.
A Study 1 (n = 120) and 2 (n = 114) detection variables and paranoia. Significance tests are based on regression coefficients from simple logistic regressions predicting paranoia (high = 1, low = 0) with each x-axis behavioural variable. B Study 2 detection variables and teleology. Significance tests are based on regression coefficients from a simple linear regression predicting teleology (BPE; continuous) with each variable. Scaled scores (z-scores) were used only for visualization purposes in (A) and (B). No p-value was adjusted because the main hypothesis was for False Alarm rates. C (Paranoia) and D (Teleology) show the probability of correct identification of the sheep or wolf as a function of condition. E (Paranoia) and F (Teleology) show confidence as a function of condition. Dashed horizontal line represents correctness at random. Panels C–F showed collective data from Studies 3 (between-subjects; n = 200) and 4 (4a within-subjects, n = 102; 4b within-subjects direct replication, n = 87). Error bars in A and B represent standard errors of the mean. ***p < 0.001, **p < 0.01, and *p < 0.05. Colour palette by Daniel Barreto (dbarreto.com).
This study consisted of two preregistered studies: 4a and 4b, with the latter serving as a direct replication of the former—and with all details matching those of Study 3 except as noted. Participants saw 1 chase-present example trial and then were exposed to 5 practice trials with feedback (correct/incorrect) provided. After practice and instruction, test trials began with 35 chasing-present and 35 chasing-absent trials, and on each trial, participants completed the same wolf/sheep identification task from Study 3. The order of sheep-identification and wolf-identification tasks was randomized across participants. We calculated the sample size based on the results of Study 3 (Study 4a, https://aspredicted.org/wr7hb.pdf; preregistered on 17/10/2023), to detect a small-medium effect size (f2) of ~0.13 with a power of 80% and significance level of 0.05, we required at least 80 good participants. But when collecting the online data in two simultaneous batches we acquired 102 participants, thus, we analysed all of them (see below).
After running Study 4a, we noticed that some people in the high paranoia group had poor performance, but we had not pre-registered any exclusion criteria to account for this. To address this, we conducted a sensitivity analysis without participants with lower correctness, the results hold, and subsequently replicated the study with pre-registered exclusion criteria (Study 4b, https://aspredicted.org/4tz4d.pdf; preregistered on 22/10/2023). Participants were excluded if their performance on chase-present trials was worse than chance (using a one-tail binomial test against a null hypothesis of 1/8 correct level, with an α level of 0.1). Ninety-four participants finished the Study but only 87 passed the exclusion criterion. Qualitatively similar results were found excluding the poor-performing participants of Study 4a (Supplementary Sensitivity Analysis: “Study 4a (N = 86) using exclusion criteria from Study 4b”).
Revised Green et al. Paranoid Thought Scale (R-GPTS12): Participants responded to 18 items on a scale of 0 (Not at all) to 4 (Totally). Two subscales comprised the R-GPTS: referential delusions (e.g., “I often heard people referring to me”) and persecution delusions (e.g., “People wanted me to feel threatened, so they stared at me”). Paranoia is defined as the clinical cutoff > 10 in the persecution subscale12. We used paranoia (high = 1, low = 0) in the statistical models as we have done previously37.
Belief in the Purpose of Events (BPE36): Participants responded to 24 items, each of which asked them to picture a situation (e.g., “A person that you are attracted to kisses you in the middle of the street”)—and then presented them with a scenario (e.g., “You start going out together”). Subjects were asked to rate the extent to which they believed that the first situation happened for a purpose—on a scale of 1 (“The event definitely did not have a purpose”) to 5 (“The event clearly had a purpose”).
Cardiff Anomalous Perception Scale (CAPS38): Participants responded to 32 yes/no items, and for each positive item, they responded to 5 Likert questions relating to distress, distraction, and the frequency of the anomalous experiences (e.g., “Do you ever sense the presence of another being, despite being unable to see any evidence?”).
For Studies 1 and 2, we ran logistic models to predict paranoia (high = 1) and linear models to predict teleology (continuous). The regressors were behavioural variables taken from the task performance: sensitivity (d’), response criterion (C), false-alarm rate, and hit rate. The impact of those behavioural variables was measured by the slope (β) and a two-sided test with an α = 0.05. C and d’ were calculated as follows (see also ref. 39):
where p(h) and p(fa) are called hit rate and false alarm rate, respectively, and are calculated as follows:
where hit is detecting a chase in a chasing-present display, miss is failing to detect a chase in a chasing-present display, false alarm is detecting a chase in a chasing-absent display, and correct rejection is failing to detect a chase in a chasing-absent display.
We ran logistic mixed models to predict detection (correct = 1) for Studies 1 and 2, and predicting identification (correct = 1) for Studies 3, 4a and 4b. Also, we ran linear mixed models predicting confidence for Studies, 2, 3, 4a, and 4b. For each study, trials were nested in participants as random intercepts. When combining multiple studies in one model, participants were also nested in studies. The models also had the conditions (chase-present or chase-absent) for each participant as a random slope. All models had the condition (chasing-absent = 1; β1) as a regressor, paranoia (High = 1) or BPE (continuous; β2), and their interactions (β3). The logistic mixed models for Studies 1 and 2 were:
And for Studies 3, 4a and 4b, the logistic mixed models were:
Finally, when predicting confidence for Studies 2, 3, 4a, and 4b, the linear mixed models were:
In the main text, we presented the estimated coefficients [β], p values [p; two-sided p-values], and standard coefficient [Std. Coef.] with their 95% confidence intervals (95% CI) as effect sizes40. The exact effect sizes and 95% confident intervals for each Study can be found in Table S1. In order to fit linear mixed models, we assumed confidence was normally distributed.
To test whether correct identification and confidence in both chase-present and chase-absent trials explained unique variance of paranoia and teleology, we ran simple logistic (paranoia) and linear (teleology) regressions predicting the questionnaires with the four regressors: correct identification in chase-present (IdCP), correct identification in chase-absent (IdCA), confidence in chase-present (CoCP), and confidence in chase-absent (CoCA). As with the previous models, we reported β, two-sided p-values, and Std. Coef. with 95% CI. The models were:
For the exploratory analysis we used Bayesian Gaussian graphical models (BGGM), which estimates all the conditional (in)dependencies among the pre-selected variables. A BGGM allowed us to test effects (ρ), in form of posterior means with their 95% credible interval, while controlling for all possible relationships between a set of variables. This method uses matrix-F as prior distribution41. We used only the data from Studies 4a and 4b, given our intention to dissect the specific effects of wolf vs. sheep identification.
For the first BGGM, the preselected variables were the probability of correct identification in chasing-present of the wolf, of the sheep, paranoia, and teleology. For the second network, we incorporated anomalous perceptions (CAPS total scores). The third and fourth models were similar to the first and second, but instead of using identification, we used confidence in chase-absence trials for both wolf and sheep tasks. The reliable connections were the ones where the posterior mean was outside the 95% credible interval.
Finally, another advantage of these models is that we can estimate how likely the possible patterns are with respect to the partial correlations (ρ); H0(null): ρ = 0, H1: ρ > 0, H2: ρ < 0. Thus, we also used these probabilities to support the absence of an adjusted connection ρ. To correlate teleology and persecution and the BGGM interaction (Supplementary Fig. 1), we used Spearman correlations (given that the continuous variables did not meet normality criteria via Shapiro–Wilk tests).
All the statistics were conducted in R. Logistic and linear mixed models were conducted with the package lmerTest42. The estimated coefficients, p values, and standardized coefficients with their 95% confidence intervals as effect sizes were obtained with the report package40. Finally, the Bayesian Gaussian graphical models were fitted with the package BGGM41. Finally, we used the next packages for visualizations and data handling: ggplot243, ggpubr44, and dplyr45.
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
We used simple logistic regressions to predict paranoia (high = 1, low = 0, as defined by the clinically meaningful cut-off on R-GPTS12) and simple linear regressions to predict teleology (measured with the Beliefs in Purpose of Events [BPE] questionnaire36) with the scaled behavioural detection variables, including sensitivity (d’), response criterion (C), false alarm rate, hit rate, correctness, and reaction time (Fig. 3A and B). These analyses first revealed positive relationships between false alarm rates and both paranoia [β = 4.7, p < 0.001, Std. Coef. = 0.60, 95% CI (0.25, 0.97)], and teleology [β = 1.3, p = 0.021, Std. Coef. = 0.22, 95% CI (0.03, 0.4)]. As a result, the probability of chasing detection was linked positively with both paranoia [β = 5.64, p = 0.005, Std. Coef. = 0.53(0.16, 0.91)] and teleology [β = 1.54, p = 0.04, Std. Coef. = 0.19(0.01,0.38)]—and both were also associated with a lower response criterion, indicating a bias towards perceiving chasing [paranoia: β = −1.77, p = 0.006, Std. Coef. = −0.54(−0.95,−0.16); teleology: β = −0.52, p = 0.021, Std. Coef. = −0.22(−0.4,−0.03)]. Finally, paranoia was associated with poorer detection performance as indicated by a lower sensitivity [β = −1.02, p = 0.015, Std. Coef. = −0.51(−0.93,−0.11)], fewer correct trials [β = −6.39, p = 0.013, Std. Coef. = −0.47(−0.86,−0.10)], and lower confidence in chase-present trials [β = −0.73, p = 0.033, Std. Coef. = −0.48(−0.96,−0.05)]. These patterns are summarized graphically in Fig. 3A and B. No p-value was adjusted because the main hypothesis was for false-alarm rates, although we present all other behavioural indicators to have a full picture.
People with paranoia and excessive teleological thinking were also worse at identifying the wolf and sheep. We tested the interaction between condition (chasing-present or chasing-absent) and the questionnaire score (paranoia or teleology) with four logistic mixed models—two models for paranoia (one for each task: wolf or sheep) and two for teleology (wolf or sheep).
Interactions (depicted in Fig. 3C) revealed that compared to those with low paranoia, people with high paranoia were worse in chasing-present trials at selecting both the sheep [β3 = 1.29, p < 0.001, Std. Coef. = 0.53(0.41,0.65)] and the wolf [β3 = 0.83, p < 0.001, Std. Coef. = 0.33(0.21,0.45)]. And (Fig. 3D), reliable positive interactions were also observed between condition and teleology for the identification of both the sheep [β3 = 0.51, p < 0.001, Std. Coef. = 0.40(.28,.53)] and wolf [β3 = 0.41, p < 0.001, Std.Coef. = 0.33(0.22,0.45)]. In all four models, we found significant main effects of condition [Paranoia-Sheep β1 = −2.53, p < 0.001, Std. Coef. = −2.25 (−2.41,−2.08); Paranoia-Wolf β1 = −2.54, p < 0.001, Std. Coef. = −2.37 (−2.64, −2.10); Teleology-Sheep β1 = −3.73, p < 0.001, Std. Coef. = −2.25 (−2.46, −2.04); Teleology-Wolf β1 = −3.53, p < 0.001, Std. Coef. = −2.37 (−2.67, −2.08)]. And we also observed the main effects of questionnaires, implying that people with paranoia or excessive teleology were worse at ascribing roles to the discs [Paranoia-Sheep β2 = −1.30, p < 0.001, Std. Coef. = −0.54 (−0.65,−0.43); Paranoia-Wolf β2 = −0.94, p < 0.001, Std. Coef. = −0.38 (−0.48,−0.27); Teleology-Sheep β2 = −0.53, p < 0.001, Std. Coef. = −0.41 (−0.53,−0.30); Teleology-Wolf β2 = −0.47, p < 0.001, Std. Coef. = −0.39 (−0.50,−0.45)].
People with paranoia and excessive teleology exhibited higher confidence in chase-absent trials. As before, we tested the interactions between condition and the questionnaires with four linear mixed models [2 (questionnaire type: paranoia or teleology)×2 (condition: chase-present and chase-absent)]. We found that people with paranoia were more confident in chase-absent trials when selecting the sheep [β3 = 0.68, p < 0.001, Std. Coef. = 0.19 (0.14, 0.24)] or the wolf [β3 = 0.61, p < 0.001, Std. Coef. = 0.16 (0.11, 0.22)]; Fig. 3E]. In the case of teleology, we also found significant interactions between teleology and condition, with those high in teleology confidently misidentifying the role of the disc in the absence of chasing [sheep: β3 = 0.42, p < 0.001, Std. Coef. = 0.22 (0.17, 0.27)]; wolf: β3 = 0.33, p < 0.001, Std. Coef. = 0.18 (0.13, 0.24); Fig. 3F]. Hence both paranoia and excess teleology involved a confident mis-ascription of intentions in the absence of chasing —social hallucinations. These results, as well as the previous section (identification) holds when we used persecution as a continuous variable, instead of binarizing it in paranoia high and low (Supplementary Sensitivity Analysis: “Paranoia-persecution (from R-GPTS) as continuous”).
Finally, to test the possibility that the effect on paranoia and teleology is not a general effect on confidence, we regress confidence and the questionnaires to predict correct identification in chase-present trials. We found that confidence predicted correct identification, but more interestingly, we found a significant negative interaction between confidence and paranoia/teleology. Thus, confidence is indeed mis-calibrated in paranoia and high teleology (Supplementary Sensitivity Analysis: “Identification and Confidence”).
In this wide-format analysis, we tested the specificity of the behavioural variables when predicting either paranoia (binary) or teleology (continuous), while adjusting for correct identification in chase-present and chase-absent trials, as well as for confidence in chase-present and chase-absent trials (see “Methods” subsection “Unique variance and wide-format analysis” Eqs. (11) and (12)). We observed that lower correct identification in chase-present trials was associated with paranoia [β1 = −4.45, p < 0.001, Std. Coef. = −0.98 (−1.31, −0.68)] and teleology [β1 = −0.81, p < 0.001, Std. Coef. = −0.22 (−0.32, −0.12)] and higher confidence in chase-absent trials was also related to paranoia [β4 = 0.71, p < 0.001, Std. Coef. = 0.66 (0.25,1.07)] and teleology [β4 = 0.35, p < 0.001, Std. Coef. = 0.4 (0.28,0.53)], but no other regressor was significant, nor correct in chase-absent nor confidence in chase-present [Paranoia Model: correct identification in chase-absent β2 = −0.88, p = 0.7, Std. Coef. = −0.05 (−0.30,0.20), and confidence in chase-present β3 = 0.41, p = 0.16, Std. Coef. = 0.32 (−0.12,0.77); Teleology Model: correct identification in chase-absent β2 = −0.46, p = 0.38, Std. Coef. = −0.03 (−0.10,0.04), and confidence in chase-present β3 = −0.09, p = 0.15, Std. Coef. = −0.08 (−0.20,0.03)]. This supports the idea of social hallucinations, given that the same individuals displayed incorrect chase-present identification and higher confidence in chase-absent trials.
As described in methods and in preregistrations, the effects of interest from all mixed models were the interactions between questionnaire and condition. We display all studies by dependent variable (vertical panels) and coefficients (horizontal panels) in Fig. 4. In general, we found relevant interactions in the expected direction in all three dependent variables and for paranoia and teleology models. We next report the interactions β3 and effect sizes (Std. Coef.) see Eqs. (5)–(10) (more details in Supplementary Table 1).
Linear (Confidence) and logistic (Detect and Identification) mixed models’ results for each experiment and dependent variable. The y-axis shows the model’s estimates (or betas), and the x-axis the studies. Vertical panels are the types of dependent variables, binary detection for Studies 1 (n = 120) and 2 (n = 114), correct identification for Studies 3 (n = 200) and 4 (4a, n = 102; 4b, n = 87), and confidence for Studies 2, 3, and 4. Horizontal panels represent the relevant factor to interact with the Condition, either task type (sheep = 0, wolf = 1), paranoia (low = 0, high = 1), or teleology (continuous). All models had participants as random intercept and conditions as a random slope (see details in methods). Also, the effect of condition (chase-present = 1, chase-absent = 0) was estimated for all models. Size of circles represents the absolute estimate (larger = far from 0). “+” and “−“ signs represent that the 95 confident interval of the standardized effect size did not include the 0 (null effect).
The positive interaction in Studies 1 and 2 suggests more false alarms in paranoia [β3 = 0.14, p = 0.008, Std. Coef. = 0.14 (0.04, 0.24)], but this was not the case for teleology [β3 = 0.19, p = 0.062, Std. Coef. = 0.13 (−0.01, 0.27), Fig. 4, left panel “Detect”]. However, for paranoia this effect was only found in Study 1 [β3 = 0.93, p = 0.005, Std. Coef. = 0.21 (0.06, 0.36)], but not in Study 2 [β3 = 0.25, p = 0.14, Std. Coef. = 0.1 (−0.03, 0.24)].
For Studies 3 and 4 we tested if there was a difference in correct identification between wolf and sheep tasks (upper horizontal panel). There were no significant interactions when modelling correct identification [Study 3, β3 = −0.20, p = 0.077, Std. Coef. = −0.20 (−0.42, 0.02); Study 4a, β3 = 0.03, p = 0.771, Std. Coef. = 0.03 (−0.15, 0.20); Study 4b, β3 = −0.17, p = 0.08, Std. Coef. = −0.17 (−0.35,0.02)] nor confidence [Study 3, β3 = −0.03, p = 0.718, Std. Coef. = −0.02 (−0.14, 0.10); Study 4a, β3 = −0.03, p = 0.401, Std. Coef. = −0.02 (−0.07,0.03); Study 4b, β3 = 0.009, p = 0.83, Std. Coef. = 0.006 (−0.05, 0.06)] in any study (Fig. 4; upper row all panels). However, these models suggest an overall lower correct identification for the wolf task (Studies 3 and 4), as a main effect [β2 = 0.09, p = 0.027, Std. Coef. = −0.09 (−0.16, −0.01)].
When we tested the interaction between questionnaires and condition, we found that both paranoia and teleology were related to poorer correct identification in Study 3 [paranoia: β3 = 0.37, p = 0.018, Std. Coef. = 0.13(0.02, 0.25); teleology: β3 = 0.19, p = 0.015, Std. Coef. = 0.14 (0.03, 0.25)], Study 4a [paranoia: β3 = 1.71, p < 0.001, Std. Coef. = 0.8 (0.63, 0.98); teleology: β3 = 0.71, p < 0.001, Std. Coef. = 0.63 (0.43, 0.84)], and Study 4b [paranoia: β3 = 0.63, p = 0.011, Std. Coef. = 0.23 (0.05, 0.4); teleology: β3 = 0.34, p = 0.003, Std. Coef. = 0.26 (0.08, 0.43)] (Fig. 4; middle vertical panel). Also, for the confidence models, both questionnaires positively interacted with the condition in Study 4a [paranoia: β3 = 0.85, p < 0.001, Std. Coef. = 0.28 (0.2, 0.36); teleology: β3 = 0.50, p < 0.001, Std. Coef. = 0.31 (0.23, 0.38)] and Study 4b [paranoia: β3 = 0.71, p < 0.001, Std. Coef. = 0.17 (0.07, 0.27); teleology: β3 = 0.39, p < 0.001, Std. Coef. = 0.2 (0.1, 0.29)], but in Study 3, only teleology [β3 = 0.13, p = 0.037, Std. Coef. = 0.06 (0.00, 0.12)] but no paranoia [β3 = 021, p = 0.089, Std. Coef. = 0.05 (−0.01, 0.11)] interacted with confidence.
We found that male participants had lower false alarm rates, were better at discriminating chases, and had more distinct confidence between chase-present and chase-absent displays (Supplementary Fig. 2). We found that female participants (27%) had more paranoia in comparison than male participants (18%) [χ2(1) = 6.26, p = 0.012]. Also, female participants exhibit higher teleological thinking (Mean 3.03) than male participants (Mean 2.74) [t(536.7) = 4.31, p < 0.001, d = 0.37 (0.20, 0.54)].
People with paranoia and excess teleology scores were worse at correctly ascribing intentional roles to the wolves and sheep and were overly confident in those ascriptions—with a similar pattern of errors for both paranoia and teleology. However, subtler differences could well have been masked by the simple fact that persecution (the continuous variable underlying paranoia) and teleology were themselves highly correlated [ρ(690) = 0.34, p < 0.001]. As we and others have reported previously, people who believe that there is purpose in random events often also tend to believe that other agents have malintent towards them46 or that conspiracies are afoot11. So, to delineate the specific relations more carefully between the two identification tasks (wolf, sheep) and the two questionnaire scores (paranoia, teleology), we conducted the following analysis to dissociate any specific effect given that most of the variables are correlated.
To dissociate the highly correlated data set, we fit three Bayesian Gaussian Graphical models (BGGM41). Given that we are interested in how people perceive both the wolf and the sheep, we fit the within-subjects data from Studies 4a and 4b (N = 189). A BGGM allowed us to test for such effects (ρ), while controlling for all possible relationships between a set of variables. It also allowed us to quantify the evidence in favour one of the three possible patterns: H0: ρ = 0 (no connection), H1: ρ > 0 (positive connection), H2: ρ < 0 (negative connection). The results are summarized in Fig. 5, with connections not shown when the null effect is within the 95% Credible Interval. See all ρ for each within-subject study in Supplementary Table 2.
A Variables used were: paranoia (binary), teleology (continuous), correct sheep identification in chasing-present trials (continuous) and correct wolf identification in chasing-present trials (continuous). B Variables used were: paranoia (binary), teleology (continuous), confidence in sheep for chasing-absent trials (continuous) and confidence in wolf for chasing-absent trials (continuous). C Same patterns as A when we added anomalous perceptions (continuous), which load to both paranoia and teleology. D Same patterns as B when we added anomalous perceptions (continuous), which load to both paranoia and teleology. All the continuous variables were scaled by subtracting the mean and dividing by the standard deviation. The line thickness represents how far the estimate is away from 0 (null effect). And all the connections between nodes were estimated but not shown if the 0 (null effect) was inside the 95% credible interval. Green edges with a “+” sign are positive relations, whereas red edges with a “−” sign are negative relations.
We found a positive relation between paranoia and teleology [ρ = 0.36 (0.16, 0.54); p(H1) = 0.99; Fig. 5A]. However, we observed a sharp and categorical difference between the ways in which paranoia and teleology were specifically related to wolf vs. sheep identification in chase-present trials. On the one hand, paranoia was negatively related to the correct identification of the sheep [ρ = −0.43 (−0.59, −0.24); p(H2) ~ 1], but not of the wolf [ρ = −0.03 (−0.25, 0.18); p(H0) = 0.6]. On the other hand, teleology was negatively related to the correct identification of the wolf [ρ = −0.22 (−0.37, −0.07); p(H2) ~ 1], but not of the sheep [ρ = 0.08 (−0.09, 0.24); p(H0) = 0.74]. Correct identification of sheep and wolf were nevertheless positively related [ρ = 0.54 (0.42, 0.67); p(H1) ~ 1]. We also looked at the relation between paranoia, teleology, and confidence in chase-absent trials for sheep and wolf (Fig. 5B). Paranoia and teleology were also related [ρ = 0.27 (0.05, 0.46); p(H1) = 0.96], as well as sheep with wolf confidence [ρ = 0.79 (0.71, 0.85); p(H1) ~ 1]. Here confirming the dissociation, we found that paranoia was positively related to confidence when identifying the sheep [ρ = 0.28 (0.05, 0.48); p(H1) ~ 0.99], but not for wolf [ρ = 0.08 (−0.14, 0.31); p(H0) = 0.55]. However, we did not observe an exacerbated confidence for wolves in teleology.
We believe these behavioural effects reflect instances of social hallucinations, thus we explored whether they were related to anomalous hallucination-like percepts in everyday life, measured by CAPS38 (given CAPS is not normally distributed, tested with Shapiro–Wilk, we present Spearman correlations), as this self-reported questionnaire does not measure directly social perception. Unsurprisingly, we found that CAPS correlated with persecution [ρ(187) = 0.52, p < 0.001] and teleology [ρ(187) = 0.39, p < 0.001]. But also, it is negatively correlated with correct identification of wolf [ρ(187) = −0.22, p < 0.01] and sheep [ρ(187) = −0.28, p < 0.001] in chase-present trials, and positively correlated with confidence for wolf [ρ(187) = 0.25, p < 0.001] and sheep [ρ(187) = 0.26, p < 0.001] in chase-absent trials. When we added CAPS to the network (Fig. 5C), we found the same structure: paranoia-sheep [p(H1)~1] and teleology-wolf [p(H1) = 0.99], as in Fig. 5A. Interestingly, paranoia and teleology are not linked anymore [p(H0) = 0.39], but they both relate to CAPS [teleology: p(H1) ~ 1; paranoia: p(H1) ~ 1] (Fig. 5C). Anomalous perceptions were not related to either identification of wolf [p(H0) = 0.74] nor sheep [p(H0) = 0.75]. Same patterns of results when we add CAPS to the network with confidence in chase-absent trials (Fig. 5D). Paranoia positively relates with sheep [p(H1) = 0.99] and not with wolf [p(H0) = 0.67], as well as anomalous perceptions were linked with paranoia [p(H1) ~ 1] and teleology [p(H1) ~ 1].
In summary, the BGGM results suggest (i) two negative relationships on chasing-present trials—paranoia-sheep and teleology-wolf; (ii) two null relationships on chasing-present trials—paranoia-wolf and teleology-sheep; and (iii) null relations between anomalous perceptions and performance in chasing-present trials. To test explicitly the interaction between questionnaires and performance for chasing-present trials, we also correlated the difference between the probability of correct identification of sheep and wolf on chasing-present trials with the differences between persecution and teleology z-scores (Supplementary Fig. 2). We obtained a significant negative correlation [Spearman ρ(186) = −0.17, p = .02], which supports the same pattern: higher teleology is associated with worse wolf (but not sheep) identification, and higher paranoia is associated with worse sheep (but not wolf) identification.
Taken together, our data suggest that participants experience social hallucinations during visual perception—making confident errors in the perception of agency when none is present. Moreover, their errors depend on their specific concerns about the agency, and their proclivity to make errors depends on their propensity toward other types of hallucinations outside the laboratory.
Our key result was a striking connection between seemingly higher-level phenomena of reasoning (involving aberrant beliefs related to paranoia and teleological thinking) and a seemingly lower-level phenomenon of vision (involving the perception of animacy from motion). People with paranoia and excess teleology (i) were more likely to report perceiving chasing when there was none, (ii) had particular difficulties accurately identifying which moving disc (in a display filled with many distractors) was the ‘wolf’ or ‘sheep’, and (iii) were more confident in ascribing chasing when no chase was in fact present. High-confidence perceptual errors are considered an empirical hallmark of hallucinations17,18. And indeed, both types of identification errors (involving the ‘wolf’ and ‘sheep’) were correlated with reports of perceptual aberration and hallucination-like experiences in everyday life. On these bases, we suggest that aberrant beliefs about intentions may relate to social hallucinations in vision.
Our data further suggest that such connections may differ in important ways for different kinds of aberrant beliefs about intentions since we observed dissociations between paranoia and excessive teleology—despite their general relatedness. In particular, paranoia was more associated with impaired sheep identification and higher confidence in the absence of a chase, whereas teleology was more associated with impaired wolf identification. This pattern suggests that the relationship between perception and thought identified here may not be a simple matter of aberrant reasoning leading to a general improvement or impairment; rather, different kinds of aberrant thought about intentions may be associated with strikingly specific forms of social perception.
When we asked participants to ascribe particular roles in the chase, they were highly confident in the absence of real chasing. The high-confidence misattributions of agency reported here were robust and generalizable, across between- and within-subject designs, as well as sensitivity analyses (e.g., when removing participants due to correctness, the results hold). They occurred in both a pre-registered study (4a) and in a second pre-registered direct replication (4b). They held across multiple different regression models (both mixed models, and when collapsing to the participant level). Most directly, they were replicated multiple times (in Studies 3, 4a, and 4b; as detailed in Fig. 4).
Our results relating to paranoia were obtained with general online samples, but are, nevertheless, consistent with observations in clinical populations. People with schizophrenia are less able to discriminate different types of social vs. non-social animacy from motion47,48, particularly with ambiguous visual cues49. They use fewer mental-state words to describe what they perceive in such displays50, consistent with hypomentalizing, whereas our data are consistent with hypermentalizing in paranoia and teleology. However, the effects of schizophrenia may have been a function of cognitive deficits or motivational dysfunction, and thus may not necessarily reflect a relationship between unusual beliefs48,51. Nevertheless, if the animacy perception dysfunction in schizophrenia does involve visual processing, per se, this could be a potential and add-on explanation to why congenital blindness may protect from schizophrenia52,53.
The last point raises an interesting and controversial issue: is it aberrant bottom-up visual perception that fuels complex forms of higher cognition (such as teleology and paranoia)? Or is the causal influence the other way around, with higher cognitive configuration (paranoia and teleology) producing top-down changes in how the visual system processes the world? Our data cannot distinguish between these two alternatives. Some research suggests that cognition cannot affect perception54, while others have argued for cases of top-down influence in perception17,20. A third possibility would allow both views to coexist, where these social hallucinations are located somewhere between vision (bottom-up) and inference (top-down)55. While these chasing experiences have an irresistible perceptual quality, they are also experienced with a degree of confidence—and social hallucinations are high-confidence false alarms. Metacognition, as measured by confidence, has been shown to dissociate from performance56, such as false alarms.
Finally, we suspect that when participants were falsely alarmed, this was because they noticed a randomly moving object incidentally moving toward another object, and subsequently discounted moments in which that object did not head toward that other object, resulting in their maintaining an incorrect predictive model of the object’s movements. We agree that future work should test the mechanism of these false alarms more directly. Also, future work in this domain could localize deficits in the perception of animacy to particular psychotic symptoms, disorganized thinking, flat affect, or low motivation. The past work in schizophrenia has involved broad measures of perceived animacy47,48, but to our knowledge has never separately explored the identification of wolves vs. sheep in chasing displays. The key dissociation between paranoia and excessive teleology in the current project, however, suggests that people with schizophrenia may have particular difficulty with the identification of what is being chased (as opposed to what is chasing).
Cognitive behavioural therapy for psychosis may not alleviate all delusions57, but specific worry interventions do mollify paranoia58. Furthermore, paranoia and other delusions may have different underlying cognitive and perceptual mechanisms59. The present paradigm may provide a simple visual test to predict which patients may benefit from the worry intervention for paranoia, particularly focusing on the perceptual data that are taken as grounds for the delusion58.
There were also inconsistencies, however. In Study 2 (unlike Study 4a and 4b), we did not detect more high-confidence false alarms during chasing-absent trials in people with paranoia and excess teleology. This could have been a power issue since, when collapsed across studies (e.g., Study 1 and 2), we observed more false alarms, and the trend was in the expected direction even in the un-collapsed data.
Another limitation is that in all of the displays, the sheep were not fleeing the wolf explicitly. Thus, to obtain a full dissociation between teleology and paranoia at the experimental level, we may want in future work to develop displays where the sheep is reacting to the wolf, and actively attempting to escape. Finally, another interesting possibility for future experiments is to use a chasing task where participants directly control the actual movement of either the sheep5 or the wolf27. Treviño et al. 27. develop a technique to analytically dissect interceptive and pursuit strategies in chase behaviour—where interceptive strategies imply an estimation of the future location of the sheep (in a top-down manner), whereas pursuit strategies simply involve the current location of the sheep (in a bottom-up manner). Thus, we could hypothesize that in new chasing experiments where participants control the wolf, paranoia would be related to a pursuit strategy and teleology with an interceptive strategy.
In summary, paranoia and teleological beliefs are correlated with each other, and both involve the perception of agency when there is none—essentially social visual hallucinations. Yet they also involve distinct high-confidence misattributions of intention during social perception of animacy from motion. These different routes to hallucination in the same task setting highlight the role of different perceptual experiences and capabilities in the genesis and impact of anomalous beliefs. In the broadest sense, they suggest new links between seeing and thinking—with aberrant thought about the social world being connected to aberrant social perception, in ways not previously appreciated.
De-identified human data for all the studies and analysis scripts are accessible in our GitHub repository: https://github.com/santiagocdo/socialHallucinations [folder data].
Supporting Animations S1 and S2 found in https://github.com/santiagocdo/socialHallucinations [folder supporting_animation]. Analysis scripts are found in https://github.com/santiagocdo/socialHallucinations. The tasks scripts are available online: https://github.com/belieflab/perceivedAnimacy [branch chase (Study 1), chase-confidence (Study 2), and sheep-and-wolf (Study 4a and 4b)].
Heider, F. & Simmel, M. An experimental study of apparent behavior. Am. J. Psychol. 57, 243 (1944).
Article Google Scholar
Scholl, B. J. & Tremoulet, P. D. Perceptual causality and animacy. Trends Cogn. Sci. 4, 299–309 (2000).
Article PubMed Google Scholar
Gao, T., Newman, G. E. & Scholl, B. J. The psychophysics of chasing: a case study in the perception of animacy. Cogn. Psychol. 59, 154–179 (2009).
Article PubMed Google Scholar
McMahon, E. & Isik, L. Seeing social interactions. Trends Cogn. Sci. 27, 1165–1179 (2023).
Article PubMed PubMed Central Google Scholar
Gao, T., McCarthy, G. & Scholl, B. J. The wolfpack effect: perception of animacy irresistibly influences interactive behavior. Psychol. Sci. 21, 1845–1853 (2010).
Article PubMed Google Scholar
Hafri, A. & Firestone, C. The perception of relations. Trends Cogn. Sci. 25, 475–492 (2021).
Article PubMed Google Scholar
Schultz, J. & Frith, C. D. Animacy and the prediction of behaviour. Neurosci. Biobehav. Rev. 140, 104766 (2022).
Article PubMed Google Scholar
Isik, L., Koldewyn, K., Beeler, D. & Kanwisher, N. Perceiving social interactions in the posterior superior temporal sulcus. Proc. Natl Acad. Sci. USA 114, E9145–E9152 (2017).
Article PubMed PubMed Central Google Scholar
Schultz, J., Friston, K. J., O’Doherty, J., Wolpert, D. M. & Frith, C. D. Activation in posterior superior temporal sulcus parallels parameter inducing the percept of animacy. Neuron 45, 625–635 (2005).
Article PubMed Google Scholar
Pitcher, D. & Ungerleider, L. G. Evidence for a third visual pathway specialized for social perception. Trends Cogn. Sci. 25, 100–110 (2021).
Article PubMed Google Scholar
Wagner-Egger, P., Delouvée, S., Gauvrit, N. & Dieguez, S. Creationism and conspiracism share a common teleological bias. Curr. Biol. 28, R867–R868 (2018).
Article PubMed Google Scholar
Freeman, D. et al. The revised Green et al., Paranoid Thoughts Scale (R-GPTS): psychometric properties, severity ranges, and clinical cut-offs. Psychol. Med. 51, 244–253 (2021).
Article PubMed Google Scholar
Roberts, A. J., Wastell, C. A. & Polito, V. Teleology and the intentions of supernatural agents. Conscious Cogn. 80, 102905 (2020).
Article PubMed Google Scholar
Brüne, M. “Theory of Mind” in schizophrenia: a review of the literature. Schizophr. Bull. 31, 21–42 (2005).
Article PubMed Google Scholar
Langdon, R., Coltheart, M., Ward, P. B. & Catts, S. V. Mentalising, executive planning and disengagement in schizophrenia. Cogn. Neuropsychiatry 6, 81–108 (2001).
Article Google Scholar
Perner, J., Priewasser, B. & Roessler, J. The practical other: teleology and its development. Interdiscip. Sci. Rev. 43, 99 (2018).
Article PubMed PubMed Central Google Scholar
Corlett, P. R. et al. Hallucinations and strong priors. Trends Cogn. Sci. 23, 114–127 (2019).
Article PubMed Google Scholar
Haarsma, J., Deveci, N., Corbin, N., Callaghan, M. F. & Kok, P. Expectation cues and false percepts generate stimulus-specific activity in distinct layers of the early visual cortex Laminar profile of visual false percepts. J. Neurosci. JN-RM-0998-23 (2023) https://doi.org/10.1523/JNEUROSCI.0998-23.2023.
Schmack, K., Bosc, M., Ott, T., Sturgill, J. F. & Kepecs, A. Striatal dopamine mediates hallucination-like perception in mice. Science (1979) 372, 33–34 (2021).
Powers, A. R., Mathys, C. & Corlett, P. R. Pavlovian conditioning-induced hallucinations result from overweighting of perceptual priors. Science (1979) 357, 596–600 (2017).
Google Scholar
Haarsma, J. et al. Influence of prior beliefs on perception in early psychosis: effects of illness stage and hierarchical level of belief. J. Abnorm. Psychol. 129, 581 (2020).
Article PubMed PubMed Central Google Scholar
Fleming, L. M. et al. Using dimensionality-reduction techniques to understand the organization of psychotic symptoms in persistent psychotic illness and first episode psychosis. Sci. Rep. 13, 1–12 (2023).
Article Google Scholar
Bell, V. A Community of one: social cognition and auditory verbal hallucinations. PLoS Biol. 11, 1–4 (2013).
van Buren, B., Gao, T. & Scholl, B. J. What are the underlying units of perceived animacy? Chasing detection is intrinsically object-based. Psychon. Bull. Rev. 24, 1604–1610 (2017).
Article PubMed Google Scholar
Dittrich, W. H. & Lea, S. E. Visual perception of intentional motion. Perception 23, 253–268 (1994).
Article PubMed Google Scholar
Gao, T. & Scholl, B. J. Chasing vs. stalking: interrupting the perception of animacy. J. Exp. Psychol. Hum. Percept. Perform. 37, 669–684 (2011).
Article PubMed Google Scholar
Treviño, M., Medina-Coss y León, R., Támez, S., Beltrán-Navarro, B. & Verdugo, J. Directional uncertainty in chase and escape dynamics. J. Exp. Psychol. Gen. https://doi.org/10.1037/XGE0001510 (2023).
van Buren, B., Uddenberg, S. & Scholl, B. J. The automaticity of perceiving animacy: goal-directed motion in simple shapes influences visuomotor behavior even when task-irrelevant. Psychon. Bull. Rev. 23, 797–802 (2016).
Article PubMed Google Scholar
Scholl, B. J. & Gao, T. Perceiving animacy and intentionality: visual processing or higher-level judgment? In Social Perception: Detection and Interpretation of Animacy, Agency, and Intention (eds Rutherford, M. D. & Kuhlmeier, V. A.) 197–230 (MIT Press, Cambridge, MA, 2013).
Ji, M., Ward, E. J. & Green, C. S. Realistic and complex visual chasing behaviors trigger the perception of intentionality. PLoS ONE 18, e0284485 (2023).
Article PubMed PubMed Central Google Scholar
McMahon, E., Bonner, M. F. & Isik, L. Hierarchical organization of social action features along the lateral visual pathway. Curr. Biol. https://doi.org/10.1016/J.CUB.2023.10.015 (2023).
Bell, V., Raihani, N. & Wilkinson, S. Derationalizing delusions. Clin. Psychol. Sci. 9, 24–37 (2021).
Article PubMed Google Scholar
Gao, T., Scholl, B. J. & McCarthy, G. Dissociating the detection of intentionality from animacy in the right posterior superior temporal sulcus. J. Neurosci. 32, 14276–14280 (2012).
Article PubMed PubMed Central Google Scholar
Rushworth, M. F. S., Mars, R. B. & Sallet, J. Are there specialized circuits for social cognition and are they unique to humans? Curr. Opin. Neurobiol. 23, 436–442 (2013).
Article PubMed Google Scholar
Association, W. M. World Medical Association Declaration of Helsinki: ethical principles for medical research involving human subjects. JAMA 310, 2191–2194 (2013).
Article Google Scholar
Lindeman, M. & Aarnio, K. Superstitious, magical, and paranormal beliefs: an integrative model. J. Res. Pers. 41, 731–744 (2007).
Article Google Scholar
Suthaharan, P. et al. Paranoia and belief updating during the COVID-19 crisis. Nat. Hum. Behav. 5, 1190–1202 (2021).
Article PubMed PubMed Central Google Scholar
Bell, V., Halligan, P. W. & Ellis, H. D. The Cardiff Anomalous Perceptions Scale (CAPS): a new validated measure of anomalous perceptual experience. Schizophr. Bull. 32, 366–377 (2006).
Article PubMed Google Scholar
Castiello, S. et al. Benefiting from trial spacing without the cost of prolonged training: frequency, not duration, of trials with absent stimuli enhances perceived contingency. J. Exp. Psychol. Gen. https://doi.org/10.1037/XGE0001166 (2022).
Makowski, D., Ben-Shachar, M. S., Patil, I. & Lüdecke, D. Automated results reporting as a practical tool to improve reproducibility and methodological best practices adoption. R package, CRAN, https://cran.r-project.org/web/packages/report/index.html (2021).
Williams, D. & Mulder, J. BGGM: Bayesian Gaussian Graphical Models in R. J. Open Source Softw. 5, 2111 (2020).
Article Google Scholar
Kuznetsova, A., Brockhoff, P. B. & Christensen, R. H. B. lmerTest package: tests in linear mixed effects models. J. Stat. Softw. 82, 1–26 (2017).
Article Google Scholar
Wickham, H. Ggplot2: Elegant Graphics for Data Analysis (Springer-Verlag New York, 2016).
Kassambara, A. ggpubr: ‘ggplot2’ based publication ready plots. R package, CRAN, https://cran.r-project.org/web/packages/report/index.html (2020).
Wickham, H., François, R., Henry, L. & Müller, K. dplyr: a grammar of data manipulation. R package, CRAN, https://cran.r-project.org/web/packages/dplyr/index.html (2022).
Ongchoco, J. D. K., Castiello, S. & Corlett, P. R. Excessive teleological thinking is driven by aberrant associations and not by failure of reasoning. iScience 26, 1–16 (2023).
Roux, P., Passerieux, C. & Ramus, F. An eye-tracking investigation of intentional motion perception in patients with schizophrenia. J. Psychiatry Neurosci. 40, 118 (2015).
Article PubMed PubMed Central Google Scholar
Muthesius, A. et al. Preserved metacognition despite impaired perception of intentionality cues in schizophrenia. Schizophr. Res. Cogn. 27, 100215 (2022).
Article PubMed Google Scholar
Langdon, R., McMillan, J. & Gao, T. Detecting motion signals of intent in schizophrenia. Psychiatry Res. 219, 700–702 (2014).
Article PubMed Google Scholar
Langdon, R., Flynn, M., Connaughton, E. & Brüne, M. Impairments of spontaneous and deliberative mentalizing co-occur, yet dissociate, in schizophrenia. Br. J. Clin. Psychol. 56, 372–387 (2017).
Article PubMed Google Scholar
Lisøy, R. S. et al. Seeing minds—a signal detection study of agency attribution along the autism-psychosis continuum. Cogn. Neuropsychiatry 1–17 (2022) https://doi.org/10.1080/13546805.2022.2075721.
Morgan, V. A. et al. Congenital blindness is protective for schizophrenia and other psychotic illness. A whole-population study. Schizophr. Res. 202, 414–416 (2018).
Article PubMed Google Scholar
Pollak, T. A. & Corlett, P. R. Blindness, psychosis, and the visual construction of the world. Schizophr. Bull. 46, 1418–1425 (2020).
Article PubMed Google Scholar
Firestone, C. & Scholl, B. J. Cognition does not affect perception: evaluating the evidence for “top-down” effects. Behav. Brain Sci. 39, e229 (2016).
Article PubMed Google Scholar
Westfall, M. Other minds are neither seen nor inferred. Synthese 198, 11977–11997 (2021).
Article Google Scholar
Di Luzio, P., Tarasi, L., Silvanto, J., Avenanti, A. & Romei, V. Human perceptual and metacognitive decision-making rely on distinct brain networks. PLoS Biol. 20, 1–19 (2022).
Jauhar, S. et al. Cognitive-behavioural therapy for the symptoms of schizophrenia: systematic review and meta-analysis with examination of potential bias. Br. J. Psychiatry 204, 20–29 (2014).
Article PubMed Google Scholar
Freeman, D. et al. Comparison of a theoretically driven cognitive therapy (the Feeling Safe Programme) with befriending for the treatment of persistent persecutory delusions: a parallel, single-blind, randomised controlled trial. Lancet Psychiatry 8, 696–707 (2021).
Article PubMed PubMed Central Google Scholar
Rossi-Goldthorpe, R. et al. Different learning aberrations relate to delusion-like beliefs with different contents. Brain 139, 16–17 (2024).
Google Scholar
Download references
This project was funded by a grant from the Templeton Foundation to P.R. Corlett and B.J. Scholl; this grant funded J. Ongchoco and S. Castiello de Obeso postdocs. The funders had no role in study design, data collection and analysis, decision to publish or preparation of the manuscript. For assistance with running Study 3, we thank Minerva Pappu. For helpful comments and/or comments on previous drafts, we thank members of the Yale Belief, Learning, and Memory Laboratory. Study 1 was conducted in a graduate research visit at the Belief Lab in 2022 funded with a Study Visit Grant to S. Castiello de Obeso by the Experimental Psychology Society, UK.
Yale University, New Haven, CT, USA
Santiago Castiello, Brian J. Scholl & Philip R. Corlett
University of British Columbia, Vancouver, BC, Canada
Joan Danielle K. Ongchoco
The New School of Social Research, New York, NY, USA
Benjamin van Buren
PubMed Google Scholar
PubMed Google Scholar
PubMed Google Scholar
PubMed Google Scholar
PubMed Google Scholar
Conceptualization: P.R. Corlett and S. Castiello de Obeso. Data curation: S. Castiello de Obeso. Formal analysis: S. Castiello de Obeso and P.R. Corlett. Funding acquisition: B.J. Scholl and P.R. Corlett. Investigation: S. Castiello de Obeso, J.D.K. Ongchoco and P.R. Corlett. Methodology: S. Castiello de Obeso and P.R. Corlett. Project administration: P.R. Corlett. Resources: B.J. Scholl and P.R. Corlett. Software: S. Castiello de Obeso, J.D.K. Ongchoco, B. van Buren and B.J. Scholl. Supervision: B.J. Scholl and P.R. Corlett. Validation: S. Castiello de Obeso, J.D.K. Ongchoco, B. van Buren, B.J. Scholl and P.R. Corlett. Visualization: S. Castiello de Obeso and P.R. Corlett. Writing—original draft: S. Castiello de Obeso and P.R. Corlett. Writing—review & editing: J.D.K. Ongchoco, B. van Buren and B.J. Scholl.
Correspondence to Philip R. Corlett.
S. Castiello de Obeso, J.D.K. Ongchoco, B. van Buren and B.J. Scholl do not have any competing interests. P.R. Corlett is co-founder of Tetricus Labs, a precision psychiatry company. They did not fund this work.
Communications psychology thanks Joost Haarsma, Johannes Schultz and Philipp Sterzer for their contribution to the peer review of this work. Primary Handling Editor: Jennifer Bellingtier. A peer review file is available.
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.
Reprints and permissions
Castiello, S., Ongchoco, J.D.K., van Buren, B. et al. Paranoid and teleological thinking give rise to distinct social hallucinations in vision. Commun Psychol 2, 117 (2024). https://doi.org/10.1038/s44271-024-00163-9
Download citation
Received:
Accepted:
Published:
Version of record:
DOI: https://doi.org/10.1038/s44271-024-00163-9
Anyone you share the following link with will be able to read this content:
Sorry, a shareable link is not currently available for this article.
Provided by the Springer Nature SharedIt content-sharing initiative
Collection
Advertisement
Communications Psychology (Commun Psychol)
ISSN 2731-9121 (online)
© 2026 Springer Nature Limited
Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.