The Bodily Self

Abstract

The embodied self emerges from dynamic interactions between internal bodily signals and external sensory inputs, including those of social origin. Interoception (the perception and interpretation of internal bodily states) plays a central role in bodily self-consciousness and social cognition. Although socially salient contexts have been shown to modulate interoceptive accuracy (IAcc), the influence of passive social presence remains largely unexplored. In this study, we examined how the social exteroception modulates cardiac interoceptive accuracy using a heartbeat counting task in a Cave Automatic Virtual Environment (CAVE). Neurologically healthy participants (N = 32; ages 19–45) completed the task under both social and non-social exteroceptive conditions. Results showed that the mere presence of social exteroceptive cues significantly decreased IAcc. These findings indicate that even minimal social cues can influence internal bodily awareness, supporting the view that the embodied self is inherently relational and dynamically influenced by its social environment. This effect may reflect attentional competition between interoceptive and exteroceptive processes, consistent with predictive coding and attentional switching models. By leveraging immersive virtual reality, we created ecologically valid yet precisely controlled social contexts, minimizing the confounds of real-life physiological coregulation. Beyond theoretical implications, this study raises methodological considerations regarding the potential influence of social context, including experimenter presence, on interoceptive performance, and may inform clinical research on mental health conditions characterized by altered interoceptive and social processing.

Highlights

Passive viewing of social stimuli in VR could modulates perceived cardiac signals.

Individuals perform more poorly in the presence of a virtual other.

This study sheds new light on the role of the experimenter in interoception paradigms.

Introduction

The internal representation of the body, shaped through multiple feedback and feedforward loops, interacts with the experience of the external world, including the social world (Critchley & Garfinkel, 2018), to construct the self as an embodied agent (Christoff et al., 2011). This internal representation emerges notably through interoception which refers to the integration and interpretation of signals from within the body by the nervous system (Khalsa et al., 2018). In this study, we aimed to determine whether the external environment, and more specifically social context, influences interoceptive accuracy (IAcc), which captures our ability to detect internal sensations and track physiological states (Garfinkel et al., 2015).

Interoception, in conjunction with attention-demanding exteroception (Christoff et al., 2011), contributes to bodily self-consciousness and the distinction between self and other (Palmer & Tsakiris, 2018). The embodied self thus emerges from dynamic interactions between an acting organism and its environment, and is fundamentally shaped by interpersonal functioning (for philosophical reference: Heidegger’s being-with or Mitsein (Heidegger, 1962); for cognitive sciences and clinical references: (Gao et al., 2019; Graziano & Kastner, 2011; Legrand & Briend, 2015; Tajadura-Jiménez et al., 2012). Empirical research has shown that interoception and social cognition are tightly linked. On the one hand, interoceptive information modulates higher-order social cognition, such as social decision-making and emotional egocentricity bias (von Mohr et al., 2021; Mohr et al., 2023). On the other hand, social cognition modulates interoceptive accuracy, with prior work suggesting that the effects of social context on interoception are heterogeneous and depend on the nature of the social situation (Palmer & Tsakiris, 2018). For example, self-relevant social cues can enhance IAcc (Maister et al., 2017) and contexts involving social evaluation, such as being observed or exposed to direct gaze, have been associated with changes in interoceptive processing, potentially enhancing self-referential attention (Arnold et al., 2019; Durlik et al., 2014; Hazem et al., 2017; Isomura & Watanabe, 2020). In contrast, socially demanding or exclusionary contexts may impair interoceptive accuracy by reallocating attentional resources toward external monitoring and social threat detection (Durlik & Tsakiris, 2015). These findings indicate that different forms of social exteroception, ranging from active engagement to passive presence, may differentially modulate interoceptive processing.

From a broader theoretical perspective, symbolic interactionism and related philosophical traditions have long argued that the self develops through social exchanges and shared meaning-making, rather than existing as an isolated mental construct. This notion resonates with recent developmental and neuroscientific work emphasizing the Social Self First Hypothesis (SSFH) (Grossmann, 2025), which posits that the sense of self is inherently social from the outset of life. Neuroimaging evidence suggests that even infants recruit brain networks involved in self-referential and social cognition when processing self-relevant cues (Grossmann, 2015, 2025). The early emergence of the default-mode network, a key substrate of self-related processing, appears to be socially embedded, highlighting that the social environment from infancy scaffolds the self. Importantly, relational and socially scaffolded accounts of the self do not entail a unidirectional prediction regarding interoceptive accuracy. Instead, the impact of social context is likely to depend on how attentional resources are allocated between internal and external signals (Durlik & Tsakiris, 2015; Maister et al., 2017).

In the context of social exteroception, social presence refers to the subjective experience that another agent is present and accessible (Biocca et al., 2003; Short et al., 1978). This experience can arise even in the absence of direct interaction and includes passive co-presence, defined as the mere presence of another individual without interaction, evaluation, or explicit self-relevance. We refer to it as a “minimal phenomenal social experience”, defined as the simple perception of being in the presence of another individual, inspired by the concept of “minimal phenomenal selfhood” (Blanke & Metzinger, 2009).

To our knowledge, this minimal form of social experience has never been examined in relation to interoception, despite its ecological relevance in everyday contexts and its potential to engage attentional and self-referential processes. Notably, being aware of another’s presence, even without direct interaction, can trigger subtle physiological and attentional shifts (Durlik et al., 2014; Legrand & Briend, 2015), potentially redirecting cognitive resources away from interoceptive signals. Attending to social versus non-social stimuli recruits partially distinct but interconnected neural systems (Hurliman et al., 2005; Zuo et al., 2023), notably through the insular cortex (Blanke, 2012; Critchley et al., 2004), which integrates interoceptive and exteroceptive inputs to sustain bodily self-awareness.

In the present study, we specifically focus on passive social co-presence, and we propose that this form of minimal social context primarily engages exteroceptive attentional processes without reinforcing self-focused processing. Consistent with an attentional account of interoceptive–exteroceptive competition, we expected that cardiac interoceptive accuracy would decrease in the presence of social exteroceptive stimuli compared to non-social conditions, reflecting a shift in attentional resources toward externally oriented processing. While this prediction is compatible with broader models of cross-modal modulation between interoceptive and exteroceptive systems (Failla et al., 2020; Tajadura-Jiménez & Tsakiris, 2014; von Mohr et al., 2021), the present design primarily tests an attentional mechanism rather than these broader frameworks directly.

To examine this question under ecologically valid yet controlled conditions, we used virtual reality (VR) in a five-sided Cave Automatic Virtual Environment (CAVE). VR allows the creation of immersive and embodied experiences (Herbelin et al., 2016) while avoiding the physiological co-regulation that occurs in real dyadic settings (von Mohr et al., 2023), thereby ensuring experimental reproducibility. Participants performed a heartbeat counting task (HCT) either in the presence or absence of a 3D video of a human conspecific. We hypothesized that even passive exposure to social exteroceptive stimuli would modulate cardiac interoception through psychophysiological mechanisms involving attention and self-related processing.

Methods

Participants

33 volunteers partook in the study; one was excluded from the analyses as they were familiar with the actor in the social video (Tsakiris et al., 2011). The final sample consisted of 32 volunteers (50% male) aged 19–45 years (Table 1, M = 25.19, SD = 5.82).

Table 1.

Demographic and neuropsychological information of the sample

Age (years)	Mean (± SD)	n
	25.19 (5.82)	32
Gender (% females)	50	32
MAIA		31
Noticing	3.60 (0.89)
Not-Distracting	2.27 (1.04)
Not-Worrying	3.14 (0.88)
Attention Regulation	3.06 (0.80)
Emotional Awareness	3.82 (0.70)
Self-Regulation	2.59 (0.94)
Body Listening	2.9 (1.03)
Trusting	3.64 (0.80)
Total score	25.08 (2.94)
EDI-2		32
Interoceptive awareness	3.67 (4.08)
STAI		32
Total score	41.53 (10.42)
TAS20		32
Total score	50.40 (10.67)
AQ		31
Total score	15.19 (6.51)
VR Questionnaire		32
Q1: Presence of Video	5 (0)
Q2: Surprised by social stimuli	1.46 (0.80)
Q3: Disturbed by social stimuli	1.78 (1.03)
Valence for social stimuli	5.21 (0.97)
Arousal for social stimuli	4.12 (1.94)

MAIA-2 multidimensional assessment of interoceptive awareness 2, EDI-2 eating disorder inventory-2, STAI state-trait anxiety inventory, TAS20 Toronto alexithymia ccale TAS-20, AQ autism quotient, Q virtual reality environment questions (cf. methods part)

Sample size was evaluated with a priori power analysis based on the most similar interoception paradigm where effect sizes were reported (Maister et al., 2017) (blank-screen vs. partner(familiar)-screen condition; d = 0.56). The analysis was designed to detect a within-subject difference in interoceptive accuracy between social and non-social conditions using a paired-sample comparison. To achieve a power of 85% at the 0.05 alpha level (two-tailed), 31 participants were required. To compensate for eventual data loss, 33 participants were recruited. Most of the participants were university students. All participants met the following criteria: ability to participate in the totality of the experimental procedure, normal or corrected-to-normal vision and good general health, not using psychotropic substances or drugs.

Heartbeat Counting Task

The present study used PsychoPy3 (Peirce et al., 2019), an open-source local application that provides an experiment builder and Python programming library. The source code is available on Gitlab.

The Heartbeat counting task (HCT) was used to evaluate cardiac interoceptive accuracy (Schandry 1981): participants were instructed to silently start counting their own heartbeat on an auditory start cue, until they received an auditory stop cue. After a brief training session, the actual experiment started. Four different time intervals of 100 s, 45 s, 35 s and 25 s were presented in random order across participants. Seated participants were asked to report orally the number of heartbeats counted at the end of each interval (Fig. 1). Subjects were not allowed to monitor their pulse or try any other physical manipulations that might facilitate the detection of heartbeats. Moreover, they were asked to count only heartbeats that they felt and not to guess (“It is very important that you only count the heartbeats you really feel, without trying to guess your heart rate” (Desmedt, Luminet, et al. 2020)). Heart rate was monitored with a wireless sensor: each participant was instructed to wear the Empatica E4 wristband (Empatica Srl, Milan, Italy) around the wrist of the non-dominant hand. The wristband is a wearable, non-invasive, research device (Milstein and Gordon 2020), which allows for real-time physiological data collection such as blood volume pulse, measured using a photoplethysmogram sensor and from which Heart Rate (HR) (HR = 60/Interbeat Interval, obtained by blood volume pressure from photoplethysmogram sensor). Data from the sensor was synced with the HCT paradigm by time triggers.

Fig. 1.

Schematic representation of the social heartbeat counting task in a computer automatic virtual environment. (Top) Representation of the social [person sitting on a chair] and non-social [empty chair] visual exteroceptive condition projected into the five screens of the cave automatic virtual environment. (Bottom) Timeline of the cardiac interoceptive task. This procedure was repeated over four different time intervals (25, 35, 45 and 100 s) and presented in a random order according to the two conditions

Participants also completed a control task as described by Nicholson and colleagues (Nicholson et al., 2019), in which they were asked to count the number of seconds of three different elapsed time intervals (19, 37, and 49 s). In each trial, reported and actual seconds were compared to calculate an index of time estimation using the following equation: 1 - (|nseconds_real - nseconds_reported|)/(nseconds_real). The resulting accuracy scores were averaged across the three time intervals. Task performance was then correlated with HCT to determine whether participants relied on strategies such as time estimation during the heartbeat counting process (the script is available here).

Experimental Procedure and Context Stimuli

First, participants were seated in a chair facing a CAVE screen positioned outside the typical range of peripersonal space (Ferri et al., 2013) and equipped with goggles. Social and non-social contexts (i.e., visual exteroceptive conditions) were presented through videos. In the social context, participants viewed a video of a person with a neutral expression (male or female, non-significant other (Maister et al., 2017; Terasawa et al., 2014) seated on a chair and reading, simulating a “face-to-face dyadic interaction” (von Mohr et al., 2023). To minimize bodily self-awareness, the person in the video did not make eye contact with the participant (Baltazar et al., 2014; Isomura & Watanabe, 2020) and exhibited minimal movement. The non-social condition featured a video of the same background without any person present (see Figs. 1 and 2). To reduce surprise effects and heart rate reactivity (Reisenzein et al., 2019), participants viewed approximately 20 s of each video prior to beginning the experimental tasks, with no information given about the experiment’s objectives. The experimenter remained outside the CAVE, out of view of the participant. The videos were displayed continuously throughout each condition, including periods when participants reported their heartbeat counts. Following this familiarization phase, participants performed the HCT and control task within either the social or non-social context (i.e., visual exteroceptive conditions), with the order of conditions counterbalanced across participants.

Fig. 2.

The CAVE and participants’ view during the experiment. Video example of the CAVE can be found here https://www.imagin-vr.com/sante

The CAVE (Fig. 2), built by imagin-VR (https://www.imagin-vr.com/), consisted of 5 large screens (2.55 m high and 1.88 m wide) on which 2D images were rear-projected. The spatial resolution was approximately 18 pixels per degree, the horizontal field of view (FOV) was 180° to 270°, and the vertical FOV was approximately 40°. The images were computed and projected at a refresh rate of 120 Hz. The participant’s point of view (location and eye height) was estimated from head data (from a chair) and used to render the visual scene accordingly. For audio feedback, a spatial sound system with loudspeakers was placed on the CAVE structure.

Self-Report Questionnaire

Several questionnaires were evaluated, particularly because differences in behavior have been associated with differences in interoceptive processing (Desmedt et al., 2022; Jenkinson et al., 2024; Khoury et al., 2018; Williams et al., 2022).

To assess interoceptive sensibility (Garfinkel et al., 2015; Khalsa et al., 2018) and body preoccupation, participants completed the Multidimensional Assessment of Interoceptive Awareness 2 [MAIA-2] and the interoceptive awareness scale of the Eating Disorder Inventory 2 [EDI-2].

Contrasted research indicated that alexithymia, anxiety and autism are all linked to variability in interoceptive processing (Garfinkel et al., 2016; Shah et al., 2016), although recent meta-analysis has failed to return significant associations between HCT performance and either trait anxiety or alexithymia (Desmedt et al., 2022). This is the reason why we also included a series of self-reported questionnaires to assess levels of anxiety (the State-trait Anxiety Inventory), alexithymia (the 20-item Toronto Alexithymia Scale TAS-20), and self-reported autistic traits (measured by the AQ: Autism Quotient) (Baron-Cohen et al., 2001). Nevertheless, although widely used, the AQ is a self-report questionnaire and may be affected by response biases and limited introspective accuracy. It may also not fully capture the heterogeneity of autistic traits (Lin et al., 2026; Lombardo et al., 2026) and overlaps with related constructs such as alexithymia and anxiety. Finally, as a trait measure rather than a diagnostic tool, its associations with interoceptive accuracy should be interpreted with caution.

Additionally, to better comprehend participants’ experience of the VR environment, a VR questionnaire was used: participants answered a Likert scale (from 1 to 5) on three questions: Q1: “Did you notice the person in the video?”, Q2: “Were you surprised by their behavior?”, Q3: “Did you feel disturbed by their presence?”. Moreover, two seven-point Likert scales were used to rate the valence and arousal dimensions of participants’ own emotional states during the VR social environment.

Data Analysis

Interoceptive accuracy during heartbeat counting was calculated as the average accuracy across the four different time intervals (HCT; n = 4) using the following formula (Garfinkel et al., 2015; von Mohr et al., 2021):

IAcc = 1/n ∑(1 – (|recorded heartbeats – counted heartbeats|)/(recorded heartbeats – counted heartbeats)/2).

Hence, IAcc was calculated per participant and condition (social/non-social) and varied between − 1 (low IAcc) and 1 (High IAcc). Consistent with previous studies (Garfinkel et al., 2015; von Mohr et al., 2021), we included counted heartbeats in the denominator to mitigate overestimation of human performance.

Statistical Evaluation

To compare IAcc between conditions (social vs. non-social), we used a Wilcoxon signed-rank test (paired, non-parametric) complemented with rank-biserial correlation effect sizes for non-parametric differences.

To go further in the understanding of IAcc, although linear mixed-effect model analysis (LMM) and Wilcoxon signed rank test are two means to test the same hypothesis, we used LMM as the primary analysis to simultaneously test for the influence of covariates. Participants were used as a random intercept and Conditions, time estimation and gender (Desmedt et al., 2022; Prentice & Murphy, 2022) as fixed-effect factors. The time estimation score was added to our model as a covariate to control for potential confounds associated with the HCT.

To test whether recorded heartbeats were similar in the “social” (e.g., due to increased arousal) and “non-social” conditions, and to ensure that changes in heartbeat estimates rather than changes in recorded heartbeats were the main driver of the condition effect, we used another and separate LMM where participants were used as a random intercept and recorded heartbeats as a fixed effect factor.

In an exploratory analysis, to assess the relationship between interoceptive measure (in social and non-social contexts) and self-report measures (from the questionnaires), we used correlations with Pearson coefficient (or Spearman’s correlation as a non-parametric version of Pearson’s correlations). Bonferroni correction was used to correct significance level for multiple tests.

Statistical analyses were performed in R (http://cran.r-project.org). We used the R package lme4/lmer (Kuznetsova et al., 2017) to test the integrative LMM for main effects and interactions by approximating restricted maximum likelihood estimation to compare cardiac IAcc between conditions. Significance level was set at 0.05.

Results

Demographic Data

In the current work, we report data from 32 participants. Table 1 summarizes demographic and neuropsychological data.

Cardiac Interoceptive Accuracy

In non-social context and social context, mean IAcc scores were 0.31 (SD = 0.42) and 0.22 (SD = 0.45), respectively (Fig. 3). The Wilcoxon signed rank exact test testing the difference between both conditions showed that IAcc in the non-social context was significantly higher than in the social context (W = 376.00, p = 0.036; r (rank biserial) = 0.42, 95% CI [0.06, 0.69]).

Fig. 3.

Cardiac interoceptive accuracy during the HCT across the two exteroceptive modalities. Cardiac IAcc observed for the whole sample. Dots correspond to individual data points, while the boxplots represent the median and standard deviations for each condition

The linear mixed model analysis (statistical data are provided in Table 2) showed that context (social/non-social) modulated significantly IAcc, b = 0.04, SE = 0.02, p = 0.045, 95% CI [8.75e-04, 0.08]. No significant gender and time estimation parameters were found (respectively: b = −0.12, SE = 0.07, p = 0.12, 95% CI [−0.27, 0.03] and b = 0.25, SE = 0.55, p = 0.65, 95% CI [−0.85, 1.36]), suggesting that participants’ HCT performance was not related to gender and ability to estimate time.

Table 2.

Linear mixed model results on the cardiac IAcc

Parameter					95% CI
	Estimate	SE	t	P	Lower	Upper
Fixed Effect
Intercept	0.07	0.45	0.14	0.88	−0.84	0.97
Conditions (S/NS)	0.04	0.02	2.04	0.045	0	0.08
Gender	−0.12	0.07	−1.55	0.12	−0.27	0.03
Time estimation	0.25	0.55	0.45	0.65	−0.85	1.36
Random Effect
Participants	Variance	0.16	sd	0.40

Number of observations = 64. Number of groups = paired, 32. Significant main effects are highlighted in bold. Conditions (S/NS) are dummy coded, with the first level (non-social [NS] condition) as the reference category. The estimate represents the difference between conditions, with its sign indicating the direction of the effect

The effect of context (social/non-social) on recorded heartbeats was statistically non-significant (b = 0.48, SE = 0.01, p = 0.853), suggesting that the condition effect was effectively driven by changes in heartbeat estimates.

Self-Report Questionnaire Results

Table 1 shows statistics of self-reports for all participants.

Relations Between IAcc and Report

Association between cardiac interoceptive accuracy and self-report questionnaires failed to reach significance after Bonferroni correction. Bonferroni correction for the computations of the multiple variable correlations was applied with an α level = 0.003 (i.e. 0.05/13; as 13 correlations were computed). The correlation coefficients and corresponding p-values are presented in Table 3.

Table 3.

Pearson/spearman correlation coefficients between the cardiac interoceptive accuracy score and the reports of participants

MAIA	IAcc non-social score	n	Iacc social score	N
		31		31
Noticing	−0.152		0.1
Not-Distracting	−0.01		−0.11
Not-Worrying	0.15		0.32
Attention Regulation	−0.06		−0.04
Emotional Awareness	0.24		0.29
Self-Regulation	−0.02		−0.11
Body Listening	0.37*		0.38*
Trusting	−0.15		−0.05
Total score	0.2		0.25
EDI-2		32		32
Interoceptive awareness	0.27		0.11
STAI		32		32
Total score	−0.11		−0.12
TAS20		32		32
Total score	0.27		0.16
AQ		31		31
Total score	0.21		0.08

IAccinteroceptive accuracy, MAIA multidimensional assessment of interoceptive awareness, EDI-2 eating disorder inventory-2, STAI state-trait anxiety inventory, TAS20 Toronto alexithymia scale TAS-20, AQ autism quotient; *p < 0.05, **p < 0.01 (before bonferroni correction)

Discussion

The present study examined how minimal social experience cues modulate cardiac interoceptive accuracy using immersive VR. As predicted, interoceptive accuracy decreased in the presence of social exteroceptive information. This effect is most parsimoniously explained by an attentional account, whereby the presence of socially relevant stimuli biases attentional resources toward exteroceptive processing at the expense of interoceptive signals (Arnold et al., 2019). However, it remains possible that increased perceptual and attentional load in the social condition, rather than social processing per se, contributed to the observed reduction in interoceptive accuracy. These findings extend our understanding of how minimal social contexts influence the bodily self through changes in sensorimotor integration, and raise important methodological questions regarding the influence of experimenter presence in interoception paradigms.

Implication for Fundamental Science

Bodily self-consciousness arises from the integration of interoceptive and exteroceptive sensory inputs. Our findings suggest that interoceptive accuracy (IAcc) can be modulated by the mere presence of another individual, even in the absence of active interaction. The observed effect (b = 0.04) appears modest, suggesting that the modulation of interoceptive accuracy by social context is subtle rather than large in scale. This is consistent with the notion that interoception is sensitive to contextual and attentional influences, which may induce relatively small but reliable shifts in performance. While the impact of social environments on interoception has often been assumed, it remains empirically underexplored. Prior studies have shown that social situations can transiently alter interoceptive ability (Durlik & Tsakiris, 2015; Maister et al., 2017), likely through attentional switching between internal and external sensory domains (Arnold et al., 2019).

The present findings are best accounted for by an attentional framework, in which externally oriented processing competes with access to internal bodily signals. Within this perspective, predictive coding and free-energy frameworks (Ainley et al., 2016) may be understood as providing a broader theoretical context in which such attentional reallocations are implemented, even if not directly tested by the present design. In this view, interoceptive-exteroceptive competition may reflect precision-weighting processes, whereby the increased salience of external stimuli leads to a relative down-weighting of interoceptive signals.

Our results align with this attentional account, showing that passive social presence reduced IAcc, consistent with the idea that attention was reallocated towards socially salient external cues. Importantly, this interpretation remains compatible with a more general perceptual-load account, whereby increased external stimulation, rather than social content specifically, competes with interoceptive processing (Arnold et al., 2019). This effect parallels previous findings on emotional egocentricity bias and self-other processing (von Mohr et al., 2021) and supports the concept of alteroception; the interpenetration of self- and other-related perceptions (Palmer & Tsakiris, 2018). In contrast, other studies, such as Isomura & Watanabe (Isomura & Watanabe, 2020), reported enhanced IAcc during exposure to direct gaze, suggesting that specific social cues (e.g., direct eye contact) can facilitate rather than hinder interoceptive awareness. These discrepancies likely reflect differences in the nature of the social experience tested: while direct gaze may increase self-referential attention, passive presence in VR might instead diffuse attentional focus across external sensory channels. Together, these findings illustrate the nuanced and bidirectional nature of the relationship between social exteroception and bodily self-awareness.

From a psychophysiological perspective, the observed modulation of IAcc by social presence may involve activation of the anterior insula, anterior cingulate cortex, and amygdala, regions central to both interoceptive awareness and social cognition (Critchley et al., 2004; Khalsa et al., 2018; Zuo et al., 2023). These structures integrate visceral afferents with social and emotional cues, thereby coordinating physiological arousal and attentional focus. Subtle changes in autonomic balance, induced by perceived social presence, could thus transiently reduce interoceptive sensitivity.

Developmentally, the integration of bodily self-related signals precedes and scaffolds representations of others, which themselves rely on multisensory self-based encoding (Durlik & Tsakiris, 2015; Hazem et al., 2017; Maister et al., 2017; van Elk & Blanke, 2011). However, complementary theoretical accounts emphasize that the self may also be fundamentally shaped by social input from early in development. In particular, the SSFH (Grossmann, 2025) posits that self-representation is inherently social from the outset of life, supported by early recruitment of brain networks involved in both self-referential and social processing (Grossmann, 2015, 2025). Extending this framework to adult embodiment, the present findings suggest that passive social co-presence increases the salience of exteroceptive inputs, thereby shifting this balance away from interoceptive processing and reducing access to internal bodily signals (Zuo et al., 2023). Rather than reflecting a unidirectional effect, this pattern supports the idea that the self is continuously shaped by dynamic interactions between interoceptive and exteroceptive inputs. Our findings suggest that even minimal social cues can modulate bodily self-awareness, reinforcing the notion that the self is inherently relational and socially situated. This interpretation resonates with symbolic interactionism, which conceives the self as continuously formed through real or imagined interactions with others. From this perspective, interoception and exteroception are not isolated perceptual systems but dynamically co-regulated components of an embodied and social mind. Importantly, this co-regulation is bidirectional: relational accounts (including symbolic interactionism and the SSFH) suggest socially salient cues may enhance interoceptive accuracy, whereas minimal social contexts may reduce it through attentional competition between exteroceptive and interoceptive processing.

Another implication of our study concerns the role of the experimenter in interoception paradigms. This raises the intriguing possibility that interoceptive accuracy (IAcc) could be modulated by the mere presence of the experimenter in the experimental room, beyond the well-known Pygmalion or experimenter effects (Rosenthal, 1994). Future research should further investigate the relationship between IAcc and exteroceptive social processing, considering the potential influence of the experimenter’s presence.

Possible Implication for Mental Health Conditions

The evidence for atypical IAcc in mental health conditions (Khalsa et al., 2018) was not consistently in support of either hypo- or hyper- IAcc (for example in autism: (Garfinkel et al., 2016; Shah et al., 2016), even in meta-analyses, the highest level of evidence synthesis (Ahn & Kang, 2018), with contrasting results (Desmedt et al., 2022; Jenkinson et al., 2024; Khoury et al., 2018; Williams et al., 2022).

One possible explanation is that these inconsistencies reflect differences in the relative weighting of interoceptive and exteroceptive signals across contexts and populations (Noel et al., 2018; Yang et al., 2022). In this framework, the present findings suggest that even passive social context can shift this balance, by increasing attentional allocation toward external cues at the expense of interoceptive processing. Accordingly, conditions characterized by altered orientation to social stimulation may show differential sensitivity to such contextual modulation. For example, heightened sensitivity to social cues (as in anxiety-related hypervigilance to social evaluation) may amplify the reduction in interoceptive accuracy in socially present contexts, whereas reduced sensitivity or atypical orienting toward social stimuli (as in autism spectrum conditions) may attenuate or abolish such effects, leading to more comparable performance across social and non-social contexts. In addition, discrepant findings may reflect the influence of the experimenter’s presence in the experimental setup, which may vary across mental health conditions.

Within interoception paradigms, the presence of others, as a form of exteroceptive input, has been largely overlooked in laboratory settings. Yet, altered social cognitive processes are a common feature across multiple psychiatric conditions and are highlighted in several transdiagnostic frameworks for understanding mental disorders (American Psychiatric Association, 2013; Insel et al., 2010). Furthermore, as said before, it is well established that others contribute to the multisensory foundations of self-representation (Durlik & Tsakiris, 2015; Hazem et al., 2017; Maister et al., 2017). Although not observed here through linear correlations with self-report questionnaires, the contrasting interoceptive accuracy (IAcc) results may reflect a form of “minimal phenomenal social experience” elicited during the IAcc paradigm, particularly in mental health populations. As suggested by Failla and colleagues in the context of autistic people, the integration of interoceptive signals with social information may serve as a distinguishing feature in autism, even prior to the dynamic interplay between interoception and the salience of external social cues (Failla et al., 2020; Yao et al., 2018).

While these interpretations remain speculative, they suggest that passive social context manipulations, such as those used here, may provide a useful experimental approach to probe how social environments interact with interoceptive processing in both typical and clinical populations. Future research should investigate the relationship between social cues and interoception in individuals with mental health conditions to see if the manifestation of clinical symptoms is also driven by this imbalance.

Limitations

An important limitation concerns the use of a single interoceptive measure, the Heartbeat Counting Task (HCT), restricted to the cardiac domain. Although the psychometric validity of the HCT has been debated (Corneille et al. 2020), it remains widely used and comparable to other heartbeat perception tasks. Moreover, alternative methods remain scarce and often face similar limitations (Desmedt et al. 2023; Garfinkel et al. 2022). Importantly, the HCT provides an indirect index of interoceptive accuracy and may be influenced by factors beyond pure interoceptive sensitivity, such as counting strategies, prior beliefs about heart rate, and demand characteristics (Desmedt, Corneille, et al. 2020). To minimize biases, participants were explicitly instructed to report only perceived heartbeats rather than estimating their heart rate (Desmedt, Luminet, et al. 2020), and performance on the HCT was not correlated with time estimation ability, which partly mitigates, though does not fully eliminate, the contribution of such higher-order cognitive processes. While the inclusion of a time estimation control helps account for general temporal estimation abilities, it cannot entirely rule out these influences, and the present findings should therefore be interpreted with these measurement-specific limitations in mind. The relatively lower mean IAcc scores compared to previous studies (von Mohr et al. 2021) may stem from the fact that our task was conducted directly within a virtual environment, which could slightly alter interoceptive focus. In addition, the number of trials was limited to reduce fatigue and maintain attentional engagement within the immersive VR environment, as prolonged exposure may impact cognitive and interoceptive processing (Bohil et al. 2011). While some protocols include a larger number of intervals, previous studies have shown that reliable estimates of interoceptive accuracy can be obtained with fewer trials when interval durations are varied (Desmedt, Corneille, et al. 2020). Future research may systematically investigate the impact of trial number on interoceptive measures, particularly in immersive settings.

A key methodological limitation lies in the potential confounding between the social and non-social conditions regarding perceptual complexity. The two videos differed not only in social content (presence vs. absence of a person) but also in visual richness and motion dynamics, which may have influenced attention and thereby IAcc. Nevertheless, both stimuli were filmed under identical environmental, lighting, and auditory conditions from the same perspective, which minimizes, but does not entirely eliminate, low-level perceptual differences. The present work should therefore be considered a proof of concept illustrating that interoceptive accuracy can be modulated by the mere presence of another person, rather than a definitive dissociation between social and perceptual factors. Future studies should incorporate control conditions equating perceptual features such as motion, luminance, and spatial complexity, for example, by using dynamic but non-social stimuli (e.g., a moving object) to disentangle social from purely visual or attentional influences. This refinement would clarify whether the observed modulation of cardiac interoception originates specifically from social exteroceptive processing or from general cognitive load and perceptual demands. A conceptual limitation of the present study concerns our operationalization of “minimal phenomenal social experience”. While this notion was intended to capture a basic form of social awareness based on passive co-presence, it does not fully align with the second-person neuroscience framework, which emphasizes that social cognition is fundamentally rooted in dynamic, reciprocal interactions rather than passive observation (Schilbach et al., 2013). Nevertheless, the strength of our approach lies precisely in isolating a minimal, non-interactive form of social presence, allowing us to test whether low-level social cues, independent of reciprocity or interaction, are sufficient to modulate interoceptive processing. As such, our findings complement second-person approaches by suggesting that even in the absence of interactive dynamics, the mere perception of another individual may be enough to influence bodily self-awareness.

Immersive VR itself may influence interoceptive processing independently of social content. VR environments are known to alter multisensory integration through changes in the weighting of visual, vestibular, and proprioceptive signals (Bohil et al., 2011; Kilteni et al., 2012). Given the close functional links between vestibular and interoceptive systems, which jointly contribute to bodily self-consciousness (Serino et al., 2013), such alterations may impact the perception of internal bodily states via overlapping neural substrates, including the insular cortex (Blanke, 2012; Critchley et al., 2004). Future studies could include non-VR control conditions, or designs involving a real person physically present but not visible or non-interacting, or a social agent outside the participant’s field of view, to better disentangle social-specific effects, reduce perceptual confounds, and improve ecological validity.

Conclusions

The present study showed that minimal social exteroceptive information, manipulated through VR, can influence cardiac interoception. These findings support the idea that being an embodied agent, composed of afferent information from within one’s body, is influenced by the social environment. Through this lens, they also question the role of the experimenter in interoception paradigms.

Biographies

Julia Verbe

is a Hospital Physician affiliated with the Autism and Neurodevelopment team at the iBraiN laboratory (Inserm U1253, University of Tours, France). Her research focuses on neurodevelopment, cognitive neuroscience, and the study of brain mechanisms underlying perception and social cognition.

Raphaël Gautier

is a PhD candidate in the Autism and Neurodevelopment team at the iBraiN laboratory (Inserm U1253, University of Tours, France). His research focuses on interoception, cardiac signal processing, and electrophysiological markers of brain function, with a particular interest in heartbeat-evoked potentials and their role in cognition.

Marianne Latinus

is a researcher at the iBraiN laboratory (Inserm U1253, University of Tours, France). Her research investigates the neural mechanisms underlying auditory and social perception, with a particular interest in voice processing, multisensory integration, interoception, and autism spectrum disorder.

Frédérique Bonnet-Brilhault

is Professor of Child and Adolescent Psychiatry at the University of Tours and Head of the Child Psychiatry Department at Tours University Hospital. She is a senior researcher at the iBraiN laboratory (Inserm U1253), where her work focuses on autism spectrum disorder, neurodevelopment, and the development of innovative diagnostic and therapeutic approaches.

Frédéric Briend

is an Assistant Professor at the University of Tours and a researcher in the Autism and Neurodevelopment team at the iBraiN laboratory (Inserm U1253). His research interests include cognitive neuroscience, psychophysiology, interoception, and the neural mechanisms underlying autism and schizophrenia.

Author Contributions

Julia Verbe: Conceptualization, Formal analysis, Writing – review & editing. Raphaël Gautier: Investigation. Marianne Latinus: Writing – review & editing. Frédérique Bonnet-Brilhault: Supervision. Frédéric Briend: Conceptualization, Methodology, Resources, Writing – original draft, statistics, Supervision.

Funding

This research received no external funding.

Data Availability

The datasets and code generated and/or analysed during the current study are available in the OSF repository (https:/osf.io/w8mur/?view_only=18e9f9b5b84542e5a5caf5715a7ece85).

Declarations

Ethics Approval and Consent to Participate

All study procedures were in accordance to the Declaration of Helsinki and approved by the French ethics committee “Comité de Protection des Personnes” (PROSCEA2017/23; ID RCB: 2017-A00756-47). Each participant received an information letter about the study. In accordance with the Helsinki declaration, written informed consent was obtained from each participant before inclusion.

Competing interests

The authors declare no competing interests.