Real-Time Embodied Experience Shapes High-Level Reasoning Under Altered Gravity

  1. Hélène Grandchamp des Raux
  2. Tommaso Ghilardi
  3. Elisa Raffaella Ferrè
  4. Ori Ossmy

  • Curated by eLife

    eLife logo

    eLife Assessment

    This important study investigates whether high-level physical reasoning is grounded in real-time bodily and vestibular signals using an innovative combination of virtual tool-use tasks and galvanic vestibular stimulation. The evidence is solid for the specific conclusion that ongoing vestibular information can influence selected aspects of physical reasoning in a context-dependent manner, while is incomplete for the broader mechanistic interpretations.

Reviewed by

Read the full article

Listed in

Log in to save this article

Abstract

A critical aspect of human cognition is the ability to use our knowledge about the laws of physics to make predictions about physical events. Whether this ability is based on abstract processes or is grounded in our body-environment interactions remains an open debate. We used physical reasoning under altered gravity as a model system to show that humans’ real-time embodied experience modifies their high-level physical reasoning. Specifically, we tested participants in computerised reasoning games, while disrupting their gravitational signalling using Galvanic Vestibular Stimulation (GVS). Participants failed more and had suboptimal strategies under the GVS condition compared to no-GVS in games requiring reasoning about terrestrial gravity. However, the effects of GVS were reduced when the games included reasoning about altered gravity. Our findings demonstrate how the physical experience of the body shifts high-level cognitive skill as reasoning, suggesting that humans’ mental representation of the world is grounded in adaptable physical mechanisms.

Article activity feed

  1. Version published to 10.7554/elife.111166.2 on eLife
  2. Version published to 10.7554/elife.111166 on eLife
  3. eLife

    eLife Assessment

    This important study investigates whether high-level physical reasoning is grounded in real-time bodily and vestibular signals using an innovative combination of virtual tool-use tasks and galvanic vestibular stimulation. The evidence is solid for the specific conclusion that ongoing vestibular information can influence selected aspects of physical reasoning in a context-dependent manner, while is incomplete for the broader mechanistic interpretations.

  4. eLife

    Reviewer #1 (Public review):

    Summary:

    This study investigates a fundamental question in cognitive science: is our ability to reason about the physical world an abstract mental process, or is it "embodied"-directly rooted in our real-time physical interactions with the environment? The authors compared participants' performance in computerized reasoning games with and without Galvanic Vestibular Stimulation (GVS). They suggest that participants failed more often and utilized suboptimal strategies under GVS compared to a sham stimulation condition. Furthermore, they suggest that this detrimental effect of GVS was reduced when the games were governed by altered gravity (hyper- and hypo-gravity). Consequently, the authors conclude that the physical experience of the body modifies high-level cognitive skills, such as reasoning.

    Strengths:

    The manuscript is well-written, organized, and easy to follow. Also, combining a specialized physical reasoning task with real-time vestibular disruption (GVS) is an intriguing approach to testing the boundaries of embodied cognition.

    Weaknesses:

    The empirical evidence supporting the conclusions, as currently stated in the Abstract, remains weak. The overall effect of GVS is limited, and the conclusions regarding its impact on strategy use rely on sophisticated clustering and classification methods whose reliability and validity are not clearly demonstrated in the current manuscript. The differences between Studies 1 and 2, which are critical to establishing the effect of altered gravity, are also not robust overall.

  5. eLife

    Reviewer #2 (Public review):

    Summary

    The paper investigates whether the real-time physical experience of the body shapes high-level physical reasoning. Participants played a set of computerized tool-use reasoning games (the Virtual Tools paradigm) in which they must use knowledge of physical laws - including gravity, collisions, and inertia - to guide a ball into a target area. In Study 1, participants played the games under terrestrial gravity while receiving either Galvanic Vestibular Stimulation (GVS), which introduces noise into the vestibular organ and disrupts gravitational signalling, or a Sham condition with matched skin sensation. In Study 2, a separate cohort played the same games redesigned under hypogravity (0.5 g - half Earth g) or hypergravity (2 g - double Earth g), again with concurrent GVS or Sham stimulation. Performance was assessed through success rate, number of attempts, and time per attempt; strategy was assessed through the spatial distance between successive tool placements and the frequency of tool switching across attempts. A post-hoc gravity-weighted index (GWI) was computed to compare the effect of vestibular perturbation across the two studies. The main finding is that GVS impairs performance in gravity-dependent games under terrestrial gravity, yet the same perturbation appears to be neutral or even beneficial when the game environment involves non-terrestrial gravity - a result the authors interpret as evidence for an adaptable, body-grounded internal model of physics.

    Strengths

    One of the most notable strengths of this work is its conceptual positioning at the intersection of embodied cognition and physical reasoning. Rather than treating the human body either as an abstract information-processing device or as a purely biomechanical system, the authors take seriously the idea that cognition is scaffolded by ongoing sensorimotor state - and they test this idea with a paradigm that is both tractable and theoretically motivated. The use of the Virtual Tools paradigm is well-suited to this goal: the games vary systematically in their reliance on gravitational predictions, allowing selective impairment (rather than general disruption) to serve as a signature of embodied physical reasoning.

    The dual-study design is another strength. Testing the same vestibular perturbation under terrestrial and altered game-gravity conditions and observing a reversal in its effect depending on context, provides a form of internal control that is conceptually compelling. The additional clustering analyses (Dirichlet Process Gaussian Mixture Model and leave-one-out kernel density classification) strengthen the strategy results beyond raw distance measures, confirming that GVS systematically shifts participants' spatial exploration strategies.

    The paper is also clearly written and engages meaningfully with relevant theoretical frameworks - predictive coding, embodied cognition, and stochastic resonance - making it accessible and stimulating for a broad audience.

    Weaknesses

    (1) Absence of multiple-comparisons correction. A large number of game-level pairwise t-tests are conducted in both studies (upward of twenty per study) without correction for familywise error rate. The game-level effects that anchor the main narrative - in Study 1 alone: Remove, GoalMove, Spiky, Falling_A, Shafts_B, Gap, and Chaining - arise from an uncorrected pool of comparisons. The probability that some of these constitute false positives is non-trivial. The authors should apply a correction (e.g., Benjamini-Hochberg) or at minimum discuss this limitation explicitly.

    (2) The facilitation claim rests on a post-hoc and arbitrarily parameterized index. The gravity-weighted index (GWI), which drives the central cross-study comparison uses integer coefficients (1, 2, 3) to weight games by gravity dependency level. These coefficients are entirely arbitrary and bear no principled relationship to the actual gravitational magnitudes used in the study. Why not use the gravity dependency ratings themselves, or the empirically estimated gravity impact scores from the computational modelling mentioned in the Methods? The choice of weights should be either principled or tested across a range of values to demonstrate robustness. Furthermore, the notation in equation (1) as currently typeset reads as "Gravity minus Weighted Index" rather than "Gravity-Weighted Index"; this should be corrected.

    (3) The "facilitation" interpretation exceeds what the data in Study 2 directly support. Across all games in Study 2, GVS versus Sham differences in absolute performance are non-significant in all directions. The facilitation claim derives entirely from the GWI being higher in Study 2 than in Study 1 - a between-subjects comparison involving different participant groups and a non-pre-registered metric. The language of "facilitation" should be tempered accordingly, or the authors should provide additional analyses to support this framing.

    (4) Gravitational manipulation is visual only, and the vestibular system is only one component of the gravity-sensing network. Gravity perception results, as the authors very well know, from a distributed multisensory integration process that involves, in addition to the vestibular system, visual, proprioceptive, and visceral inputs. The present paradigm manipulates gravitational context solely through visual cues and targets the vestibular system through GVS - a point the authors acknowledge but do not discuss in sufficient depth. It is important to distinguish clearly between real gravitational alterations (as achieved in parabolic flight or centrifuge environments, where the entire body is physically subjected to a different gravitational vector) and virtually altered gravity, where only one sensory modality is targeted while others remain anchored to 1 g. The scope of the conclusions should reflect this distinction.

    (5) The choice of 0.5 g and 2 g may lack sensitivity. Combining the two altered-gravity conditions in Study 2 because no significant effect of hypo versus hypergravity was found is statistically pragmatic but conceptually unsatisfying. There is evidence in the space physiology literature that gravitational processing is not linearly symmetric around 1 g: threshold effects exist below and above terrestrial gravity that may not be captured by modest deviations (half and double g) - see refs below. It is worth discussing whether the absence of a hypo/hyper distinction in Study 2 reflects a genuine equivalence or a lack of sensitivity, and whether more extreme conditions (e.g., near-zero g or 4-5 g) might reveal different processing regimes. Whether 0.5 g and 2 g were sufficient to saturate the system or merely insufficient to perturb it remains an open question with direct implications for the interpretation of the null GWI effects on strategy measures.

    Lee SMC, Ribeiro LC, Martin DS, Zwart SR, Feiveson AH, Laurie SS, Macias BR, Crucian BE, Krieger S, Weber D, Grune T, Platts SH, Smith SM, and Stenger MB. Arterial structure and function during and after long-duration spaceflight. J Appl Physiol (1985) 129: 108-123, 2020.

    de Winkel KN, Clément G, Groen EL, and Werkhoven PJ. The perception of verticality in lunar and Martian gravity conditions. Neurosci Lett 529: 7-11, 2012.

    Clément G, Moore ST, Raphan T, and Cohen B. Perception of tilt (somatogravic illusion) in response to sustained linear acceleration during spaceflight. Exp Brain Res 138: 410-418, 2001.

    Benson AJ, Kass JR, and Vogel H. European vestibular experiments on the Spacelab-1 mission: 4. Thresholds of perception of whole-body linear oscillation. Exp Brain Res 64: 264-271, 1986.

    (6) High-level reasoning is not defined with sufficient precision. The term "high-level reasoning" appears from the title onward and in the heading of the Study 1 results section (line 138), but it is never formally defined. The reader needs a clearer account of what distinguishes high-level physical reasoning from low-level sensorimotor prediction, and where the games used here fall along that continuum. What specific physical competencies - ballistic trajectories, free-fall predictions, collision dynamics, frictional forces, inertial effects - are required across the game set? When describing the subset of games that drive key effects, this information is critical for evaluating whether effects are specific to gravity reasoning or to some other physical concept.

    (7) Performance measures are disconnected from underlying kinematics. The performance measures (success rate, number of attempts, time per attempt) are coarse, high-level summaries. Time per attempt is used as a proxy for performance efficiency, yet participants received no instructions regarding speed, and different individuals may have adopted systematically different speed-accuracy trade-offs. It would be valuable to know whether time per attempt correlates with attempt number within a given game (which would indicate within-game learning) and whether mouse movement data - trajectory, velocity, hesitation - were recorded and could be analysed to provide more mechanistic insight into strategy formation.

    Comments on revised version.

    My overall assessment remains positive. I consider the findings important and the evidence solid. The study uses an original experimental approach to address a theoretically significant question, and I consider the data sufficient to support the more constrained conclusion that ongoing vestibular information can influence physical reasoning in a context-dependent manner.

    I agree with the other reviewers that the game-level multiple comparisons require appropriate statistical treatment and that the Gravity-Weighted Index should not carry more inferential weight than justified by its construction. I also agree that some of the broader mechanistic interpretations should be presented cautiously. These issues concern primarily the strength and scope of interpretation rather than the validity of the experimental findings themselves.

    I do not, however, consider the absence of preregistration of the exploratory analyses to constitute a methodological weakness in itself. Preregistration is desirable when analyses are intended to provide confirmatory evidence, but it is not a prerequisite for a scientifically informative exploratory analysis. The important distinction is that exploratory analyses should be transparently identified as such and their conclusions should be appropriately qualified. In my view, the relevant concern here is therefore the inferential weight placed on these analyses, rather than the fact that they were not preregistered.

    I would consequently retain a rating of solid for the strength of evidence rather than incomplete. The main claims are supported provided they are formulated at an appropriate level of specificity; the evidence is less secure for the broader mechanistic claims concerning general recalibration of internal models of gravity.

  6. eLife

    Reviewer #3 (Public review):

    Summary

    This manuscript investigates whether high-level physical reasoning is influenced by ongoing vestibular and bodily signals rather than operating as a fully abstract process. The authors combine Galvanic Vestibular Stimulation (GVS) with the Virtual Tools task to test the effects of vestibular perturbation during reasoning under terrestrial gravity in Study 1 and visually simulated altered gravity in Study 2. The experimental approach is original and provides a potentially useful bridge between research on vestibular processing, physical cognition, and embodied reasoning.

    The revised manuscript is substantially improved. The authors have responded carefully to many of the methodological concerns raised in the previous review. In particular, they have removed the original arbitrarily weighted Gravity-Weighted Index, introduced mixed-effects models that account for participant-level and game-level variability, clarified the basis of the gravity-impact classification, and more clearly distinguished preregistered from exploratory analyses. The interpretation of the strategy measures and the treatment of the 0.5g and 2g conditions have also been improved. These revisions provide a substantially stronger and more transparent analytical framework for evaluating the hypotheses.

    Strengths

    The principal strength of the study is its conceptual and methodological originality. Combining vestibular stimulation with a relatively high-level physical problem-solving task provides an innovative approach to testing whether ongoing bodily information is associated with physical reasoning. The comparison of games that differ in their dependence on gravity is also potentially informative for distinguishing general effects of vestibular perturbation from effects more specifically associated with gravity-relevant reasoning.

    The revised analyses represent another important strength. The mixed-effects models provide a more principled basis for inference than reliance on individual game-level comparisons alone, while the revised Gravity Index removes the arbitrary weighting used in the original manuscript. The clearer treatment of exploratory analyses also makes it easier to distinguish the primary evidence from complementary observations. Overall, the revision has addressed many of the methodological concerns raised previously and has made the empirical pattern considerably easier to evaluate.

    Weaknesses

    The main remaining concern is the correspondence between the strength of the empirical evidence and the strength of the manuscript's central claims.

    In Study 1, the aggregate measures provide little evidence for an overall effect of GVS on performance or strategy. More importantly for the proposed gravity-specific account, the newly introduced mixed-effects analyses do not show significant stimulation-by-gravity-impact interactions for the principal performance or strategy measures. Some gravity-level-specific and game-level effects are present, including a difference in placement distance for high-gravity-impact games, and the exploratory spatial analyses provide additional evidence that vestibular perturbation may influence aspects of strategy. These findings are potentially informative, but the overall pattern is selective rather than a consistent and convergent demonstration that greater dependence on gravity systematically increases the detrimental effect of GVS.

    The evidence in Study 2 is more limited. Aggregate differences between GVS and Sham are small across the main performance and strategy measures, and the mixed-effects analyses do not identify reliable stimulation effects across the low-, medium-, or high-gravity-impact categories. Study 2 therefore does not directly demonstrate that GVS facilitates physical reasoning under altered gravity.

    The cross-study Gravity Index analyses provide some evidence that the effects may differ across gravitational contexts, but the pattern remains heterogeneous. For success rate, the high-gravity-impact contrast is in the predicted direction but remains a directional trend, with the credible interval including the point of equivalence. Other performance measures do not provide convergent support for the same pattern. The distance index shows reliable cross-study contrasts at some gravity-impact levels, but these form a crossover pattern rather than a simple reduction of the GVS effect as gravity dependence increases. The principal study-by-gravity-impact interactions are also not robust. In addition, this Gravity Index analysis was not preregistered and compares two separate participant samples. Taken together, these findings are more consistent with suggestive evidence for a context-dependent influence of vestibular perturbation than with a definitive demonstration of facilitation or rapid adaptation under altered gravity.

    A second concern is the mechanistic interpretation. The experiment directly measures behavioral performance and strategy under GVS and Sham conditions. It does not directly measure the strength of a terrestrial gravity prior, prediction errors, sensory reweighting, internal-model updating, or stochastic resonance. Predictive coding and related frameworks provide plausible interpretations of the findings, but the present behavioral data cannot distinguish among these mechanisms. Statements that GVS downregulates a terrestrial gravity prior, accelerates revision of an internal model, recalibrates cognitive expectations, or provides a vestibular "teaching signal" therefore imply a level of mechanistic specificity that is not directly established by the current experiments.

    Overall, the study remains innovative and potentially valuable, and the revision has substantially strengthened the analytical framework. The empirical results support a more circumscribed conclusion that vestibular perturbation can influence selected aspects of physical reasoning and strategy, with suggestive evidence that these effects may vary across gravitational contexts. The evidence is less conclusive for the broader claims that real-time embodied experience generally determines high-level physical reasoning or that vestibular perturbation facilitates rapid adaptation through specific mechanisms of gravity-prior updating.

  7. eLife

    Author response:

    The following is the authors’ response to the original reviews.

    Reviewer #1 (Public review):

    (R1.1) Lack of Overall Effects and Inflated Type I Error for Game-Level Effects For Study 1, to identify condition differences at the game level (i.e., Figure 2), the authors effectively conducted 70 independent t-tests (5 measures × 14 games). While 7 significant results were reported, this large number of independent tests invites an inflated Type I error rate, as no multiple-comparison correction appears to have been applied. A similar inflation is expected in Study 2, where 50 independent t-tests (5 measures × 10 games) yielded 5 significant comparisons (Figure 4). Although the authors might argue the direction of the differences is systematic, implying GVS generally impairs performance, at least one significant comparison shows the opposite effect: tool switching indicates that GVS led to better performance for the 'Table_A' game in Study 2 (Figure 4d), whereas the same variable indicated GVS led to worse performance in Study 1 (Figure 2d). I suspect that none of the significant game-level results would survive a proper statistical correction. If possible, the authors can redo statistical testing with corrections (FDR or Bonferroni) or with LMM using game as a random effect. Before proper statistical analyses, I strongly encourage the authors to refrain from drawing broad conclusions based on these isolated game-level results.

    We thank the reviewer for this important statistical point. We have revised the manuscript to clarify that each game was treated as a distinct physical-reasoning problem, and these were used to localize the games in which the stimulation effect was expressed, not to compare games with one another or to draw broad conclusions from isolated contrasts (and therefore, in this case, no multiple-comparison corrections are required; lines 637-640). That said, to address the reviewer’s concern directly and to strengthen our conclusions in the paper, we also added mixed-effect models using gravity impact level and stimulation type as fixed factors and participant and game as random effects for each study (lines 168-177, 202-207, 257-259, 301-303, 641-651, Tables 2 and 4).

    (R1.2) Furthermore, when analyzing data across all games, the study found no significant effect of GVS on overall performance or strategy measures in either Study 1 or Study 2. This lack of an aggregate effect contradicts the authors' conclusion that participants failed more often or utilized suboptimal strategies under GVS.

    We thank the reviewer for raising this issue. We agree that the analyses collapsed across all games did not reveal significant overall effects. However, this pattern is informative rather than contradictory: the full-game analysis combining gravity-dependent problems with games in which gravity does plays little or no role in the solution (8 low-gravity-impact games in Study 1 and 4 in Study 2; Extended Data Table 1). If GVS had produced a general dizziness-related, physiological, attentional, or arousal effect, we would expect a broad impairment across the task. Instead, the revised mixed-effects models show that the effects were tied to gravity-relevant contexts: in Study 1, GVS was associated with a trend-level increase in attempts for medium-gravity-impact games and a significant increase in placement distance for high-gravity-impact games. These effects were not observed in Study 2. Together with the cross-study Gravity Index analyses, this pattern strengthens the interpretation that GVS did not cause a general impairment, but selectively altered physical reasoning as a function of the gravitational context. We have now clarified this point in the manuscript (lines 398-403).

    (R1.3) Missing Rationale for Classification Analysis.

    It is puzzling why the authors pursued two exploratory analyses on tool placement after revealing that the two related primary measures (tool positioning and switching) did not generate significant condition differences in Study 1. These additional analyses-the Dirichlet Process Gaussian Mixture Model and leave-one-out classification-were not pre-registered. In the absence of overall condition differences, the authors appear to be "doubling down" by applying sophisticated classification tools to the raw data without a clear prior rationale.

    We thank the reviewer for this important comment. We agree that the rationale for these additional analyses needed to be made clearer. Our aim was not to “double down” on the primary measures, but to test a distinct question that the preregistered summary measures cannot address. The distance and tool-switching measures reduce each participant’s behaviour to two averages; they therefore do not capture the multivariate structure of spatial placements, whether participants move between distinct placement strategies across successive attempts, or whether the overall placement distributions differ systematically between stimulation conditions. Our analyses, therefore, included these two complementary analyses, adapted from prior Virtual Tools work (Allen et al., 2024), to characterise the spatial and sequential organisation of tool-use strategies. The Dirichlet Process Gaussian Mixture Model is well suited to this question because it identifies strategy clusters directly from x/y placement coordinates without imposing an a priori number of strategies, allowing us to quantify transitions between spatial strategies across attempts (in relation to the environment in the game which is not taken into account in the more basic measures). The leave-one-out kernel-density analysis then asks a related but independent question: whether the placement patterns contain sufficient condition-specific structure to classify stimulation condition above chance. These analyses therefore add value by testing the organisation of strategy use at a level not captured by the primary summary measures. We have revised the manuscript to clarify this rationale and to present these analyses as complementary evidence about strategy organisation, while keeping the mixed-effects models as the primary inferential basis for the manuscript (lines 216-218, 721-731, 752-753).

    (R1.4) Insufficient Evidence for the Reduced Effect of GVS Under Altered Gravity… To compare Study 1 and Study 2, the authors devised a "gravity-weighted index," but its definition is not sufficiently justified. The index assigns weights of 1, 2, and 3 to low-, medium-, and high-gravity-dependent games, respectively. The choice of these specific weights appears arbitrary, making the quantitative results difficult to interpret. More importantly, there is no citation or explanation regarding how these three levels of "gravity impact" were defined in the first place (Line 468). This index was also not pre-registered.

    We thank the reviewer for this important comment. We agree that the original gravity-weighted index was not sufficiently justified because it combined the three gravity-impact levels, using numerical weights of 1,2, and 3. Although these weights followed the ordinal classification of the games, they were not independently validated and could therefore make the cross-study comparison difficult to interpret. To address this concern directly, we removed the gravity-weighted index entirely. In the revised manuscript, we instead use a Gravity Index that is calculated separately for each measure and each gravity-impact level, as the ratio between performance or strategy outcomes under GVS and Sham. This removes the arbitrary weighting scheme and provides a direct estimate of the stimulation effect within each gravity-impact category.

    We then compared these Gravity Indexes between Study 1 and Study 2 using mixed-effect models that include study and gravity-impact level as fixed factors, rather than collapsing across gravity-impact levels with our defined weights. We used participant as random factor. Because the indexes are bounded and showed clustering at structural values, we analysed them using Bayesian cumulative logit models, which are appropriate for bounded, non-normally distributed outcomes and allow the cross-study contrasts to be estimated without imposing a linear weighting structure. We also clarified how the gravity-impact categories were defined, based on piloting and previous computational modelling work in the lab, and report the classification of each game in Extended Data Table 1. These revisions directly address the reviewer’s concern by replacing the arbitrary weighted composite with a transparent, category-specific analysis of the GVS effect across terrestrial and altered-gravity contexts, in line with frameworks outlined by James et al. (2021), Li et al. (2023), Liu et al. (2017) (lines 140-147, 276-292, 307-317, 588-594, 658-666, 692-717, Table 5, Fig 5, Extended data Fig 3).

    James, N. T., Harrell Jr, F. E., & Shepherd, B. E. (2021). Bayesian cumulative probability models for continuous and mixed outcomes. arXiv preprint arXiv:2102.00330.

    Li, C., Tian, Y., Zeng, D., & Shepherd, B. E. (2023). Asymptotic properties for cumulative probability models for continuous outcomes. Mathematics, 11(24), 4896.

    Liu, Q., Shepherd, B. E., Li, C., & Harrell Jr, F. E. (2017). Modeling continuous response variables using ordinal regression. Statistics in medicine, 36(27), 4316-4335.

    Manuguerra, M., Heller, G. Z., & Ma, J. (2020). Continuous ordinal regression for analysis of visual analogue scales: The R package ordinalCont. Journal of Statistical Software, 96, 1-25.

    (R1.5) The authors state that for the success rate index, a value close to -1 indicates a large negative difference for GVS, 0 indicates no difference, and 1 indicates a large positive difference. These are theoretical bounds; the actual distribution of each index should be examined to validate such claims. However, the paper lacks descriptive statistics for this composite index.

    We removed the composite index (See R1.6) and included the actual distribution of each index for the two studies in Extended Data Fig. 3.

    (R1.6) Notably, the "reduction" of the GVS effect in altered gravity was only demonstrated in one of the five available indices (success rate, p = 0.046). In fact, the success rate in Study 2 was 66.7(sham) vs 67.3 (GVS) in Table 2. It is highly debatable whether this marginal result justifies the conclusion that GVS effects "were reduced when the games included reasoning about altered gravity.

    We thank the reviewer for this important point. We agree that the original wording placed too much weight on a single marginal comparison and could be read as implying a direct overall benefit of GVS in Study 2. We have therefore revised both the analysis and the interpretation. The conclusion no longer rests on the absolute GVS–Sham difference in Study 2 alone, which was small when averaged across all games. Instead, we now ask a more targeted question: whether the relative effect of GVS differs between terrestrial and altered-gravity contexts as a function of the games’ gravity impact. To address this, we replaced the previous gravity-weighted index with separate Gravity Indexes for each measure and each gravity-impact level, and compared these indexes across the two studies.

    This revised analysis provides a more precise pattern. For success rate, the high-gravity-impact games showed evidence in the predicted direction, with a higher GVS/Sham index in Study 2 than in Study 1, suggesting that the negative effect of vestibular disruption observed under terrestrial gravity was reduced when participants reasoned about altered gravity (lines 285-287, Fig. 5, Table 5). For strategy, the distance index also showed gravity-impact-dependent differences between studies, indicating that GVS affected spatial strategy organisation differently depending on the gravitational context (lines 307-315; Fig. 5; Table 5). We have also tempered the relevant claims in the Discussion—rather than arguing for a broad facilitation effect, we now conclude that GVS has a context-dependent influence on physical reasoning, with detrimental effects under terrestrial gravity and reduced or relatively more favourable effects when reasoning must be adapted to altered gravity. Finally, we explicitly note that the time index should be interpreted cautiously because participants were not instructed to prioritise speed, making speed–accuracy trade-offs difficult to interpret (lines 335-336, 420-436).

    (R1.7) Questionable Assumptions Regarding Strategy. The authors assume that "big changes in tool positioning and frequent tool switching indicate poor evaluation of the failed outcome". This assumption is questionable. In solving this cognitive task, participants must explore and exploit solutions based on feedback. Large shifts in positioning or frequent tool switching might reflect active, adaptive exploration based on failed outcomes rather than a failure to evaluate them.

    We thank the reviewer for this comment. We agree that large changes in tool placement or tool switching are not inherently maladaptive, because exploration is an essential part of trial-and-error learning in the Virtual Tools task. Our interpretation is more specific: when these measures are averaged across attempts within a game, persistent large shifts in placement and frequent switching provide a proxy for reduced convergence on an effective strategy, rather than for any single exploratory move being “poor”. We have revised the manuscript to make this interpretation more nuanced (lines 194-197).

    We also clarified that this interpretation is grounded in our and others’ previous work using the same Virtual Tools task, in which these strategy measures systematically tracked improved reasoning across development and reduced reasoning efficiency in ageing (Grandchamp des Raux et al., 2024; Grandchamp des Raux et al., 2025; Dexter et al., 2025; Allen et al. 2023). Thus, although exploration can be adaptive, the pattern captured by these averaged measures has empirical validity as an index of how efficiently participants refine their strategy after feedback. We have therefore revised the wording to avoid implying that all switching or repositioning reflects poor evaluation, while retaining the use of these measures as established indicators of strategy organisation in this task (lines 194-197).

    (R1.8) Confounding Factors in GVS Interpretation… GVS does not selectively target the vestibular nerve; it also activates distributed fronto-parietal attention networks and hippocampal circuits essential for any reasoning task. Additionally, the vestibular system is linked to the limbic system and the cerebellum, which regulate emotional reactivity and arousal. Because attention and emotion are likely affected by GVS, the authors should be much more cautious in attributing their behavioral findings solely to changes in the "physical experience of the body.

    We thank the reviewer for raising this important point. We agree that GVS should not be interpreted as targeting the vestibular nerve in isolation, because vestibular signals are embedded within a distributed cortical and subcortical network that also contributes to attention, spatial cognition, arousal, emotion, and executive control. We have therefore revised the manuscript to acknowledge this broader vestibular-cognitive network explicitly, citing Ferrè and Haggard (2020), and to avoid attributing the behavioural effects solely to a narrow vestibular mechanism (lines 406-407).

    At the same time, the pattern of results is not consistent with a general attentional, arousal-related, or discomfort-driven impairment. If GVS had mainly disrupted participants’ general cognitive state, we would expect a broad reduction in performance across games and across gravitational contexts. Instead, we did not observe a general GVS effect across all games, and the effects were tied to the gravity relevance of the task and to the gravitational context: in Study 1, GVS affected measures most clearly in games with medium or high gravity impact, whereas the same pattern was not observed in Study 2 under altered-gravity reasoning. This selectivity strengthens the interpretation that the effects are related to gravity-dependent physical reasoning rather than to a non-specific physiological consequence of stimulation. We have revised the Discussion to make this point more explicit while still acknowledging that GVS acts on a broader vestibular-cognitive system rather than on a single isolated pathway (lines 398-410, 416-419).

    Ferrè, E. R., & Haggard, P. (2020). Vestibular cognition: State-of-the-art and future directions. Cognitive neuropsychology, 37(7-8), 413–420. https://doi.org/10.1080/02643294.2020.1736018

    Reviewer #2 (Public review):

    (R2.1) Absence of multiple-comparisons correction. A large number of game-level pairwise t-tests are conducted in both studies (upward of twenty per study) without correction for familywise error rate. The game-level effects that anchor the main narrative - in Study 1 alone: Remove, GoalMove, Spiky, Falling_A, Shafts_B, Gap, and Chaining - arise from an uncorrected pool of comparisons. The probability that some of these constitute false positives is non-trivial. The authors should apply a correction (e.g., Benjamini-Hochberg) or at a minimum discuss this limitation explicitly.

    We thank the reviewer for their comment. As we answered to R1.1, We have revised the manuscript to clarify that each game was treated as a distinct physical-reasoning problem, and these were used to localize the games in which the stimulation effect was expressed, not to compare games with one another or to draw broad conclusions from isolated contrasts (and therefore, in this case, no multiple-comparison corrections are required; lines 637-640). That said, to address the reviewer’s concern directly and to strengthen our conclusions in the paper, we also added mixed-effect models using gravity impact level and stimulation type as fixed factors and participant and game as random effects for each study (lines 168-177, 202-207, 257-259, 301-303, 641-651, Tables 2 and 4).

    (R2.2) The facilitation claim rests on a post-hoc and arbitrarily parameterized index…These coefficients are entirely arbitrary and bear no principled relationship to the actual gravitational magnitudes used in the study. Why not use the gravity dependency ratings themselves, or the empirically estimated gravity impact scores from the computational modelling mentioned in the Methods? The choice of weights should be either principled or tested across a range of values to demonstrate robustness.

    The reviewer raises an excellent point. As described in our response to R1.4, we agree that the original gravity-weighted index was not the best way to test the cross-study effect, because assigning numerical weights of 1, 2, and 3 to low-, medium-, and high-gravity-impact games imposed an interval structure that was not independently validated. We therefore removed this weighted index entirely.

    We also considered the reviewer’s suggestion of using the gravity-dependency ratings or computational gravity-impact estimates directly. However, these values were designed to classify and balance games according to the extent to which their solution depends on gravity; they were not calibrated as an interval-scale measure of the expected GVS effect, nor are they commensurable across the different outcome measures. Using them as continuous weights would therefore reintroduce the same problem: it would treat a stimulus-classification variable as a quantitative weighting parameter. Similarly, weighting by the physical gravity magnitudes used in Study 2 would not be appropriate, because gravity magnitude (0.5g vs 2g) and gravity dependency of a game are conceptually distinct: the former defines the simulated environment, whereas the latter defines how much solving a given game relies on gravity-based reasoning.

    To address the issue more directly, we now compute a Gravity Index separately for each measure and each gravity-impact level. We then compare these indices between Study 1 and Study 2 using models that include study and gravity-impact level as factors, rather than collapsing across games with author-defined weights. This approach preserves the theoretical distinction between low-, medium-, and high-gravity-impact problems while avoiding arbitrary parameterisation. We have revised the Methods, Results, and Discussion accordingly (lines 140-147, 276-292, 307-317, 588-594, 658-666, 692-717; Table 5; Fig. 5; Extended Data Fig. 3).

    (R2.3) Furthermore, the notation in equation (1) as currently typeset reads as "Gravity minus Weighted Index" rather than "Gravity-Weighted Index"; this should be corrected.

    We followed the reviewer’s comment and adjusted the notation in equation 1 (lines 702-706).

    (R2.4) The "facilitation" interpretation exceeds what the data in Study 2 directly support. Across all games in Study 2, GVS versus Sham differences in absolute performance are non-significant in all directions. The facilitation claim derives entirely from the GWI being higher in Study 2 than in Study 1 - a between-subjects comparison involving different participant groups and a non-pre-registered metric. The language of "facilitation" should be tempered accordingly, or the authors should provide additional analyses to support this framing.

    The reviewer is correct that more support is needed. We now strengthened the facilitation argument by adding a new mixed-effect models that includes gravity impact as a factor, by showing that: (1) while in study 1, GVS has a detrimental effect in some of the performance and strategy measures, in study 2 none of the GVS vs Sham differences is significant (lines 172-177, 205-207, 257-259, 301-303); and (2) the comparison of the GI between the 2 studies pinpoint to a facilitation role limited to games where gravity matters (284-287, 308-315). At the same time, we have tempered the language in the Discussion by avoiding a broad claim of facilitation and instead stating that the evidence points to a context-dependent effect of GVS, with detrimental effects under terrestrial gravity and reduced or relatively more favourable effects under altered-gravity reasoning, with moderate evidence for the success-rate index (lines 333-336). Thus, the revised manuscript directly addresses the reviewer’s concern in two ways: the interpretation no longer depends on a single aggregate Study 2 comparison, and the facilitation language has been replaced by a more precise conclusion supported by the new gravity-impact-level analyses.

    (R2.5) Gravitational manipulation is visual only, and the vestibular system is only one component of the gravity-sensing network. Gravity perception results, as the authors very well know, from a distributed multisensory integration process that involves, in addition to the vestibular system, visual, proprioceptive, and visceral inputs. The present paradigm manipulates gravitational context solely through visual cues and targets the vestibular system through GVS - a point the authors acknowledge but do not discuss in sufficient depth. It is important to distinguish clearly between real gravitational alterations (as achieved in parabolic flight or centrifuge environments, where the entire body is physically subjected to a different gravitational vector) and virtually altered gravity, where only one sensory modality is targeted while others remain anchored to 1 g. The scope of the conclusions should reflect this distinction.

    We thank the reviewing for raising this point and clarified the distinction in the introduction and discussion (lines 104-108, 353-356).

    (R2.6) The choice of 0.5 g and 2 g may lack sensitivity. Combining the two altered-gravity conditions in Study 2, because no significant effect of hypo versus hypergravity was found, is statistically pragmatic but conceptually unsatisfying. There is evidence in the space physiology literature that gravitational processing is not linearly symmetric around 1 g: threshold effects exist below and above terrestrial gravity that may not be captured by modest deviations (half and double g) - see refs below. It is worth discussing whether the absence of a hypo/hyper distinction in Study 2 reflects a genuine equivalence or a lack of sensitivity, and whether more extreme conditions (e.g., near-zero g or 4-5 g) might reveal different processing regimes. Whether 0.5 g and 2 g were sufficient to saturate the system or merely insufficient to perturb it remains an open question with direct implications for the interpretation of the null GWI effects on strategy measures.

    We thank the reviewer for this helpful point. We agree that gravitational processing may be non-linear around 1g, and we have revised the manuscript to acknowledge that more extreme gravitational manipulations may reveal effects that are not captured by 0.5g and 2g. In the present study, however, our aim was not to characterise the psychophysical function across different levels of altered gravity, but to test whether reasoning in a non-terrestrial gravitational context changes the effect of vestibular perturbation. We therefore tested whether hypogravity and hypergravity should be analysed separately by adding gravity level (0.5g vs 2g) to the mixed-effects models. Gravity level was not a significant predictor for any performance or strategy measure, and this was consistent with previous Virtual Tools work showing no reliable difference between hypo- and hypergravity effects. On this basis, we aggregated the two conditions as “altered gravity” for the main analyses (lines 248-253). We have now also clarified that this aggregation should not be interpreted as evidence that 0.5g and 2g are cognitively equivalent in all contexts, but rather that no reliable hypo/hyper distinction was detected in the present task and parameter range (lines 248-253, 462-463, 651-656).

    (R2.7) High-level reasoning is not defined with sufficient precision. The term "high-level reasoning" appears from the title onward and in the heading of the Study 1 results section (line 138), but it is never formally defined. The reader needs a clearer account of what distinguishes high-level physical reasoning from low-level sensorimotor prediction, and where the games used here fall along that continuum.

    We thank the reviewer for pointing this out and added a clear definition of high-level reasoning at the beginning of the Introduction section (lines 41-44).

    (R2.8) What specific physical competencies - ballistic trajectories, free-fall predictions, collision dynamics, frictional forces, inertial effects - are required across the game set? When describing the subset of games that drive key effects, this information is critical for evaluating whether effects are specific to gravity reasoning or to some other physical concept.

    The reviewer is correct that the Virtual Tools games require multiple physical competencies, including gravity-based prediction, collision dynamics, object support, containment, and the anticipation of object-to-object interactions. Importantly, most games do not isolate a single physical principle; rather, successful solutions typically require combining several physical predictions. For this reason, our aim was not to attribute each effect to one isolated physical primitive, such as ballistic motion or friction, but to test whether GVS effects were stronger in games whose solutions depended more heavily on gravity-based reasoning.

    To make this clearer, we revised the manuscript to state that the game set varied systematically in the extent to which gravity-related predictions were required for solution, and that this classification was based on piloting and previous computational modelling work in the lab (lines 112-116, 426-431, 588-594; Extended Data Table 1). This design is central to the logic of the study: if GVS produced a general disruption, effects should appear broadly across the task; if it affected gravity-dependent reasoning, effects should be most evident in games with higher gravity impact. We now clarify this point and avoid implying that the task isolates gravity from all other physical competencies.

    (R2.9) Performance measures are disconnected from underlying kinematics. The performance measures (success rate, number of attempts, time per attempt) are coarse, high-level summaries.

    We agree that success rate, number of attempts, and time per attempt are high-level measures. This is, however, aligned with the logic of the Virtual Tools task. Participants do not physically manipulate objects or perform complex reaching movements; they simply select a tool with one click and place it on the screen with another. The critical demand is therefore not manual dexterity or movement execution, but the ability to predict the physical cascade that will follow from a chosen tool and placement. For this reason, kinematic measures would not provide a direct index of the reasoning process targeted here. We have clarified this in the manuscript by emphasising that the task minimises motor demands and that performance depends primarily on tool selection and spatial placement, allowing us to focus on high-level physical reasoning rather than movement kinematics (lines 116-121).

    (R2.10) Time per attempt is used as a proxy for performance efficiency, yet participants received no instructions regarding speed, and different individuals may have adopted systematically different speed-accuracy trade-offs.

    We thank the reviewer for this helpful comment. We agree that time per attempt should not be interpreted as a straightforward performance-efficiency measure. Participants were told that each game had an overall time limit, but they were not instructed explicitly to respond as quickly as possible. Attempt duration could therefore reflect several processes, including deliberation, hesitation, task engagement, or individual speed–accuracy trade-offs, rather than better or worse reasoning. We have revised the manuscript accordingly: time per attempt is now treated cautiously as a secondary descriptive measure of attempt duration, rather than as a direct index of performance efficiency. We also clarified in the Methods that participants were given a time limit but no speed instruction, and in the Discussion that the time results should be interpreted with caution for precisely this reason (lines 121-125, 431-436, 558-560).

    (R2.11) It would be valuable to know whether time per attempt correlates with attempt number within a given game (which would indicate within-game learning) and whether mouse movement data - trajectory, velocity, hesitation - were recorded and could be analysed to provide more mechanistic insight into strategy formation.

    We thank the reviewer for this suggestion and agree with the reviewer that the insight into strategy formation (and hesitation in particular) could be examined more precisely, potentially using eye tracking to capture where participants look. We add a comment in our discussion (lines 434-436). We also included the correlation in our analyses (lines 183-186, 265-267, Extended Data Fig.1 and Fig.2).

    Reviewer #2 (Recommendations for the authors):

    (R2.12) Introduction, end of paragraph 2 (applied fields): The authors mention artificial intelligence and robotics as fields that could benefit from this work. Pedagogy and educational neuroscience are equally relevant: understanding how the brain conceptualizes gravity through embodied interaction could inform how physics is taught, and could motivate more sensorimotor-grounded instructional approaches.

    This is a great point! We have now added pedagogy and educational neuroscience as relevant fields, along with corresponding references (line 52).

    (R2.13) The gravitational acceleration is typeset as "ms2," which appears to conflate milliseconds-squared with the correct unit, m·s-2 (or m/s2). Please verify the superscript convention used throughout.

    We corrected the manuscript accordingly (line 78).

    (R2.14) When introducing GVS as the tool for gravitational manipulation, the authors should more explicitly clarify that GVS produces noise in the vestibular organ rather than a physical change in the gravitational environment. Readers familiar with space physiology will notice the asymmetry between this approach and paradigms involving true gravitational alteration (parabolic flight, centrifuges). The introduction should acknowledge this asymmetry upfront rather than leaving it to the Discussion.

    We agree that this distinction should be made explicit from the outset. We have therefore revised the Introduction to clarify that GVS does not physically alter the gravitational environment, as in parabolic flight or centrifuge paradigms, but instead introduces noise into vestibular signalling while participants remain in terrestrial gravity. This asymmetry is now acknowledged before the study rationale is presented, so that the scope of the manipulation is clear to readers from the beginning (lines 102-108).

    (R2.15) “covered by the same ethical approval than study 1" - should read "as study 1.”

    We corrected the manuscript accordingly (line 512).

    (R2.16) Line 422: Was the red text label the only cue indicating the gravity condition in the altered-gravity games? If so, this is an important detail with potential implications for how participants updated their expectations. Please clarify.

    We thank the reviewer for this comment. Yes, the red text label was the explicit cue indicating the gravity condition for each altered-gravity game. Participants were told before the task that the gravity condition would vary across games and that they should attend to the red sentence at the top of the screen to identify whether the current game was in hypo- or hypergravity. We have clarified this in the Methods, including that the visual label was presented for each game and served as the instruction cue for the gravitational context (lines 560-562).

    (R2.17) Lines 547-555 - GWI notation and coefficients: (a) Replace the subscript indices 1, 2, 3 in the equations with "low," "med," and "high" to improve readability and link the notation directly to the experimental conditions.

    We changed the manuscript accordingly (lines 697-706).

    (R2.18) Lines 547-555 - GWI notation and coefficients: (b) Justify the choice of weighting coefficients 1, 2, and 3. These are arbitrary integers that do not correspond to actual gravity levels or to the empirically estimated gravity impact scores used elsewhere in the paper. Either provide a principled rationale or demonstrate that the GWI result is robust across alternative plausible weightings.

    We agree with the reviewer that the original weighting coefficients of 1, 2, and 3 were not independently validated and therefore made the gravity-weighted index difficult to interpret. Rather than trying to justify these coefficients post hoc, we addressed the concern directly by removing the gravity-weighted index from the manuscript entirely. In the revised analysis, no numerical weights are assigned to the gravity-impact categories.

    Instead, we now calculate a Gravity Index separately for each measure and each gravity-impact level, and compare these indexes between Study 1 and Study 2 using Bayesian cumulative logit models. This approach preserves gravity impact as a categorical factor, avoids imposing an arbitrary linear weighting structure, and allows the cross-study comparison to be tested directly within each gravity-impact level. We have revised the Methods accordingly and describe this new analysis in detail (lines 140-147, 276-280, 658-666, 692-717). Thus, the revised manuscript no longer depends on the arbitrary GWI coefficients.

    (R2.19) Lines 547-555 - GWI notation and coefficients: (c) The composite expression in equation (1) - as currently typeset - reads as "Gravity minus Weighted Index"; please correct the formatting so it unambiguously reads as a gravity-weighted index.

    We corrected the manuscript accordingly (lines 702-706).

    (R2.20) P-values throughout the manuscript are reported without a leading zero (e.g., "p = .39") while Cohen's d values and other statistics include one. Please adopt a consistent convention; in line with APA 7 and most journal styles, all values between −1 and 1 should include a leading zero.

    Our statistics generally follow the APA 7th guidelines (https://apastyle.apa.org/instructional-aids/numbers-statistics-guide.pdf), which state: Put a zero before the decimal point when a number is less than 1 but the statistic can exceed 1. Do not use a zero before a decimal when the statistic cannot be greater than 1 (proportion, correlation, level of statistical significance). There was a formatting error in one of the tables (Table 3), which has now been corrected.

    (R2.21) The authors should either apply a correction for multiple comparisons to game-level t-tests or acknowledge this limitation explicitly and interpret game-level results as exploratory rather than confirmatory.

    See R1.1 and R2.1. We have revised the manuscript to clarify that each game was treated as a distinct physical-reasoning problem, and these were used to localize the games in which the stimulation effect was expressed, not to compare games with one another or to draw broad conclusions from isolated contrasts (and therefore, in this case, no multiple-comparison corrections are required; lines 637-640). That said, to address the reviewer’s concern directly and to strengthen our conclusions in the paper, we also added mixed-effect models using gravity impact level and stimulation type as fixed factors and participant and game as random effects for each study (lines 168-177, 202-207, 257-259, 301-303, 641-651, Tables 2 and 4).

    (R2.22) Discussion, circa line 274 (state of the body): The authors assert that their manipulation alters the body's gravitational state. This claim deserves qualification. Gravity perception results from distributed multisensory integration; the vestibular system is one contributor, but proprioception, vision, and visceral signals are equally involved. Moreover, altering gravity visually and disrupting vestibular signals does not approximate the full-body gravitational reweighting that occurs in real altered-gravity environments. This is not a fatal limitation, but the authors should acknowledge it more directly and discuss how their results might differ from findings in truly altered gravity contexts (e.g., parabolic flight or centrifuge experiments).

    We agree that our manipulation should not be described as equivalent to the full-body gravitational reweighting that occurs in real altered-gravity environments. GVS perturbs vestibular signalling while other sensory inputs remain anchored to 1g, whereas parabolic flight or centrifuge paradigms alter the gravitational context of the body more globally. We have revised the Discussion to make this distinction explicit and to qualify the scope of our claims accordingly (lines 353-359).

    At the same time, the distinction between full-body and partial gravitational manipulation does not undermine the logic of the present study. Parabolic flight or centrifuge paradigms would not be appropriate for the physical-reasoning task used here. Moreover, the parabolic-flight study by Kunavar et al. (2021), which the reviewer helpfully highlighted, also involved a mismatch between proprioceptive and vestibular signals. Taken together, these findings suggest that manipulating one important component of the gravity-processing system can still alter behaviour, while also leaving open the possibility that stronger or different effects may emerge when sensory conflict is reduced. We have incorporated this point in the revised Discussion (lines 356-370).

    (R2.23) Discussion, around line 290 (perturbing the body helps cognition): The authors make the interesting observation that vestibular perturbation can be beneficial in altered-gravity contexts. This resonates with a finding by Kunavar et al. (2021), who showed that restoring terrestrial gravity at the hand-environment interface while the body was immersed in real altered gravity during parabolic flight paradoxically improved performance - even though the gravitational context was incoherently distributed across body segments. That study suggests a broader principle: mismatch or perturbation within the sensorimotor system does not necessarily degrade cognition and may, in some circumstances, facilitate adaptation. The present results fit within this framework, and the authors should engage with this literature. The distinction between real and virtual gravitational manipulation is not merely semantic: even genuine immersion in microgravity during parabolic flight has been shown to be insufficient to update internal gravity representations unless paired with active sensorimotor engagement (Rannaud Monany et al., J Neurophysiol, 2022), which raises the question of whether a passive vestibular perturbation such as GVS can be expected to drive the kind of cognitive recalibration the authors propose.

    We thank the reviewer for drawing our attention to the work by Kunavar et al. (2021). We agree that this study is highly relevant, and we have now incorporated it into the Discussion (lines 360-364). We also agree that the work by Rannaud Monany et al. (2022) is important for interpreting our findings, and we now discuss it explicitly in the revised manuscript (lines 365-370).

    We have clarified our interpretation of the reviewer’s point about whether passive vestibular perturbation can drive cognitive recalibration. Our study did not involve passive exposure to GVS alone. Rather, participants reasoned about physical events while vestibular signalling was being perturbed. This is conceptually aligned with the conclusion of Rannaud Monany et al. (2022), where mere exposure to microgravity was not sufficient to update internal models, but exposure coupled with task-relevant motor imagery was. Similarly, in the present study, GVS was coupled with active physical reasoning, which we argue is critical for interpreting the observed effects.

    (R2.24) Discussion, circa line 295 (internal model of gravity): A natural reference point for the predictive coding discussion is McIntyre, Zago, Berthoz, and Lacquaniti (2001, Nature Neuroscience, "Does the brain model Newton's laws?"), which provides direct neurophysiological evidence for an internalized Newtonian physics engine in the brain. This paper should be cited here.

    We agree with the reviewer that this reference is relevant and added it to the manuscript (line 373).

    (R2.25) Discussion, circa line 352 (sensory modalities and adaptation): The proposal that vestibular perturbation accelerates the revision of gravity priors is intriguing. The authors should consider whether an analogous argument applies to visual and proprioceptive channels. They should also acknowledge that sustained adaptation to a novel environment requires not only a perturbation mechanism but also veridical, graded feedback so that the system can converge on an accurate updated model - a condition that is only partially met in the present paradigm.

    See R2.22. We revised the manuscript accordingly (lines 356-370).

    (R2.26) Discussion, closing paragraph (variability and adaptability): The authors raise the positive role of noise in living systems. This connects to a well-developed body of work on optimal movement variability (Stergiou and colleagues), which formalizes how neither too little nor too much variability is optimal for adaptive performance. This framework may help situate the authors' findings within a broader theoretical account of why controlled noise injection is beneficial in both biological and artificial systems. On the computational side, the same principle underlies several foundational algorithms: stochastic gradient descent (and its mini-batch variants) outperforms deterministic gradient descent in non-convex optimization landscapes; simulated annealing uses controlled random perturbation to escape local minima; Langevin dynamics leverages noise for efficient Bayesian sampling; and dropout regularization in deep neural networks injects noise during training to improve generalization. These are canonical examples where noise is not merely tolerated but functionally necessary - a point that strengthens the authors' broader theoretical message.

    We thank the reviewers for their suggestion and revised our manuscript accordingly (lines 385-390)

    Welling, M., & Teh, Y. W. (2011). Bayesian learning via stochastic gradient Langevin dynamics. In Proceedings of the 28th international conference on machine learning (ICML-11) (pp. 681-688).

    Stergiou, N., & Decker, L. M. (2011). Human movement variability, nonlinear dynamics, and pathology: is there a connection?. Human movement science, 30(5), 869–888. https://doi.org/10.1016/j.humov.2011.06.002

    Moss, F., Ward, L. M., & Sannita, W. G. (2004). Stochastic resonance and sensory information processing: a tutorial and review of application. Clinical neurophysiology : official journal of the International Federation of Clinical Neurophysiology, 115(2), 267–281. https://doi.org/10.1016/j.clinph.2003.09.014

    Sherman, S. O., Greenstein, M., Basner, M., Clark, T. K., & Anderson, A. P. (2023). Effects of additive sensory noise on cognition. Frontiers in human neuroscience, 17, 1092154. https://doi.org/10.3389/fnhum.2023.1092154

    Reviewer #3 (Public review):

    (R3.1) The main weakness of the manuscript is that its central conclusion is not strongly supported by the data. The key finding depends on a marginally significant cross-study comparison, whereas direct GVS-versus-Sham differences in Study 2 are minimal across aggregate measures. In addition, many game-level analyses involve a large number of uncorrected multiple comparisons, raising the possibility that some of the reported effects may reflect chance findings. The manuscript's most important metric, the Gravity-Weighted Index, was not preregistered and is exploratory in nature, yet it is treated as a primary basis for confirmatory conclusions.

    See R1.4 and R2.2. We moved away from the GWI method to apply a more robust comparison between the two studies. We agree that the original gravity-weighted index was not sufficiently justified because it combined the three gravity-impact levels, using numerical weights of 1,2, and 3. Although these weights followed the ordinal classification of the games, they were not independently validated and could therefore make the cross-study comparison difficult to interpret. To address this concern directly, we removed the gravity-weighted index entirely. In the revised manuscript, we instead use a Gravity Index that is calculated separately for each measure and each gravity-impact level, as the ratio between performance or strategy outcomes under GVS and Sham. This removes the arbitrary weighting scheme and provides a direct estimate of the stimulation effect within each gravity-impact category.

    We then compared these Gravity Indexes between Study 1 and Study 2 using mixed-effect models that include study and gravity-impact level as fixed factors, rather than collapsing across gravity-impact levels with our defined weights. We used the participant as a random factor. Because the indexes are bounded and showed clustering at structural values, we analysed them using Bayesian cumulative logit models, which are appropriate for bounded, non-normally distributed outcomes and allow the cross-study contrasts to be estimated without imposing a linear weighting structure. We also clarified how the gravity-impact categories were defined, based on piloting and previous computational modelling work in the lab, and report the classification of each game in Extended Data Table 1. These revisions directly address the reviewer’s concern by replacing the arbitrary weighted composite with a transparent, category-specific analysis of the GVS effect across terrestrial and altered-gravity contexts, in line with frameworks outlined by James et al. (2021), Li et al. (2023), Liu et al. (2017) (lines 140-147, 276-292, 307-317, 588-594, 658-666, 692-717, Table 5, Fig 5, Extended data Fig 3).

    James, N. T., Harrell Jr, F. E., & Shepherd, B. E. (2021). Bayesian cumulative probability models for continuous and mixed outcomes. arXiv preprint arXiv:2102.00330.

    Li, C., Tian, Y., Zeng, D., & Shepherd, B. E. (2023). Asymptotic properties for cumulative probability models for continuous outcomes. Mathematics, 11(24), 4896.

    Liu, Q., Shepherd, B. E., Li, C., & Harrell Jr, F. E. (2017). Modeling continuous response variables using ordinal regression. Statistics in medicine, 36(27), 4316-4335.

    Manuguerra, M., Heller, G. Z., & Ma, J. (2020). Continuous ordinal regression for analysis of visual analogue scales: The R package ordinalCont. Journal of Statistical Software, 96, 1-25.

    (R3.2) The cross-study comparison is also difficult to interpret because the two studies differ in participant samples, number of games, and partially in the stimulus set.

    We thank the reviewer for this comment. We have now clarified that the cross-study comparison was based on games that were common to Study 1 and Study 2 (lines 713-714). In addition, the indices were calculated in the same way in both studies, allowing us to compare the relative effect of GVS across the two experiments rather than relying on differences in raw performance between the two participant samples. We have emphasised this point in the revised manuscript (lines 712-715).

    (R3.3) Finally, the mechanistic claims in the Discussion-particularly those invoking predictive coding, stochastic resonance, or updating of internal gravity models-go well beyond what can be directly inferred from the present behavioral data.

    We took the reviewer’s advice, along with previous comments by the other reviewers and now added further findings in the literature, strengthening our claims in the Discussion (lines 356-370, 385-390).

    Reviewer #3 (Recommendations for the authors):

    (R3.4) Moderate the scope of the main conclusions. As currently presented, the results seem to support at most the possibility that vestibular gravity signals influence a limited subset of physical reasoning tasks under specific conditions, rather than physical reasoning more broadly. The manuscript would benefit from either a more cautious interpretation or additional supporting evidence from follow-up experiments.

    We revised the manuscript accordingly (lines 398-410, 423-436).

    (R3.5) Apply an appropriate correction for multiple comparisons to all game-level analyses (for example, FDR correction) and report clearly which effects remain significant after correction…A mixed-effects modeling approach, with the game treated as a random effect within participants, would also provide a more principled and statistically powerful analysis.

    See R1.1. We have revised the manuscript to clarify that each game was treated as a distinct physical-reasoning problem, and these were used to localize the games in which the stimulation effect was expressed, not to compare games with one another or to draw broad conclusions from isolated contrasts (and therefore, in this case, no multiple-comparison corrections are required; lines 637-640). That said, to address the reviewer’s concern directly and to strengthen our conclusions in the paper, we also added mixed-effect models using gravity impact level and stimulation type as fixed factors and participant and game as random effects for each study (lines 168-177, 202-207, 257-259, 301-303, 641-651, Tables 2 and 4).

    (R3.6) Provide a clearer justification for the classification of gravity dependence (low/medium/high) and for the weighting scheme used in the Gravity-Weighted Index. It would be helpful to report the quantitative criteria or validation procedure used to assign these categories, and to include sensitivity analyses showing whether the main findings are robust to alternative weighting schemes (for example, equal or binary weights). Treating gravity dependence as a continuous predictor could also strengthen the analysis.

    See R3.1. We abandoned the weighting system and use a Baysian cumulative logit model which does not imply any arbitrary weighting instead (lines 140-147, 276-292, 307-317, 588-594, 658-666, 692-717 , Table 5, Fig 5, Extended data Fig 3).

    (R3.7) Discuss the limitations of the cross-study comparison more explicitly. The two studies differ in participant samples, numbers of games, and partially in the task set, which makes the comparison difficult to interpret. At a minimum, the manuscript would benefit from a fuller comparison of participant demographics and other relevant characteristics.

    See R3.1. We used an indexing process which normalizes the data and isolates a relative effect. While raw scores could not be used to directly compare across the two studies, the use of a standardized scale allows valid comparisons between the experiments. We revised the manuscript accordingly (line 142-144).

    (R3.8) Ideally, future work should test Earth-gravity and altered-gravity conditions within the same participants.

    We agree that testing terrestrial- and altered-gravity conditions within the same participants would be a valuable design. However, we could not implement this approach for safety reasons, as it would have substantially increased each participant’s total exposure to GVS. The current design kept GVS exposure within acceptable safety limits, and we have now clarified this point in the revised manuscript and cited the work by Utz et al. related to safety limits (lines 619-620).

    Utz, K. S., Korluss, K., Schmidt, L., Rosenthal, A., Oppenländer, K., Keller, I., & Kerkhoff, G. (2011). Minor adverse effects of galvanic vestibular stimulation in persons with stroke and healthy individuals. Brain Injury, 25(11), 1058-1069.

    (R3.9) Address the lack of a direct manipulation check for GVS more explicitly. If any subjective reports related to dizziness, discomfort, or vestibular disturbance were collected, these could be reported as an indirect validation of the manipulation. If no such data are available, this limitation should be acknowledged clearly, and future work should consider including an independent validation measure such as postural stability.

    We thank the reviewer for raising this point. In our experiment, the maximum intensity was set at 1mA, which is well tolerated. There are no side effects with these GVS parameters. The stimulation protocol used a stochastic waveform, consisting of an alternating sum-of-sines voltage with dominant frequencies at 0.16, 0.32, 0.43, and 0.61 Hz, which has been used in previous studies, both in our and other labs. We have now added this information to the manuscript by citing the work of Zanchi et al. that used the same GVS protocol (lines 613-614).

    Zanchi, S., Cuturi, L. F., Sandini, G., Gori, M., & Ferrè, E. R. (2023). Vestibular contribution to spatial encoding. European Journal of Neuroscience, 58(9), 4034-4042.

    (R3.10) Reconsider the decision to collapse the 0.5g and 2g conditions into a single altered-gravity category. Even if the two conditions do not differ significantly in the current dataset, they are cognitively distinct manipulations, and combining them may obscure potentially meaningful interactions. Separate analyses, or at least explicit interaction tests, would strengthen the interpretation.

    We agree that 0.5g and 2g are cognitively distinct manipulations and that collapsing them should be justified statistically. We therefore conducted additional mixed-effects models including gravity level (0.5g vs 2g) as a factor (see responses R1.1 and R3.5). These analyses did not reveal significant differences between the two gravity levels for the relevant performance or strategy measures (lines 248-253). On this basis, we retained the aggregation of the 0.5g and 2g conditions as a single altered-gravity category in the main analyses.

    (R3.11) Clarify the interpretation of the tool-switching results. At present, the manuscript appears to invoke different post hoc explanations for different patterns of switching behavior, including increased exploration noise and reduced cognitive flexibility. The authors should either provide a more constrained interpretation or acknowledge more clearly that the current data do not distinguish between these possibilities.

    Increased switching can reflect a less stable or less efficient strategy, but in the context of a small number of attempts, indeed it may also reflect adaptive exploration or cognitive flexibility. The present data do not allow us to distinguish conclusively between these possibilities. We have therefore revised the Discussion to provide a more constrained interpretation and to acknowledge explicitly that the meaning of the switching-rate effects remains uncertain (lines 448-452).

    (R3.12) Expand the discussion of alternative explanations. In particular, the manuscript should address whether the observed effects could reflect non-vestibular influences of GVS, such as distraction, discomfort, changes in arousal, or anxiety-related effects, rather than a specific influence on physical reasoning.

    We thank the reviewer for this excellent suggestion. We agree that non-vestibular effects of GVS, including distraction, discomfort, arousal, or anxiety-related influences, should be considered. We have therefore revised the Discussion to acknowledge this possibility more explicitly. At the same time, the pattern of results argues against a primarily non-specific explanation: if GVS had mainly affected general attention, arousal, or discomfort, we would expect a broader impairment across games. Instead, the effects were limited to specific games and were related to the gravity relevance of the task. We have clarified that this selective pattern supports an interpretation in terms of gravity-dependent physical reasoning rather than a general consequence of stimulation, while still acknowledging the broader possible influences of GVS (lines 398-411).

    (R3.13) Acknowledge more explicitly that the Virtual Tools task depends on multiple cognitive processes in addition to physical reasoning, including visuospatial ability, executive function, and trial-and-error learning. A non-physical visuospatial control task would be helpful in future work for assessing the specificity of the reported effects.

    We agree with the reviewer and revised the manuscript accordingly (lines 437-438).

  8. Version published to 10.64898/2026.03.16.712090v2 on bioRxiv
  9. Version published to 10.7554/elife.111166.1 on eLife
  10. eLife

    eLife Assessment

    This valuable study investigates whether high-level physical reasoning is grounded in real-time bodily and vestibular signals using an innovative combination of virtual tool-use tasks and galvanic vestibular stimulation. The evidence is incomplete, as the main claims rely on limited and partially exploratory effects, including uncorrected multiple comparisons and cross-study comparisons that weaken the strength of the conclusions. The work, if it can be supported by clearer statistical support and more cautious interpretation, will be of interest to researchers in embodied cognition and physical reasoning.

  11. eLife

    Reviewer #1 (Public review):

    Summary:

    This study investigates a fundamental question in cognitive science: is our ability to reason about the physical world an abstract mental process, or is it "embodied"-directly rooted in our real-time physical interactions with the environment? The authors compared participants' performance in computerized reasoning games with and without Galvanic Vestibular Stimulation (GVS). They suggest that participants failed more often and utilized suboptimal strategies under GVS compared to a sham stimulation condition. Furthermore, they found that this detrimental effect of GVS was reduced when the games were governed by altered gravity (hyper- and hypo-gravity). Consequently, the authors conclude that the physical experience of the body modifies high-level cognitive skills, such as reasoning.

    Strengths:

    The manuscript is well-written, organized, and easy to follow, making complex concepts accessible. Also, combining a specialized physical reasoning task with real-time vestibular disruption (GVS) is an intriguing approach to testing the boundaries of embodied cognition.

    Weaknesses:

    (1) Lack of Overall Effects and Inflated Type I Error for Game-Level Effects

    The study utilizes a within-subject design. Taking Study 1 as an example, each subject participated in a familiarization session (4 games), a baseline session (12 games without stimulation), a GVS session (14 games), and a sham session (14 games). No game was repeated for any single subject. Performance was quantified using three primary measures (success rate, number of attempts, and time per attempt) and two strategy measures (tool switching and the distance between tool placements).

    For Study 1, to identify condition differences at the game level (i.e., Figure 2), the authors effectively conducted 70 independent t-tests (5 measures × 14 games). While 7 significant results were reported, this large number of independent tests invites an inflated Type I error rate, as no multiple-comparison correction appears to have been applied.

    A similar inflation is expected in Study 2, where 50 independent t-tests (5 measures × 10 games) yielded 5 significant comparisons (Figure 4). Although the authors might argue the direction of the differences is systematic, implying GVS generally impairs performance, at least one significant comparison shows the opposite effect: tool switching indicates that GVS led to better performance for the 'Table_A' game in Study 2 (Figure 4d), whereas the same variable indicated GVS led to worse performance in Study 1 (Figure 2d). I suspect that none of the significant game-level results would survive a proper statistical correction. If possible, the authors can redo statistical testing with corrections (FDR or Bonferroni) or with LMM using game as a random effect. Before proper statistical analyses, I strongly encourage the authors to refrain from drawing broad conclusions based on these isolated game-level results.

    Furthermore, when analyzing data across all games, the study found no significant effect of GVS on overall performance or strategy measures in either Study 1 or Study 2. This lack of an aggregate effect contradicts the authors' conclusion that participants failed more often or utilized suboptimal strategies under GVS.

    (2) Missing Rationale for Classification Analysis

    It is puzzling why the authors pursued two exploratory analyses on tool placement after revealing that the two related primary measures (tool positioning and switching) did not generate significant condition differences in Study 1. These additional analyses-the Dirichlet Process Gaussian Mixture Model and leave-one-out classification-were not pre-registered. In the absence of overall condition differences, the authors appear to be "doubling down" by applying sophisticated classification tools to the raw data without a clear prior rationale.

    (3) Insufficient Evidence for the Reduced Effect of GVS Under Altered Gravity

    To compare Study 1 and Study 2, the authors devised a "gravity-weighted index," but its definition is not sufficiently justified. The index assigns weights of 1, 2, and 3 to low-, medium-, and high-gravity-dependent games, respectively. The choice of these specific weights appears arbitrary, making the quantitative results difficult to interpret. More importantly, there is no citation or explanation regarding how these three levels of "gravity impact" were defined in the first place (Line 468). This index was also not pre-registered.

    The authors state that for the success rate index, a value close to -1 indicates a large negative difference for GVS, 0 indicates no difference, and 1 indicates a large positive difference. These are theoretical bounds; the actual distribution of each index should be examined to validate such claims. However, the paper lacks descriptive statistics for this composite index.

    Notably, the "reduction" of the GVS effect in altered gravity was only demonstrated in one of the five available indices (success rate, p = 0.046). In fact, the success rate in Study 2 was 66.7(sham) vs 67.3 (GVS) in Table 2. It is highly debatable whether this marginal result justifies the conclusion that GVS effects "were reduced when the games included reasoning about altered gravity".

    (4) Questionable Assumptions Regarding Strategy

    The authors assume that "big changes in tool positioning and frequent tool switching indicate poor evaluation of the failed outcome". This assumption is questionable. In solving this cognitive task, participants must explore and exploit solutions based on feedback. Large shifts in positioning or frequent tool switching might reflect active, adaptive exploration based on failed outcomes rather than a failure to evaluate them.

    (5) Confounding Factors in GVS Interpretation

    The central theoretical question is whether physical reasoning is grounded in physical experience. GVS is used here to manipulate that experience. However, GVS does not selectively target the vestibular nerve; it also activates distributed fronto-parietal attention networks and hippocampal circuits essential for any reasoning task. Additionally, the vestibular system is linked to the limbic system and the cerebellum, which regulate emotional reactivity and arousal. Because attention and emotion are likely affected by GVS, the authors should be much more cautious in attributing their behavioral findings solely to changes in the "physical experience of the body."

  12. eLife

    Reviewer #2 (Public review):

    Summary

    The paper investigates whether the real-time physical experience of the body shapes high-level physical reasoning. Participants played a set of computerized tool-use reasoning games (the Virtual Tools paradigm) in which they must use knowledge of physical laws - including gravity, collisions, and inertia - to guide a ball into a target area. In Study 1, participants played the games under terrestrial gravity while receiving either Galvanic Vestibular Stimulation (GVS), which introduces noise into the vestibular organ and disrupts gravitational signalling, or a Sham condition with matched skin sensation. In Study 2, a separate cohort played the same games redesigned under hypogravity (0.5 g - half Earth g) or hypergravity (2 g - double Earth g), again with concurrent GVS or Sham stimulation. Performance was assessed through success rate, number of attempts, and time per attempt; strategy was assessed through the spatial distance between successive tool placements and the frequency of tool switching across attempts. A post-hoc gravity-weighted index (GWI) was computed to compare the effect of vestibular perturbation across the two studies. The main finding is that GVS impairs performance in gravity-dependent games under terrestrial gravity, yet the same perturbation appears to be neutral or even beneficial when the game environment involves non-terrestrial gravity - a result the authors interpret as evidence for an adaptable, body-grounded internal model of physics.

    Strengths

    One of the most notable strengths of this work is its conceptual positioning at the intersection of embodied cognition and physical reasoning. Rather than treating the human body either as an abstract information-processing device or as a purely biomechanical system, the authors take seriously the idea that cognition is scaffolded by ongoing sensorimotor state - and they test this idea with a paradigm that is both tractable and theoretically motivated. The use of the Virtual Tools paradigm is well-suited to this goal: the games vary systematically in their reliance on gravitational predictions, allowing selective impairment (rather than general disruption) to serve as a signature of embodied physical reasoning.

    The dual-study design is another strength. Testing the same vestibular perturbation under terrestrial and altered game-gravity conditions, and observing a reversal in its effect depending on context, provides a form of internal control that is conceptually compelling. The additional clustering analyses (Dirichlet Process Gaussian Mixture Model and leave-one-out kernel density classification) strengthen the strategy results beyond raw distance measures, confirming that GVS systematically shifts participants' spatial exploration strategies.

    The paper is also clearly written and engages meaningfully with relevant theoretical frameworks - predictive coding, embodied cognition, and stochastic resonance - making it accessible and stimulating for a broad audience.

    Weaknesses

    (1) Absence of multiple-comparisons correction. A large number of game-level pairwise t-tests are conducted in both studies (upward of twenty per study) without correction for familywise error rate. The game-level effects that anchor the main narrative - in Study 1 alone: Remove, GoalMove, Spiky, Falling_A, Shafts_B, Gap, and Chaining - arise from an uncorrected pool of comparisons. The probability that some of these constitute false positives is non-trivial. The authors should apply a correction (e.g., Benjamini-Hochberg) or at a minimum discuss this limitation explicitly.

    (2) The facilitation claim rests on a post-hoc and arbitrarily parameterized index. The gravity-weighted index (GWI), which drives the central cross-study comparison, uses integer coefficients (1, 2, 3) to weight games by gravity dependency level. These coefficients are entirely arbitrary and bear no principled relationship to the actual gravitational magnitudes used in the study. Why not use the gravity dependency ratings themselves, or the empirically estimated gravity impact scores from the computational modelling mentioned in the Methods? The choice of weights should be either principled or tested across a range of values to demonstrate robustness. Furthermore, the notation in equation (1) as currently typeset reads as "Gravity minus Weighted Index" rather than "Gravity-Weighted Index"; this should be corrected.

    (3) The "facilitation" interpretation exceeds what the data in Study 2 directly support. Across all games in Study 2, GVS versus Sham differences in absolute performance are non-significant in all directions. The facilitation claim derives entirely from the GWI being higher in Study 2 than in Study 1 - a between-subjects comparison involving different participant groups and a non-pre-registered metric. The language of "facilitation" should be tempered accordingly, or the authors should provide additional analyses to support this framing.

    (4) Gravitational manipulation is visual only, and the vestibular system is only one component of the gravity-sensing network. Gravity perception results, as the authors very well know, from a distributed multisensory integration process that involves, in addition to the vestibular system, visual, proprioceptive, and visceral inputs. The present paradigm manipulates gravitational context solely through visual cues and targets the vestibular system through GVS - a point the authors acknowledge but do not discuss in sufficient depth. It is important to distinguish clearly between real gravitational alterations (as achieved in parabolic flight or centrifuge environments, where the entire body is physically subjected to a different gravitational vector) and virtually altered gravity, where only one sensory modality is targeted while others remain anchored to 1 g. The scope of the conclusions should reflect this distinction.

    (5) The choice of 0.5 g and 2 g may lack sensitivity. Combining the two altered-gravity conditions in Study 2, because no significant effect of hypo versus hypergravity was found, is statistically pragmatic but conceptually unsatisfying. There is evidence in the space physiology literature that gravitational processing is not linearly symmetric around 1 g: threshold effects exist below and above terrestrial gravity that may not be captured by modest deviations (half and double g) - see refs below. It is worth discussing whether the absence of a hypo/hyper distinction in Study 2 reflects a genuine equivalence or a lack of sensitivity, and whether more extreme conditions (e.g., near-zero g or 4-5 g) might reveal different processing regimes. Whether 0.5 g and 2 g were sufficient to saturate the system or merely insufficient to perturb it remains an open question with direct implications for the interpretation of the null GWI effects on strategy measures.

    Lee SMC, Ribeiro LC, Martin DS, Zwart SR, Feiveson AH, Laurie SS, Macias BR, Crucian BE, Krieger S, Weber D, Grune T, Platts SH, Smith SM, and Stenger MB. Arterial structure and function during and after long-duration spaceflight. J Appl Physiol (1985) 129: 108-123, 2020.

    de Winkel KN, Clément G, Groen EL, and Werkhoven PJ. The perception of verticality in lunar and Martian gravity conditions. Neurosci Lett 529: 7-11, 2012.

    Clément G, Moore ST, Raphan T, and Cohen B. Perception of tilt (somatogravic illusion) in response to sustained linear acceleration during spaceflight. Exp Brain Res 138: 410-418, 2001.

    Benson AJ, Kass JR, and Vogel H. European vestibular experiments on the Spacelab-1 mission: 4. Thresholds of perception of whole-body linear oscillation. Exp Brain Res 64: 264-271, 1986.

    (6) High-level reasoning is not defined with sufficient precision. The term "high-level reasoning" appears from the title onward and in the heading of the Study 1 results section (line 138), but it is never formally defined. The reader needs a clearer account of what distinguishes high-level physical reasoning from low-level sensorimotor prediction, and where the games used here fall along that continuum. What specific physical competencies - ballistic trajectories, free-fall predictions, collision dynamics, frictional forces, inertial effects - are required across the game set? When describing the subset of games that drive key effects, this information is critical for evaluating whether effects are specific to gravity reasoning or to some other physical concept.

    (7) Performance measures are disconnected from underlying kinematics. The performance measures (success rate, number of attempts, time per attempt) are coarse, high-level summaries. Time per attempt is used as a proxy for performance efficiency, yet participants received no instructions regarding speed, and different individuals may have adopted systematically different speed-accuracy trade-offs. It would be valuable to know whether time per attempt correlates with attempt number within a given game (which would indicate within-game learning) and whether mouse movement data - trajectory, velocity, hesitation - were recorded and could be analysed to provide more mechanistic insight into strategy formation.

  13. eLife

    Reviewer #3 (Public review):

    Summary:

    This manuscript investigates a theoretically important question in cognitive science: whether higher-level physical reasoning is an abstract, modular process or is grounded in real-time body-environment interactions. To address this question, the authors combine galvanic vestibular stimulation (GVS) with the Virtual Tools task to test whether perturbing vestibular gravity signals affects performance in physical reasoning. The study is conceptually innovative and has the potential to bridge embodied sensory processing and higher-level cognition. However, in its current form, the evidence only partially supports the main claims, and several aspects of the analysis and interpretation limit the strength of the conclusions.

    Strengths:

    A major strength of the manuscript is the originality of the experimental paradigm. The combination of galvanic vestibular stimulation (GVS), which perturbs gravity-related vestibular signals, with computerized game-based tasks that require physical reasoning provides a novel way to test whether ongoing bodily experience influences higher-level cognition. Conceptually, the study is highly original and meaningfully bridges two domains that are often studied separately: sensorimotor processing and higher-level cognition.

    Weaknesses:

    The main weakness of the manuscript is that its central conclusion is not strongly supported by the data. The key finding depends on a marginally significant cross-study comparison, whereas direct GVS-versus-Sham differences in Study 2 are minimal across aggregate measures. In addition, many game-level analyses involve a large number of uncorrected multiple comparisons, raising the possibility that some of the reported effects may reflect chance findings. The manuscript's most important metric, the Gravity-Weighted Index, was not preregistered and is exploratory in nature, yet it is treated as a primary basis for confirmatory conclusions. The cross-study comparison is also difficult to interpret because the two studies differ in participant samples, number of games, and partially in the stimulus set. Finally, the mechanistic claims in the Discussion-particularly those invoking predictive coding, stochastic resonance, or updating of internal gravity models-go well beyond what can be directly inferred from the present behavioral data. Overall, the study provides intriguing but limited evidence that vestibular signals may influence some physical reasoning tasks under specific conditions, rather than strong evidence for a broad account of physical reasoning as grounded in online vestibular processing

  14. Mentioned by @biorxiv-neursci.bsky.social on Bluesky

    Real-Time Embodied Experience Shapes High-Level Reasoning Under Altered Gravity https://www.biorxiv.org/content/10.64898/2026.03.16.712090v1

  15. Mentioned by @biorxivpreprint.bsky.social on Bluesky

    Real-Time Embodied Experience Shapes High-Level Reasoning Under Altered Gravity https://www.biorxiv.org/content/10.64898/2026.03.16.712090v1

  16. Version published to 10.64898/2026.03.16.712090v1 on bioRxiv