Domain-general computational integration in the Sense of Agency
Curation statements for this article:-
Curated by eLife
eLife Assessment
This important study examines how temporal and spatial discrepancies between an action and its visual feedback are combined in judgments related to the sense of agency. Using a virtual-reality reaching task and computational modeling, the authors show that both discrepancies influence self-attribution and that a multiplicative model outperforms the prespecified alternatives, a result reproduced in a preregistered replication. The evidence is convincing for the conclusion that temporal and spatial information is integrated. Other models - beyond multiplication - may in principle account equally well for the observed data; sharpening these claims and situating the findings more clearly within Bayesian accounts of cue integration would further strengthen this key contribution.
This article has been Reviewed by the following groups
Listed in
- Evaluated articles (eLife)
Abstract
The Sense of Agency (SoA), the experience of being in control of one’s own actions, is thought to emerge from the comparison of internal sensorimotor predictions with afferent feedback. Classical comparator models treat SoA as arising from a prediction-feedback comparison. Volitional action likely engages multiple forward models that predict distinct features of an outcome, such as its timing and spatial trajectory. Whether prediction errors arising from these distinct forward models are integrated into a domain-general representation of agency, and if so by what computational logic, remains unresolved. We addressed this question using a Virtual Reality reaching task. Participants observed a virtual hand replicating their movements while we independently manipulated two sensorimotor domains: temporal delay and spatial angle deviation, in isolation and in factorial combinations. After each trial, participants made an SoA judgment. We examined whether SoA responses show computational hallmarks of integration between different features of sensorimotor prediction. Specifically, we pre-registered three computational models (Multiplicative, Minimum, and Mean) and compared their fit to per-trial responses. Across an exploratory sample (N = 16) and a pre-registered replication (N = 38), SoA declined monotonically with conflict magnitude in both domains. Critically, a Multiplicative integration rule consistently outperformed the Minimum and Mean rules. These results provide direct evidence for domain-general integration between prediction errors in SoA, governed by a multiplicative computational logic.
Article activity feed
-
eLife Assessment
This important study examines how temporal and spatial discrepancies between an action and its visual feedback are combined in judgments related to the sense of agency. Using a virtual-reality reaching task and computational modeling, the authors show that both discrepancies influence self-attribution and that a multiplicative model outperforms the prespecified alternatives, a result reproduced in a preregistered replication. The evidence is convincing for the conclusion that temporal and spatial information is integrated. Other models - beyond multiplication - may in principle account equally well for the observed data; sharpening these claims and situating the findings more clearly within Bayesian accounts of cue integration would further strengthen this key contribution.
-
Reviewer #1 (Public review):
Summary:
This study investigates how temporal and spatial discrepancies between an action and its visual feedback are combined when people judge whether an observed movement corresponds to their own action. Using a virtual-reality reaching task and computational modeling, the authors show that both temporal and spatial discrepancies reduce self-attribution and that a Multiplicative model better explains judgments than the prespecified Minimum and Mean alternatives. Importantly, the main model comparison was confirmed in a preregistered replication.
Strengths:
The study addresses an important question about how multiple sources of sensorimotor information contribute to the sense of agency. The factorial manipulation of temporal and spatial discrepancies provides a clear experimental approach to this question, …
Reviewer #1 (Public review):
Summary:
This study investigates how temporal and spatial discrepancies between an action and its visual feedback are combined when people judge whether an observed movement corresponds to their own action. Using a virtual-reality reaching task and computational modeling, the authors show that both temporal and spatial discrepancies reduce self-attribution and that a Multiplicative model better explains judgments than the prespecified Minimum and Mean alternatives. Importantly, the main model comparison was confirmed in a preregistered replication.
Strengths:
The study addresses an important question about how multiple sources of sensorimotor information contribute to the sense of agency. The factorial manipulation of temporal and spatial discrepancies provides a clear experimental approach to this question, and the distinction between an exploratory study and a preregistered replication is a major strength. The behavioral effects were consistent across the two samples, and the replication provides confirmatory support for the advantage of the Multiplicative model among the prespecified candidate models.
Weaknesses:
The main limitation concerns the strength of the theoretical conclusions drawn from the model comparison. Although the Multiplicative model outperformed the prespecified Minimum and Mean models, this does not uniquely establish multiplication as the underlying computational operation, because other nonlinear integration rules were not tested. The evidence therefore more directly supports cross-domain integration than a specifically multiplicative computation.
Similarly, the behavioral results do not by themselves establish that this integration occurs within a domain-general representation of agency. The binary judgment used in the task is also closely related to judging whether the seen and performed movements match, leaving open whether the observed integration reflects agency computation itself or a later decision process. These considerations do not undermine the main empirical findings, but they suggest that the broader computational and representational claims should be stated more cautiously.
-
Reviewer #2 (Public review):
Summary:
Harduf et al. investigated the mechanisms underlying the sense of agency (SoA), the feeling of being in control of our actions. SoA is thought to arise from the comparison of action outcome predictions, emerging from forward models, with the actually observed outcomes. Current literature has not elucidated how cues from different forward models predicting different features of actions are combined to form a final agency judgement. The authors investigate this gap by introducing various combinations of spatial and temporal mismatches during a virtual reality (VR) based reaching task, and recording participants' agency judgements. Then, they fit different models of cue combination to behavioural data and compare them. The authors find compelling evidence that a model in which the effects of spatial …
Reviewer #2 (Public review):
Summary:
Harduf et al. investigated the mechanisms underlying the sense of agency (SoA), the feeling of being in control of our actions. SoA is thought to arise from the comparison of action outcome predictions, emerging from forward models, with the actually observed outcomes. Current literature has not elucidated how cues from different forward models predicting different features of actions are combined to form a final agency judgement. The authors investigate this gap by introducing various combinations of spatial and temporal mismatches during a virtual reality (VR) based reaching task, and recording participants' agency judgements. Then, they fit different models of cue combination to behavioural data and compare them. The authors find compelling evidence that a model in which the effects of spatial and temporal incongruence are multiplied provides the best fit, outperforming models that do not combine spatial and temporal incongruences, and one model that combines them by averaging rather than multiplying. The authors argue that sense of agency emerges from the domain-general integration of different forward models.
Strengths:
The experiments are well designed, and model validation is compelling thanks to the comparison with an appropriate set of alternative models and replication of the results in a first exploratory cohort and in a pre-registered validation cohort. The writing is generally very clear, and the presentation of the results is exhaustive yet compact, making the paper very readable. I also commend the authors for empirically addressing whether and how the integration of different cues contribute to the sense of agency, an idea which has been more widely proposed conceptually than empirically investigated.
Weaknesses:
(1) The terminology is not always clearly defined, and the authors are not sufficiently rigorous in specifying what gap their study aims to address exactly, and to what extent it succeeds. The authors acknowledge that the idea that cue combination underlies sense of agency was already proposed, yet frame their contribution as showing that "integration" underlies sense of agency. However, it is unclear what the authors exactly mean by "integration". Also, it is not well specified what a domain-general representation is exactly, and how such a notion is not already included in the previously proposed cue combination theory. Overall, the authors seem to imply the main contribution of the paper is to elucidate both whether and how cue integration is performed. However, the most novel contribution of the study appears to concern the "how" rather than the "whether".
(2) The authors should have discussed their findings in light of Bayesian inference, which has already been proposed as the basis of cue combination for sense of agency. The fact that a multiplication of the different cues fits the data better than an averaging makes it too tempting to think about Bayesian inference to ignore such idea. Indeed, if cues are assumed to provide statistically independent evidence about the probability of self-causation, then they should combine in a multiplicative way rather than by averaging. Even if the experimental design or pre-registration constraints preclude a formal Bayesian analysis, this point should be discussed and related to previous literature.
-
Author response:
We thank both reviewers for their constructive and thoughtful evaluations, and the editors for the positive assessment. We are grateful that the reviewers found the question important, the design clear, and the exploratory-replication structure a strength. Below we briefly outline the directions of our planned revision; a full point-by-point response to the public reviews and the recommendations will accompany our resubmission.
The reviewers raise several points we find valuable and intend to address. First, we will clarify the scope of our central claim. We will distinguish our confirmatory result, that the Multiplicative model best accounts for the data among the preregistered candidates, from stronger statements about the underlying computation, making explicit that the evidence most directly supports cross-domain …
Author response:
We thank both reviewers for their constructive and thoughtful evaluations, and the editors for the positive assessment. We are grateful that the reviewers found the question important, the design clear, and the exploratory-replication structure a strength. Below we briefly outline the directions of our planned revision; a full point-by-point response to the public reviews and the recommendations will accompany our resubmission.
The reviewers raise several points we find valuable and intend to address. First, we will clarify the scope of our central claim. We will distinguish our confirmatory result, that the Multiplicative model best accounts for the data among the preregistered candidates, from stronger statements about the underlying computation, making explicit that the evidence most directly supports cross-domain integration, with multiplication the best-supported rule among those tested. Second, we agree on the importance of situating our findings within Bayesian accounts of cue integration, and will add a discussion of how the multiplicative combination of statistically independent cues relates to probabilistic inference about self-causation and to prior cue-integration work on the sense of agency. Third, we will sharpen definitions of key terms such as ‘integration’ and ‘domain-general’ and the framing of our contribution, clarifying that it concerns how multi-feature prediction errors are combined. Finally, we will address and discuss the distinction between the integration we demonstrate and its interpretation as a domain-general representation of agency.
We believe these revisions will strengthen the manuscript, and we thank the reviewers again for their helpful guidance.
-
-
-
Domain-general computational integration in the Sense of Agency https://www.biorxiv.org/content/10.64898/2026.08.13.744643v1
-
-
Domain-general computational integration in the Sense of Agency https://www.biorxiv.org/content/10.64898/2026.08.13.744643v1
-
Domain-general computational integration in the Sense of Agency https://www.biorxiv.org/content/10.64898/2026.08.13.744643v1