Precision-weighting and the felt sense of agency: a predictive-coding account

Abstract

We propose that the subjective sense of agency — the feeling that one's actions cause events in the world — arises from a specific pattern of precision-weighted prediction errors in the motor and somatosensory hierarchy. Using a novel paradigm combining voluntary action, delayed sensory feedback, and layer-resolved fMRI, we show that individual differences in agency attribution track the relative weight given to proprioceptive versus efference-copy signals. A computational model formalizing this account makes precise predictions about how agency breaks down in conditions of sensory attenuation, confirmed in a pre-registered follow-up study.

Introduction

The sense of agency — the phenomenal quality of being the author of one's actions and their effects — is both central to self-experience and surprisingly fragile. It can be diminished by sensory delays, amplified by predictive information, and disrupted in psychopathology (Haggard, 2017; Wegner, 2002). Understanding the computational and neural mechanisms that generate this experience has direct clinical relevance for conditions ranging from schizophrenia-spectrum disorders to functional neurological disorder.

Predictive coding frameworks offer a natural starting point (Friston, 2010; Clark, 2016). In these accounts, the nervous system generates top-down predictions about the sensory consequences of action and computes prediction errors — discrepancies between expected and observed signals — used to update the generative model and, when sufficiently large, phenomenal experience. Agency might therefore arise when prediction errors in motor-sensory pathways are small: the world unfolds as the motor system anticipated, and this match registers as authorship. What has been lacking is a principled account of precision weighting in the agency context. We propose that the felt sense of agency depends not merely on prediction error magnitude but on whether the system assigns higher precision to efference-copy signals or to incoming proprioceptive and exteroceptive evidence.

Method

Participants

Forty-two right-handed adults (22 women; M age = 27.3 years, SD = 4.8) with no history of neurological or psychiatric condition participated in the fMRI component. An additional 38 participants completed the pre-registered behavioural follow-up. All participants provided written informed consent; the study was approved by the University of Ottawa Research Ethics Board (Protocol H-04-22-7814).

Paradigm and Imaging

Participants performed a delayed-action paradigm in which they pressed a key and observed a visual flash after a variable delay (0, 100, 200, or 400 ms), rating their sense of agency on a 7-point scale after each trial. A sham condition — the flash triggered externally without any action — allowed estimation of baseline agentive attribution. Functional MRI was acquired at 3T using a multiband sequence (TR = 800 ms, 2 mm isotropic). Layer-resolved BOLD responses were estimated using a surface-based deconvolution approach targeting superficial, middle, and deep laminae across primary motor cortex (M1), supplementary motor area (SMA), primary somatosensory cortex (S1), and posterior parietal cortex (PPC).

We implemented a hierarchical predictive coding model in which efference-copy signals project as descending predictions to somatosensory and visual regions, and sensory afferents generate ascending prediction errors. Precision parameters were estimated per participant using variational Bayes fitting. Agency ratings were modelled as a sigmoid function of the precision-weighted prediction error balance.

Results

Agency ratings declined monotonically with delay duration (F(3, 123) = 87.4, p < .001, η²p = 0.68). Individual variability in this decline was well predicted by the model-estimated precision balance (r = 0.61, 95% CI [0.43, 0.75]). Layer-resolved fMRI revealed that agency attribution correlated most strongly with BOLD signal in the deep layers of SMA — associated with efference-copy generation — and the superficial layers of S1, associated with incoming somatosensory prediction errors. Participants with higher efference-copy precision estimates showed proportionally greater deep-SMA relative to superficial-S1 activation during high-agency trials, a double dissociation consistent with the proposed mechanism.

The pre-registered follow-up (N = 38) tested the model's prediction that proprioceptive attenuation via vibrotactile masking would shift the precision balance and elevate agency attribution at longer delays. This prediction was confirmed (delay × masking interaction: F(3, 111) = 14.2, p < .001, d = 0.53).

Discussion

These results provide the first layer-resolved fMRI evidence for a precision-weighting account of agentive experience. The double dissociation between deep SMA and superficial S1 laminar profiles — across individuals differing in their agency-delay functions — offers a mechanistic foothold for understanding how the felt sense of authorship arises from motor-sensory hierarchies. The model's success in the pre-registered follow-up provides confirmation that goes beyond post-hoc fitting.

The account makes additional predictions about agency disruption under conditions of sensory attenuation relevant to schizophrenia-spectrum psychopathology. We are currently testing these in a clinical sample, and the computational model is available in the Institute's open model library for reuse in other paradigms.

References

Clark, A. (2016). Surfing uncertainty: Prediction, action, and the embodied mind. Oxford University Press.
Friston, K. (2010). The free-energy principle: A unified brain theory? Nature Reviews Neuroscience, 11(2), 127–138.
Haggard, P. (2017). Sense of agency in the human brain. Nature Reviews Neuroscience, 18(4), 196–207.
Seth, A. K., & Bayne, T. (2022). Theories of consciousness. Nature Reviews Neuroscience, 23(7), 439–452.
Shergill, S. S., Bays, P. M., Frith, C. D., & Wolpert, D. M. (2003). Two eyes for an eye: The neuroscience of force escalation. Science, 301(5630), 187.
Wegner, D. M. (2002). The illusion of conscious will. MIT Press.
Wolpert, D. M., Ghahramani, Z., & Jordan, M. I. (1995). An internal model for sensorimotor integration. Science, 269(5232), 1880–1882.