Sense of Agency

Abstract

Sense of agency, which the Medical Subject Headings classify under social cognition, is the feeling of controlling one's own actions and, through them, events in the outside world. It is what makes a movement feel done rather than merely happening, and it separates the actions we own from those we merely witness. Two broad accounts explain it: a comparator model in which the brain predicts the sensory consequences of its own commands and registers agency when prediction and feedback agree, and a reconstructive account in which agency is inferred after the fact from whatever cues best explain an outcome. This article traces the feeling and judgement of agency, its measurement through intentional binding, its neural basis, and its disturbance in disorder. Three interactive demonstrations model prediction, cue integration, and intentional binding.

Keywords: sense of agency, action awareness, intentional binding

The sense of agency is the experience of being the author of one's own actions and their effects — the tacit feeling, present in almost every waking moment, that it is I who am moving my hand, typing this word, turning this page (Haggard, 2017). It is so pervasive that it is noticed mainly when it fails: when a limb feels alien, when an action seems compelled from outside, or when a movement surprises the person who made it. This article follows the sense of agency from its conceptual definition, through the two great theories of how the brain produces it, to the implicit and explicit ways it is measured, its basis in the parietal and frontal cortex, and its disturbance in schizophrenia and related conditions.

What the Sense of Agency Is

The sense of agency is the subjective awareness of initiating, executing, and controlling one's own volitional actions and, through them, events in the external world (Moore, 2016). It is a component of self-consciousness that operates below reflective thought: one does not usually deliberate over whether a movement was one's own, the movement simply arrives stamped as self-produced. A useful contrast is with the **sense of ownership**, the feeling that it is my body that is moving. For voluntary action the two coincide, but they dissociate cleanly in passive movement — if one's arm is lifted by another person, one still owns the arm but does not feel that one caused its motion — which shows that agency is a distinct experience and not merely bodily awareness (Gallagher, 2000).

Agency is best treated as multi-layered rather than as a single feeling. At the lowest level is a thin, pre-reflective **feeling of agency** that accompanies fluent action without any explicit thought; above it sits a conceptual **judgement of agency**, an explicit attribution of an outcome to oneself or to another, which can be reasoned about and can be wrong (Synofzik et al., 2008). Because the two levels can come apart, an outcome can feel self-caused while being judged, on reflection, to have been someone else's doing, or the reverse (David et al., 2008). The sections that follow ask how each level is produced: first the predictive machinery that is thought to generate the low-level feeling, then the inferential process that shapes the high-level judgement.

Feeling and Judgement of Agency

The distinction between a feeling and a judgement of agency is the organising idea of the modern field (Synofzik et al., 2008). The feeling of agency is the immediate, non-conceptual sense that one is doing something; it is present whenever action runs smoothly and is registered implicitly, without the person forming any explicit belief. The judgement of agency is the reflective attribution — 'I did that,' or 'the machine did that' — that a person can report, defend, and revise in the light of evidence. The two normally agree, which is why they are easy to conflate, but they are driven by partly different information and can be separated experimentally.

Metacognition supplies one route to that separation. When people act toward a goal under varying levels of control, their explicit judgements of agency track how reliably their movements produce the intended effects, a monitoring of one's own control that is a form of metacognition rather than a direct readout of motor commands (Metcalfe & Greene, 2007). This layered picture matters because the two great theories of agency, taken up in the next two sections, are best read not as rivals but as accounts of different levels: the comparator model explains the low-level feeling that arises from prediction, while the reconstructive account explains the high-level judgement that arises from inference (Moore, 2016).

The Comparator Model

The dominant mechanistic account of the feeling of agency is the comparator model, which grows out of the theory of motor control (Frith et al., 2000). When the brain issues a motor command it is thought to retain a copy of that command, an **efference copy**, and to feed it to an internal **forward model** that predicts the sensory consequences the action should produce. That prediction is then compared with the actual sensory feedback that arrives once the action unfolds. When the predicted and actual signals match, the movement is registered as self-generated and feels agentic; when they diverge, the discrepancy — a **prediction error** — signals that something other than one's own command is at work (Blakemore et al., 2002). Figure 1 lays out this loop, and the first demonstration lets a reader introduce a mismatch and watch the agency signal fall.

The strongest evidence for a predictive mechanism is the attenuation of self-produced sensation. A self-generated touch feels weaker than the identical touch delivered by someone else, and the standard explanation is that the forward model cancels the predicted component of the sensation, leaving only the unpredicted remainder — which is why one cannot tickle oneself (Blakemore et al., 2002). The model is also directly manipulable: when the sensory feedback from an action is artificially delayed or spatially distorted, the felt sense of self-agency declines in proportion to the mismatch, exactly as a comparator that grades agency by prediction error would predict (Sato & Yasuda, 2005). The comparator thus offers a clean, quantitative story for the low-level feeling — but, as the next section argues, it cannot be the whole account.

Cue Integration and Apparent Mental Causation

The comparator model explains agency from the inside out, starting with the motor command. A second tradition runs the other way, arguing that much of the sense of agency is inferred after the fact. On the theory of **apparent mental causation**, people feel they willed an action when their conscious thought of it appeared just before the action, was consistent with it, and had no obvious rival cause — the same principles by which we infer causation between any two external events, turned inward (Wegner & Wheatley, 1999). On this view the felt will is not a direct perception of the mechanism that produced the movement but a reconstruction, a plausible story the mind tells about its own behaviour after the behaviour is under way (Wegner, 2003).

Neither pure prediction nor pure inference fits all the data, and the field has largely converged on a **cue integration** account that combines them: the brain weighs several agency cues — the internal predictive signal of the comparator, the prior thought or intention, the timing of the outcome, and its consistency with the goal — and settles on a sense of agency that reflects their reliability-weighted sum (Moore & Obhi, 2012). Crucially, the cues are not only retrospective. A **prospective** component operates before the outcome is known: when an action is selected fluently, with little conflict between competing options, the sense of agency over its result is boosted, and this prospective signal has been traced to the fluency of action selection rather than to any feedback (Chambon et al., 2014). Manipulating how an action is selected — for example with subliminal primes that ease or impede the choice — shifts the later sense of agency and produces measurable changes in the brain's response to the outcome, showing that agency is assembled from cues arriving both before and after the act (Sidarus et al., 2017). The second demonstration models this integration, letting a reader set a prospective and a retrospective cue and read off the resulting judgement.

Measuring Agency: Intentional Binding

Because the feeling of agency is thin and pre-reflective, asking people to rate it directly is a blunt instrument. The most influential solution is an implicit measure that never mentions agency at all. When a voluntary action causes an outcome, the perceived time of the action shifts later, toward the outcome, and the perceived time of the outcome shifts earlier, toward the action, so that the two feel drawn together in time — a compression called **intentional binding** (Haggard et al., 2002). The binding appears for voluntary actions but not for physically similar involuntary movements, which instead show the opposite, a repulsion, so the effect indexes something specific to self-caused action rather than to movement in general.

Intentional binding has become a workhorse of the field because it is implicit, quantitative, and hard to fake, and a large literature has used it to probe how agency varies with outcome valence, causal belief, and clinical state (Moore & Obhi, 2012). Its interpretation is not settled — binding may reflect a genuine agency signal, or a more general effect of causal or temporal prediction — but as a covert behavioural marker it has proved far more sensitive than explicit report. The measure connects the low-level, predictive feeling of agency to something a laboratory can put a number on, and the Worked Example below reproduces the classic action and outcome shifts that the third demonstration plots. Explicit judgements of agency remain useful alongside it, and the two often diverge, which is itself evidence that agency is built at more than one level.

The Neural Basis of Agency

Two cortical regions recur across studies of agency, and they map onto the two-level structure of the experience. When people feel that they, rather than another person or a machine, caused an outcome, activity in the **angular gyrus** of the inferior parietal cortex is characteristically low, and it rises when there is a mismatch between intended and observed action; the region behaves like a detector of discrepancy between prediction and feedback, signalling the loss of agency rather than its presence (Farrer & Frith, 2002). Complementing it, the pre-supplementary motor area and the wider medial frontal cortex are engaged when agency is felt and when actions are freely selected, consistent with a frontal contribution to the generative, prospective side of the experience (Chambon et al., 2014).

Bringing these together, the sense of agency in the human brain is best described as a distributed function in which frontal regions supply predictive and selection signals while the inferior parietal cortex evaluates the fit between prediction and outcome, their balance determining whether an action is experienced as one's own (Haggard, 2017). Direct tests of the felt experience, in which participants judge their own control while their brains are scanned, confirm that this parieto-frontal network tracks the subjective sense of control rather than the movement itself (Kuhn et al., 2013). The historical anchor for this whole enterprise is the finding that a readiness potential builds in the motor cortex before a person reports the conscious intention to move, which first made the timing of conscious will an experimental question and continues to frame debate over how far the felt agent is the initiator of action (Libet et al., 1983).

Agency in Disorder

The clearest evidence that the sense of agency is a distinct and fragile achievement comes from its breakdown. In the passivity or delusions-of-control experiences of schizophrenia, patients perform their own actions yet feel that the actions are made by an external force, as though their limbs were being moved by someone else. The comparator model gives this a natural reading: if the predictive signal that normally marks an action as self-generated is faulty, self-produced sensations are no longer attenuated and arrive tagged as externally caused, so the action is disowned even though the patient produced it (Frith et al., 2000). Consistent with this, patients prone to passivity symptoms show reduced attenuation of self-generated sensation and altered patterns of the parietal discrepancy response (Blakemore et al., 2002).

Disturbances of agency are not confined to schizophrenia. Altered experiences of control appear in functional movement disorders, in obsessive-compulsive disorder, and in the phenomenology of hypnosis and of anarchic-hand syndrome, and the study of these conditions alongside the sense of body ownership has become a field in its own right (Braun et al., 2018). Because a weakened or exaggerated sense of agency maps onto specific symptoms, agency measures such as intentional binding are increasingly used as objective markers of clinical state rather than as purely theoretical constructs.

Current Directions

Two currents dominate recent work. The first is the maturation of the cue-integration framework into a quantitative, computational account in which each agency cue carries a weight set by its reliability, so that the sense of agency becomes a case of the same Bayesian cue-combination the brain uses elsewhere in perception (Wen & Imamizu, 2022). This reframing makes agency continuous with the rest of perceptual inference and yields testable predictions about how prospective and retrospective cues should trade off as their reliabilities change.

The second current is applied. As people increasingly act through machines — cursors, autopilots, prosthetic limbs, and semi-autonomous systems that intervene between an intention and its effect — the question of who or what is in control has moved from the laboratory to interface design, and the sense of agency has become a variable that human-machine systems must be built to preserve rather than accidentally erode (Wen & Imamizu, 2022). The same measures developed to study schizophrenia are now used to ask whether an automated system leaves its human operator feeling, correctly, in control (Braun et al., 2018).

Criticisms and Open Questions

The comparator model, for all its explanatory reach, faces a standing objection: prediction error alone cannot generate the sense of agency, because many self-caused outcomes are poorly predicted and many well-predicted events are not self-caused. A purely predictive mechanism would misattribute agency whenever the world happened to match an expectation, which is why the field moved to a multifactorial account in which the comparator signal is one weighted cue among several rather than the sole source of the feeling (Synofzik et al., 2008). Where exactly the boundary lies between a genuine, prediction-based feeling of agency and a purely inferential judgement remains contested (David et al., 2008).

The reconstructive tradition has its own difficulty: if the felt will were merely an after-the-fact story, it is unclear why it should be as accurate and as tightly time-locked to action as it usually is, and demonstrations that thought can be made to feel like cause do not show that it always is (Wegner, 2003). The interpretation of intentional binding is likewise open, since temporal binding can be produced by causal or statistical structure that has nothing to do with agency, so a bound interval is suggestive but not decisive evidence of felt control (Moore & Obhi, 2012). Underlying all of these is the hardest question — how a set of computational signals becomes a felt experience of authorship at all — which the current science brackets rather than answers, mapping the conditions under which agency is felt while leaving the nature of the feeling itself for later.

Worked Example

The second demonstration models the sense of agency as a reliability-weighted sum of two cues, following the cue-integration account in which a prospective cue available before the outcome and a retrospective cue available after it are combined into a single judgement. Let the prospective cue be the fluency of action selection, scored p between 0 for a heavily conflicted choice and 1 for a wholly fluent one, and let the retrospective cue be the consistency of the outcome with the goal, scored r between 0 and 1. Give the retrospective cue the greater weight, taking the agency judgement to equal one hundred times the quantity zero point four times p plus zero point six times r. When selection is fluent and the outcome matches the intention, p is 1 and r is 1, so the judgement is one hundred times zero point four plus zero point six, which is one hundred times one, or 100. When the same matching outcome follows a conflicted selection, p is 0 and r is 1, so the judgement is one hundred times zero point six, which is 60: the outcome still feels largely one's own, but less so, because the prospective cue has been withdrawn. When a fluent selection is followed by an outcome inconsistent with the goal, p is 1 and r is 0, so the judgement is one hundred times zero point four, which is 40. A withdrawn prospective cue therefore costs forty points and a violated retrospective cue costs sixty, and these are exactly the values the demonstration reads out as its two sliders move.

The third demonstration reproduces the classic intentional-binding shifts of Haggard, Clark and Kalogeras (2002). Take the true interval between a voluntary action and the tone it triggers to be 250 milliseconds. In the voluntary condition the perceived time of the action moves 15 milliseconds later, toward the tone, and the perceived time of the tone moves 46 milliseconds earlier, toward the action, so the perceived interval is 250 minus 15 minus 46, which is 189 milliseconds — a binding, a compression, of 15 plus 46, or 61 milliseconds. In the involuntary condition, where the same movement is evoked artificially rather than willed, the shifts reverse: the action is perceived 27 milliseconds earlier and the tone 14 milliseconds later, so the perceived interval is 250 plus 27 plus 14, which is 291 milliseconds — a repulsion of 27 plus 14, or 41 milliseconds. The same physical sequence is thus compressed by 61 milliseconds when it is willed and stretched by 41 milliseconds when it is not, a swing of 102 milliseconds that the demonstration draws as the difference between the two conditions, and that is the behavioural signature the measure exploits.

Discussion

The sense of agency is best understood not as a single feeling with a single cause but as a layered construct assembled from several sources of evidence. At its base is a predictive mechanism: the brain forecasts the sensory consequences of its own commands through a forward model and registers an action as self-caused when prediction and feedback agree, a process visible in the attenuation of self-produced sensation and in the graded loss of agency when feedback is distorted. Layered over this is an inferential process that reconstructs authorship from cues — prior thought, the fluency of selection, the timing and consistency of outcomes — and weights each by its reliability, so that agency behaves like the rest of perceptual inference. Intentional binding gives the whole system an implicit behavioural measure, and a parieto-frontal network, with the angular gyrus detecting mismatch and medial frontal cortex supplying prospective signals, gives it a neural home. Table 1 sets the main theoretical accounts side by side.

Read this way, the study of agency has moved from a contest between prediction and inference to a single framework in which both are cues to be weighted, and from a purely theoretical construct to a measurable quantity with clinical and engineering uses. What remains open is not the catalogue of cues but their synthesis: how the weighted signals become a felt experience of authorship, why that experience is normally so accurate, and how it fails so specifically in disorder. Those questions place the sense of agency at the meeting point of motor control, perception, and the science of consciousness, a construct whose functional anatomy is increasingly clear while its subjective character remains among the open problems of the field.

Glossary

Apparent mental causation.

Wegner's theory that the feeling of consciously willing an action is inferred, from the priority, consistency, and exclusivity of a preceding thought, rather than perceived directly from the action's cause.

Comparator model.

The account on which the feeling of agency arises when the sensory consequences predicted by a forward model match the actual feedback from an action.

Cue integration.

The view that the sense of agency is a reliability-weighted combination of several cues — the predictive comparator signal, prior thought, and the timing and consistency of the outcome.

Efference copy.

A copy of an issued motor command, retained by the brain and fed to a forward model so that the sensory consequences of the action can be predicted.

Feeling of agency.

The thin, pre-reflective sense that one is causing an action as it happens, registered implicitly and without any explicit belief.

Forward model.

An internal model that, given a copy of a motor command, predicts the sensory consequences the action should produce, allowing self-caused input to be anticipated and attenuated.

Intentional binding.

The perceived compression of the interval between a voluntary action and its outcome — the action felt later and the outcome earlier — used as an implicit measure of the sense of agency.

Judgement of agency.

The explicit, conceptual attribution of an outcome to oneself or to another, which can be reported, reasoned about, and mistaken.

Passivity phenomena.

Experiences, characteristic of schizophrenia, in which patients feel that their own actions, thoughts, or sensations are produced or controlled by an external agent; a disorder of the sense of agency.

Prediction error.

The discrepancy between the sensory feedback predicted by a forward model and the feedback actually received; a large error signals that an outcome was not self-caused.

Prospective agency.

The component of the sense of agency that is generated before an outcome is known, from the fluency with which an action is selected among alternatives.

Readiness potential.

A slow build-up of electrical activity in the motor cortex that precedes a voluntary movement, and that Libet found to begin before the reported moment of conscious intention.

Retrospective agency.

The component of the sense of agency inferred after an outcome, from its timing and its consistency with the goal that was held.

Sense of agency.

The subjective awareness of initiating, executing, and controlling one's own actions and, through them, events in the external world.

Sense of ownership.

The feeling that it is one's own body that is moving or being touched, distinct from the sense of agency and preserved even during passive movement.

Sensory attenuation.

The reduced intensity of a self-generated sensation relative to an identical externally generated one, attributed to cancellation of the predicted component and illustrated by the difficulty of self-tickling.

Two-step model.

Synofzik's multifactorial account distinguishing a non-conceptual feeling of agency from a conceptual judgement of agency, built from several cues rather than from the comparator alone.

Key Researchers

Sarah-Jayne Blakemore. Professor of Psychology and Cognitive Neuroscience at the University of Cambridge; developed the central-monitoring account of action awareness and showed that the attenuation of self-generated sensation explains why one cannot tickle oneself. ORCID - Faculty Page - Google Scholar - Wikipedia

Valerian Chambon. CNRS researcher at the Institut Jean Nicod, ENS Paris; identified the prospective, action-selection contribution to the sense of agency and its basis in the angular gyrus. ORCID - Faculty Page - Google Scholar

Christopher D. Frith. Emeritus Professor at the Wellcome Centre for Human Neuroimaging, UCL; applied the forward-model comparator to the awareness and control of action and to the passivity symptoms of schizophrenia. ORCID - Faculty Page - Google Scholar - Wikipedia

Shaun Gallagher. Moss Chair of Excellence in Philosophy at the University of Memphis; drew the influential distinction between the sense of agency and the sense of ownership that frames the field. ORCID - Faculty Page - Google Scholar - Wikipedia

Patrick Haggard. Professor of Cognitive Neuroscience at UCL; introduced intentional binding as an implicit measure of agency and mapped the parieto-frontal network that supports it. ORCID - Faculty Page - Google Scholar - Wikipedia

Benjamin Libet (1916-2007). Physiologist at the University of California, San Francisco; his readiness-potential experiments on the timing of conscious intention made the conscious control of voluntary action an experimental question. Wikipedia

Janet Metcalfe. Professor of Psychology at Columbia University; studied the metacognition of agency, showing that explicit judgements of control track how reliably actions produce their intended effects. ORCID - Faculty Page - Google Scholar

James W. Moore. Reader in Psychology at Goldsmiths, University of London; reviewed intentional binding and advanced the cue-integration approach that combines predictive and inferential accounts of agency. Faculty Page - Google Scholar

Matthis Synofzik. Professor at the Hertie Institute for Clinical Brain Research, University of Tubingen; proposed the multifactorial two-step account distinguishing the feeling from the judgement of agency. ORCID - Faculty Page - Google Scholar

Daniel M. Wegner (1948-2013). Professor of Psychology at Harvard University; formulated the theory of apparent mental causation, on which the felt will is a reconstructive inference rather than a direct readout. Google Scholar - Wikipedia

Daniel M. Wolpert. Professor of Neuroscience at Columbia University; developed the forward-model theory of sensorimotor prediction that underlies the comparator account of agency. ORCID - Faculty Page - Google Scholar - Wikipedia

Frequently Asked Questions

What is the sense of agency?
It is the feeling of controlling one's own actions and, through them, events in the world; it is the experience of being the author of what one does, present in ordinary action and noticed mainly when it fails (Moore, 2016).

How is the sense of agency different from the sense of ownership?
Ownership is the feeling that it is one's own body that is moving; agency is the feeling that one is causing the movement. In passive movement one still owns the limb but does not feel one caused it, which shows the two are distinct (Gallagher, 2000).

What is the comparator model of agency?
It holds that the brain predicts the sensory consequences of its own motor commands and registers an action as self-caused when the prediction matches the actual feedback; a mismatch signals that something external is at work (Frith et al., 2000).

Why is it impossible to tickle oneself?
Because a forward model predicts the sensation one's own movement will produce and cancels the predicted part, so a self-generated touch feels weaker than the same touch from another person, a direct sign of the predictive mechanism behind agency (Blakemore et al., 2002).

What is intentional binding?
It is the perceived drawing-together in time of a voluntary action and its outcome, with the action felt later and the outcome earlier, which occurs for willed but not involuntary movements and serves as an implicit measure of the sense of agency (Haggard et al., 2002).

Is the sense of agency just prediction, or also inference?
Both. A predictive comparator generates the low-level feeling, while an inferential process reconstructs authorship from cues such as prior thought and outcome consistency; current accounts combine them as reliability-weighted cues (Moore & Obhi, 2012).

Which brain regions support the sense of agency?
A parieto-frontal network: the angular gyrus signals mismatch between intended and observed action and its loss of agency, while medial frontal regions supply the prospective, selection-related side of the experience (Haggard, 2017).

How is agency disturbed in schizophrenia?
In passivity experiences, patients perform their own actions yet feel controlled by an external force; a faulty predictive signal leaves self-produced sensations unattenuated and tagged as externally caused, so the action is disowned (Frith et al., 2000).

References