What is free will?
For Shakespeare, free will is the dilemma of "to be or not to be," a myth in a world full of darkness and uncertainty. When facing difficulties, we have the right to think and choose, not entirely controlled by instinct or external forces. Instead, we can weigh the pros and cons of different actions internally, contemplating the meaning and value of existence.
For Nietzsche, free will is the core of his Übermensch philosophy. A person who transcends mediocrity and traditional moral constraints possesses strong willpower, capable of creating their own value and meaning, unconstrained by conventional notions of good and evil. Instead, they judge the world by their own standards, shaping a unique individual existence through self-overcoming and the pursuit of power.
Throughout human history, literati and philosophers have constructed a labyrinth of free will with words and thoughts, making us both indulge in the heroic spirit of "my destiny is mine, not heaven's" and fall into contemplation before the fatalism of "everything is predetermined." But when modern neuroscience turns its microscope on the human brain, those complex and precise neural networks begin to reveal that perhaps our understanding of free will is undergoing a revolutionary upheaval.
Thus, humanity's understanding of "free will" progresses from abstract conjecture to seeking its underlying essence—a pattern also evident in John Searle's trilogy. In "Speech Acts" (1969) and "Expression and Meaning" (1979), he proposed a highly original and influential approach to language studies: assuming that the philosophy of language is essentially a branch of the philosophy of mind—speech acts, as a specific form of human action, are merely one representation of the mind's ability to connect an organism with the world. In his third work, "Intentionality" (1983), he further explored the "aboutness" or "directedness" of mental states, building the philosophical foundation for this view [1].
And connecting free will with human behavior raises a new question: Where do intentions form in the brain? How do we become aware of these intentions?
History of Research on Voluntary Movement in Brain Science
In reality, free will itself is closely related to the core issues of human behavior research. Dualistic philosophical views on consciousness and matter can be traced back to ancient Greek philosophy, but the systematic and clear articulation of ideas like "conscious intentions form in a non-physical realm" primarily developed in modern philosophy by thinkers such as Descartes. According to dualistic philosophy, the human brain is merely a receiver, and true conscious intentions form in a non-physical realm, meaning conscious intentions exist prior as the primary cause of behavior.
In response, John Searle foresightfully proposed a different view: goal-directed behavior and its impact on the environment subjectively precede motor intention. That is, although we typically believe that motor intention precedes the action itself, a reverse timing might exist at the neural level. This misalignment between subjective experience and neural mechanism timing suggests that "consciousness might merely be a post-hoc explanation." A growing body of experimental evidence supports this view and continuously challenges the dualistic hypothesis.
Are the Will and Awareness of Movement One?
In 2009, Michel Desmurge et al. conducted an experiment involving intraoperative electrical stimulation on seven patients undergoing awake craniotomy [2].
▷ Figure 1 Stimulation response sites in the premotor and parietal cortices from a 2009 Science article.
The experiment found that stimulating the right inferior parietal lobule in patients elicited strong motor intentions and desires, causing them to involuntarily want to move the contralateral hand, arm, or foot; stimulating the left inferior parietal lobule induced lip movements and the intention to speak. When the stimulation intensity in the parietal region was increased, subjects firmly believed they had executed these actions, but no electromyographic activity was detected. In contrast, stimulating the premotor area directly triggered overt oral movements and contralateral limb movements, yet patients vehemently denied having made any movements.
Therefore, the research team proposed that conscious motor intention is actually the result of enhanced activity in the premotor-parietal neural circuit, which formulates motor plans even before action. This cortical circuit also participates in motor awareness, the cognitive process of realizing one is performing an intended action. The decoupling of the concepts of intention and awareness challenges traditional notions of free will.
How Do Neural Circuits Encode Voluntary Behavior?
In 2011, Itzhak Fried et al. recorded the activity of 1019 neurons in 12 subjects performing voluntary finger movements. These subjects had deep electrodes implanted in their brains due to the treatment needs for drug-resistant epilepsy [3].
Researchers found that neurons showed progressive recruitment approximately 1500 milliseconds before subjects reported a motor decision, and the firing frequency of neurons in the supplementary motor area (SMA) progressively increased or decreased as the subjective decision moment approached. By analyzing the activity of just 256 SMA neurons, it was possible to predict subjects' upcoming motor decisions with an average accuracy of over 80% approximately 700 milliseconds before they became aware of their decision, with the precision of predicting the actual time point of voluntary motor decisions reaching hundreds of milliseconds.
This indicates that the frontoparietal neural circuit initiates before the acting subject becomes aware of the action intention. Based on these findings, scientists proposed a computational model: voluntary will emerges when changes in the firing frequency within neuronal clusters cross a specific threshold. This reignited the debate about the nature of free will. Traditional views of free will hold that conscious intention is the "first cause" of behavior, i.e., "I decide to raise my hand, which then causes me to raise my hand"; whereas the findings above support a model of "unconscious neural activity → motor execution → post-hoc generation of intention experience," directly shaking the causality of consciousness.
How Do Humans Perceive Motor Intention?
The next question then is, what are the neural mechanisms that give rise to the subjective experience of motor intention? That is, what happens in the brain when a person "perceives" an intention? Is this experience a purely passive "neural epiphenomenon," or does it have a feedback regulatory function? Only by clarifying the real-time neural mechanisms of intention perception can we truly determine whether consciousness is an "observer" or a "participant."
Research on subjective phenomena accompanying motor intention is mainly based on two time perception paradigms—"Libet's Experiment" and "Intentional Binding Effect" [4]:
(1) Libet's Task
In the Libet paradigm, subjects were asked to freely decide to perform a simple action at any moment and to record the time point when they became aware of the intention to act, as well as the changes in preparatory potentials emitted from the brain. For example, subjects first watched a clock hand, voluntarily chose a moment to perform an action, pressed a button to stop the clock, and then reported the position of the clock hand when they first "felt the urge to act"; meanwhile, scalp electrodes recorded preparatory activity in the premotor cortex.
The landmark finding of Libet's experiment is that the slow drift of neural cluster activity (i.e., "readiness potential") not only precedes the initiation of voluntary movement but also precedes the subject's conscious perception of the intention. On average, subjects reported their conscious intention to act 206 milliseconds before the onset of muscle activity, while brain preparatory activity could appear 1 second or even earlier. This indicates that the brain has been preparing for the action for a considerable time before the subject becomes aware of the intention to act.
▷ Figure 2 Libet's task paradigm and landmark findings
(2) Intentional Binding
In the intentional binding paradigm, subjects performed a simple action (e.g., pressing a key), which then caused an environmental effect (e.g., a sound). They were then asked to report the time of the action or its environmental effect. By measuring subjects' subjective estimations of the time interval between the action and its outcome, the core finding of this research was that voluntary (not passive) actions lead to a compression of the perceived time interval between the action and its effect.
▷ Figure 3. Intentional Binding Paradigm Source: Frontiers in Human Neuroscience. DOI:10.3389/fnhum.2014.00421
Research conducted based on the above paradigms has undoubtedly significantly advanced our understanding of motor intention. However, the limitation of these studies is their failure to fully examine the neural mechanisms of the entire "intention-action-environmental effect" chain—voluntary actions in Libet's task did not include environmental effects, and while the intentional binding paradigm involved intention, it did not directly measure it.
More challenging is the current lack of cell-level recordings and subjective experience studies specifically targeting the intention process, as such research can only be conducted in humans. Furthermore, invasive single-cell recordings in humans have never been used in intentional binding research to date [5]. Therefore, the temporal relationship between subjective intention experience, action execution, and environmental effects, and their corresponding neural activities, remain unclear.
Moreover, previous studies have largely focused on preparatory activity in higher-order brain regions of the frontal motor hierarchy, neglecting the role of the primary motor cortex (M1) in the intention chain. As the final cortical hub for action execution (which Sherrington called the "final common pathway"* [6]), M1 is precisely the core target for building invasive brain-machine interfaces (BMIs) to restore motor function. However, whether and how M1 represents subjective intention, and achieves its temporal binding with other elements of the intention chain, remains an unsolved mystery.
▷ *Sir Charles Scott Sherrington (1857—1952) was a renowned British neurophysiologist who, along with Edgar Douglas Adrian, received the Nobel Prize in Physiology or Medicine in 1932 for his research on the integrative function of the nervous system and neuronal function.
The "Final Common Pathway," discovered by Sherrington while studying spinal reflexes, refers to the motor neurons in the spinal cord's anterior horn as the final convergence point for various efferent effects. These motor neurons receive impulses not only from sensory nerves but also from spinal interneurons and higher brain centers. These impulses converge on the motor neurons and are ultimately transmitted through their axons to muscles, causing muscle contraction. This concept indicates that motor neurons are the final integration point for multiple neural signals, serving as the "final highway" for motor control.
(3) M1 and NMES Combined System
After designing targeted experimental paradigms, we need suitable tools to record subjects' brain signals during tasks. The development and application of brain-machine interface (BMI) technology have brought a new perspective to motor intention research [7]. Taking neuroprosthetic technology as an example, millions worldwide suffer from paralytic diseases due to interruptions in the "brain-muscle" signal pathway. Neuroprosthetic devices can build an "electronic neural bypass," circumventing disconnected pathways in the nervous system. By decoding intracortically recorded signals and extracting motor information from them, they allow patients to control computers and robotic arms based on their imagination of movements, thereby restoring motor function in paralyzed patients. Furthermore, neuroprosthetic devices can directly connect with muscles, enabling real-time control of one's own motor system.
Neuroscientists from Ohio State University utilized long-term implanted intracortical microelectrode arrays to record multi-unit activity in the motor cortex of a subject with quadriplegia due to spinal cord injury. They decoded neural signals using machine learning algorithms and employed a customized high-precision neuromuscular electrical stimulation (NMES) system to control the activation of the subject's forearm muscles. This system achieved independent finger movement, enabling the subject to continuously control six different wrist and hand actions via the cortex, and even complete daily living functional tasks.
This unique experimental system allowed scientists to implement a novel research paradigm for the intention chain—simultaneously recording M1 neural activity and collecting subjects' subjective reports on the perceived timing of intentions, actions, and environmental effects, while systematically enabling or disabling various links in the intention chain.
▷ Figure 4 Diagram of BMI and NMES combined system Source: [7]
Breakthrough Discovery in M1 Brain Region Regarding Subjective Motor Intention
An article published in PLOS Biology in 2025 utilized the aforementioned unique system to conduct a study on the intention chain [8]. Scientists first trained a dedicated decoder to distinguish the subject's "hand opening" and "hand closing" movements. The NMES system was then used to implement these two actions—for normal subjects, these actions are achieved through electromyographic control of muscle contraction; however, in this experiment, the NMES system decoded two different neural patterns to differentiate and achieve "hand opening" (HO) and "hand closing" (HC).
In the formal experiment, the HC action was bound to an external environmental factor—the subject held a small ball, and through NMES, squeezed the ball, triggering auditory feedback 300 milliseconds after the squeeze; this was initiated by the subject's own motor impulse, with their motor intention decoded by the BMI system. This formed an intention chain composed of both external and internal elements, allowing researchers to manipulate specific links in the intention chain to simulate different states of motor intention:
No-intention state: When the subject had no motor intention, NMES was used to induce involuntary hand movements;
No-action state: After decoding the intention, NMES was still not activated;
No-effect state: After detecting the intention, NMES performed the squeezing action, but the small ball did not produce sound.
During the task, subjects observed a clock animation, and after each trial, they orally reported the position of the clock hand when a certain link in the intention chain occurred, including when the motor intention was generated, when the actual movement happened, and when the tone played. As all these times had actual timestamps, it was possible to detect discrepancies between the subject's perception of these intention chain components and their actual occurrence.
▷ Figure 5 Experimental paradigm and behavioral analysis results Source: [8]
(1) Subjective Time Perception Bias and Intention-Action Binding
Under normal conditions, subjects exhibited a proactive bias of approximately 450-500 milliseconds in their time perception of their own actions and their consequences. That is, the time they self-reported generating a motor intention was earlier than the time actually decoded by the decoder; their reported time for the auditory cue was also earlier than its actual occurrence. This discrepancy might stem from their medical condition and/or sensorimotor recalibration due to long-term BMI system training, or a delay between experience and reporting. Crucially, although the absolute time points reported by the subject were not precise, the relative time between the reported hand movement and the subsequent auditory cue was accurate—indicating their ability to observe and reliably report the timing of environmental events.
When the intention link was removed and HCs were triggered randomly instead, the perceived time of the action was significantly delayed. In the no-action state, by disabling NMES and instead triggering the sound effect with a fixed delay after the decoder reached the motor execution threshold, the perceived time of intention was significantly advanced. Notably, whether the intention or action link was selectively bypassed, the estimated time of the sound effect remained unchanged. Finally, eliminating the environmental effect from the complete intention chain did not alter the perceived time of intention or action. These results reveal a new phenomenon of intentional binding—a compression effect of subjective time between intention and action.
(2) M1 Area Neural Activity Reveals Independent Encoding of Intention and Action
Traditional research has only observed temporal binding between two physical events (action and effect), while this study is the first to describe a new form of binding between a purely internal phenomenon (intention) and physical action. This intention-action binding could not be revealed in previous studies because only in subjects with disconnected brain-effector pathways could intention and motor output be independently regulated and measured. Thus, we finally had the opportunity to observe in depth what changes occur in the subject's brain when a purely internal intention arises.
We need to find an indicator to tell us when and where the brain responds, and Multi-Unit Activity (MUA) is such a signal: MUA refers to the synchronous firing activity of local neuronal populations recorded via electrodes, reflecting the integrated electrical signals from multiple neighboring neurons. By choosing different "zero points" for alignment without averaging data from the same trial, the signal can be effectively amplified after merging.
First, aligning with the objectively detected action onset time—in unconscious conditions, "action onset time" refers to the time when NMES observed exceeding the motor threshold. In the no-action state and pure intention trials, significant evoked MUA responses appeared in the M1 area; similarly, in no-intention conditions where NMES randomly triggered actions, evoked MUA was also observed. This indicates that even in the absence of overt action, intention (rather than auditory stimulation) can still evoke average neural activity in the human M1 area. Such pure intention-evoked neural responses precede action-related activity and are expected to have smaller amplitudes than action-related activity. These results suggest the presence of separable intention-related and action-related signals in the M1 area.
Next, MUA was aligned with the subjectively perceived action time—an objective achievable only by simultaneously recording extracellular activity and asking subjects to clearly report the occurrence times of intention, action, and effect. Results showed that average MUA largely followed subjective experience: evoked responses appeared 14 milliseconds after subjective intention time and 7 milliseconds after subjective action time.
Finally, MUA was aligned separately with the "objective time" of intention (determined by the BMI decoder) and the subjective reported time—results showed that the average evoked neural activity more significantly coincided with the subjective reported time of motor intention, rather than the objective intention time marked by the BMI decoder. This suggests that what the decoder captured might be anticipatory or preparatory "intention signals"; whereas the robust evoked firing activity onset in human M1 highly co-occurred with subjective intention experience, indicating a more direct association with subjective intention experience. This implies that M1's neural activity is more inclined to reflect conscious-level intention perception, rather than early motor preparation signals.
▷ Figure 6 MUA neural population activity aligned with different action times and motor intention times Source: [8]
MUA reflects whether neuronal populations respond, but it cannot reveal the content of the neuronal activity itself. To explore whether single neuron activity can track the subjective timing of intention and action in single trials, researchers effectively isolated 66 single neurons and examined whether they encoded specific types of events. The study found that in the firing rate analysis 11 seconds before the action, 8 neurons (12%) showed a significant correlation between firing frequency and subjective intention time, indicating the presence of intention time encoding; for action time encoding, only 1 neuron showed a correlation between firing rate and subjective action time; furthermore, no neuron's firing activity was found to be related to effect perception time. This further clarifies that some single neurons in the M1 area can predict the onset of intention experience, but have limited ability to encode the timing of actions and environmental effects.
(3) Dynamic Decoding and Neural Integration of Intention Chain Links
After observing M1 neuron responses to subjective intention in neuronal activity, researchers also examined the impact of each element of the intention chain on BMI real-time neural decoding behavior from the perspective of activity representation. They found that in a complete intention chain, BMI decoding signals showed a progressive increase, gradually approaching the motor initiation threshold:
When NMES-induced non-intentional actions occurred, their decoded signals showed a delayed and steep approach to the threshold;
When intention did not trigger action, the decoded signal rose normally to the threshold but sharply dropped prematurely;
And in the absence of environmental effects, the duration of the decoded signal was prolonged.
This phenomenon, on one hand, demonstrates the decoder's sensitivity, showing its specific ability to represent motor intention and its dynamic response to all links of the intention chain (intention generation → action execution → environmental feedback); on the other hand, it further enlightens us at the neural mechanism level that M1 population dynamics not only encode motor commands but also integrate the temporal binding of intention and effect, supporting the hypothesis that "the neural basis of intention-action binding exists in the primary motor cortex."
▷ Figure 7 Neural representation changes when different links of the intention chain are interrupted Source: [8]
In summary, this study offers two insights:
First, it reveals a new type of "intention-action binding," whose temporal perceptual distortion effect is stronger than the traditional action-effect binding. This reflects that there might be a long-term stable neural association between intention formation and action execution, rather than an accidental temporary connection between action and specific auditory feedback.
Second, at the level of neural activity, the M1 cortex exhibits a dual neural encoding mechanism: at the microscopic level, single neuron firing activity is highly synchronized with the subjective onset time of intention, and the firing counts of some neuronal populations can co-vary with intention experience at the single-trial level; however, at the macroscopic level, the temporal correlation with intention onset is relatively weak. This hierarchical difference suggests that consciousness-related motor intentions might be realized by precise temporal encoding of local M1 neurons, rather than being dominated by large-scale population activity.
Future Outlook
The idea that the M1 region participates in intention encoding has been hinted at by earlier studies, which indicated that this region can detect motor intentions [7]. Considering that M1 is the final central node of the motor pathway, this finding might not be surprising. However, this study is the first to clearly define the relationship between M1 evoked activity and the subjective onset time of intention, providing an important complement to existing research.
Fried et al. [3] found that neural activity in the pre-SMA, SMA, and ACC preceded subjective intention by 700-1500 milliseconds. This article fills a critical gap by recording from the downstream M1 area—showing that M1 multi-unit evoked firing activity is synchronous with, rather than preceding, the intention experience.
This finding profoundly impacts our understanding of free will. Although the pre-SMA/SMA and M1 have direct synaptic connections, the temporal difference between them and subjective intention is as high as 700-1500 milliseconds, suggesting that the intention experience might arise from a relatively slow process of neural signal accumulation.
Recent research [9] indicates that some neurons throughout the brain slowly accumulate decision evidence in a "movement-null subspace," which is then instantaneously projected to a "movement-potent subspace" to trigger action. This might explain why evidence accumulation does not necessarily trigger action, and also indirectly proves the rationality of the significant delay between pre-SMA/SMA/ACC and M1 intention signals, further explaining why there is a temporal difference between evoked MUA in M1 (possibly belonging to the movement-potent subspace) and high-dimensional population dynamics (possibly belonging to the movement-null subspace).
However, it needs to be clarified that this does not mean M1 is the origin brain region of motor intention or a crucial brain region for intention generation. This study emphasizes that the subjective experience of motor intention appears almost synchronously with M1 evoked activity, while lagging behind neural activity in other premotor and parietal areas. Due to electrode implantation positions being limited by clinical needs, activity in key intention-related brain regions such as ACC, SMA, and premotor cortex is currently difficult to record. Future research involving multi-brain region implantation or integrating heterogeneous electrode data from multiple patients may provide a more comprehensive perspective.
Neuroscience's exploration of free will is like a blind man feeling an elephant, gradually piecing together the most mysterious and complex puzzle of the brain. A seemingly simple finding like "the temporal relationship between M1 neural activity and intention onset" requires researchers to stand on the shoulders of countless scientists, design precise experiments, and meticulously conduct complex and rigorous analyses to draw conclusions. But how is human curiosity not a most primal form of freedom? It is precisely driven by such curiosity that we can approach the truth and understand the essence of freedom.
[1] SEARLE J R. Intentionality: An Essay in the Philosophy of Mind [M]. Cambridge: Cambridge University Press, 1983.
[2] DESMURGET M, REILLY K T, RICHARD N, et al. Movement intention after parietal cortex stimulation in humans [J]. Science, 2009, 324(5928): 811-3.
[3] FRIED I, MUKAMEL R, KREIMAN G. Internally Generated Preactivation of Single Neurons in Human Medial Frontal Cortex Predicts Volition [J]. Neuron, 2011, 69(3): 548-62.
[4] HAGGARD P. Human volition: towards a neuroscience of will [J]. Nature Reviews Neuroscience, 2008, 9(12): 934-46.
[5] Noel JP, Bockbrader M, Bertoni T, et al. Neuronal responses in the human primary motor cortex coincide with the subjective onset of movement intention in brain-machine interface-mediated actions. PLoS Biol. 2025;23(4):e3003118. Published 2025 Apr 17. doi:10.1371/journal.pbio.3003118
[6] The Integrative Action of the Nervous System . Nature 76, 122 (1907). https://doi.org/10.1038/076122a0.
[7] BOUTON C E, SHAIKHOUNI A, ANNETTA N V, et al. Restoring cortical control of functional movement in a human with quadriplegia [J]. Nature, 2016, 533(7602): 247-50.
[8] Noel JP, Bockbrader M, Bertoni T, et al. Neuronal responses in the human primary motor cortex coincide with the subjective onset of movement intention in brain-machine interface-mediated actions. PLoS Biol. 2025;23(4):e3003118. Published 2025 Apr 17. doi:10.1371/journal.pbio.3003118
[9] KHILKEVICH A, LOHSE M, LOW R, et al. Brain-wide dynamics linking sensation to action during decision-making [J]. Nature, 2024, 634(8035): 890-900.