Redefining the experience of will
To discuss the experience of will, we first must define what phenomenon the phrase refers to. There are two key parts: the experience and the will. Everyone has their own conception of what constitutes will, or at least has some intuitive sense of having it. However, the concept of will is rather broad and is often understood in relation to what we call "free will", certainly not a topic you would expect to find much consensus on, or the subject of this inquiry. To keep the topic focused, it will be helpful to try and abandon the term "will" and redefine the experience of will into concepts better understood and defined.
First, I would like to propose putting will in terms of volition. In this context, volition is defined as a set of cognitive processes that have a few distinct features. First, the function of volition is to "decide" on a movement; it involves an intention to make a movement. I will be focusing on movement in particular, though volition can also be used in relation to changes in more abstract terms, such as plans for later actions. The products of volition, given this movement-based definition, are voluntary actions. Second, volition is internally generated. That is, it is a process that can take place completely independent of any external stimulus, though external stimuli can certainly still play a role. Volition is also goal oriented and reasoned; we can ask why questions about voluntary actions and expect some kind of justification. A prerequisite to this feature is self-awareness; a subject must be aware of their own volition in order to reason about it. Hence it must be experiential in some way; we must have a conscious experience of volition in order to be aware and reason about it.
Perhaps the most important of these features of volition is that it involves a goal, as this is what will provide the crucial link between the cognitive process of volition and the experience of volition. Volition has at its core a neural representation of some goal. Yet introspecting that goal, identifying what is being willed is often difficult and unclear. The content of volition, the "intentional object" is certainly not always self-apparent, and is often masked by the emotional states attached to the experience of volition [Metzinger 2006]. Why did I sit down at the particular time I did, rather than a few seconds later? Perhaps because I felt tired of standing up. There is often a feeling, an experiential emotional state associated with the choice rather than the specific goal, suggesting that the conscious state of volition creates a somewhat opaque barrier between it and the mental representations of intentions, which makes sorting out the causal links difficult. Yet it cannot be that my experience of tiredness caused me to sit down, as a conscious state cannot cause a neural response directly, assuming a difference between conscious and unconscious states. There must have been an underlying goal formed in response to the same neural conditions that caused the state of tiredness, a goal which is decided upon, eventually causing the action of sitting down. Hence the intentional object of volition, though not always easy to introspect, serves as the bridge between the cognitive process and experience.
What are qualities of the experience of volition? It certainly can sometimes be quite salient. In literature and film it is common to find one distinct moment when a character must grapple with a difficult choice and makes it; where they are consciously aware of their own volition and the consequences of their voluntary actions. Common experience provides further anecdotal evidence; we must often make choices between sets of possible outcomes and can be vividly aware of our own conscious experience. We attribute qualities such as difficulty or significance to the experience of volition and will sometimes spend days or even weeks mulling over what we perceive to be a particularly important choice.
At the same time, the experience of volition can often be fleeting and very hard to pin down. There is what has been called a "thin" and "evasive" quality to the experience itself [Metzinger 2006]. Although a bit anecdotal, it will be helpful to think about some of the experiential qualities of volition to see what is meant by "thin" and "evasive." It is often hard to precisely pin down the feeling of willing an action because we don't often think about it. We make the majority of day-to-day actions without considering them in great depth; I might move my hand without consciously thinking about the movement itself or my will to move. Nonetheless, when confronted about the control of my actions, I would affirm being the one authoring them and report them as voluntary. Even if we do not explicitly think about walking, writing, waving, or typing, we still feel that we are in control and have consciously chosen to do each. Hence, an important aspect of the feeling of volition is control. It is often difficult to pin down a specific moment when we decide to take a particular action, as we often think about these moments after the action has taken place, hence the "evasive" quality. Even when considering the feeling of willing an action, it is usually not vivid or experientially rich and has a "thin" quality, as opposed to the "thick" experience of say, seeing a color. It's fairly safe to say that people who have had an experience of seeing red can identify the experience of seeing red again, and do so reliably. Is the same true for the experience of volition?
Volition has been thus far defined to be internally generated, goal-oriented and reasoned, involving an intention to move, as well as being a part of consciousness. The conscious state is most often defined by its thin and evasive qualities, while also preserving the possibility of being vivid and experientially rich in certain situations. In fact, the experience of volition can vary so wildly between a particularly salient experience to a murky one that it is not necessarily obvious that we are even discussing one single phenomenon. A likely cause of this confusion comes from one of the properties of volition itself, specifically its goal-oriented nature. In particular, we would expect that the clearer the goal, the more apparent the experience of volition becomes to the individual. Yet at the same time, the more the individual introspects and focuses on the experience itself, the more fleeting and "elusive" the experience of volition becomes [Metzinger 2006]. One possible explanation might come from a function of volition—that it provides a sense of self—and thus operates optimally when the individual is not attending to the experience itself. Such a feature would exist to preserve a constant and uninterrupted sense of agency in actions that are generally spontaneous and not the subject of much introspection. Nevertheless, we have identified two key aspects of the conscious experience of volition to look for when constructing a model: intention and control.
A model of volition
This diagram outlines what might be called the folk-psychological model of volition, with a clear causal chain between volition and action. It might not be a particularly biologically or epistemologically accurate model, but it is a good place to start because it is arguably how we all feel that volition works. We make a choice, there is likely some neural activity, and an action results. The main issue with this model stems from the first box, the volition. What is this? If it controls neural activity, it must have some neural basis, assuming a physicalist view. If it is a conscious state, it likely does not cause the action, rather it arises from some kind of other neural state which is the ultimate cause of the action. Revising the model to account for this requirement, we arrive at a more accurate folk-psychological view of volition.
Returning to Metzinger's "thin" and "evasive" qualities of the experience of volition presents difficulties for defining and experimentally studying the phenomenon. The first problem is the when. Many of the studies described below are concerned with when an action felt willed. Though time might be an intuitive metric to measure experimentally, it is not a particularly intuitive metric to introspect. We don't often think about when we choose to make actions, we usually just make them. There is plenty of reason to think that we might not be very good at the kind of introspection needed to reliably report when we were consciously aware of an intent to make an action. Try it yourself: break down a set of actions you've just made and try to precisely identify the moment you choose to do each in relation to the last. It is an extremely difficult task to introspect this kind of detail, which is what most studies of volition ask of participants.
Another problem that stems from the last is the question of the urge. When asking about a specific time you decide to make an action, there's an implicit assumption that there is a moment when you suddenly are aware of, or decide you are going to make the action. The "thin" quality of the experience of volition suggests that there might not be one conscious state that sticks out as the one associated with making the choice, but rather a vague sense of alignment between intent and action that arises over the course of deciding and acting. Many experiments and studies use the term "first felt the urge" which is suggestive of there being one moment in time where a choice is made, which, although a culturally popular notion, is not always an accurate reflection of our own experience of volition. This is not to say that time should be discarded as a metric in studies concerning volition. On the contrary, when the neural state that gives rise to the conscious state occurs in relation to the movement itself is very important, as if it were to occur anytime after the movement, the experience of will would be in some sense illusory.
The first possible model of volition suggests just this possibility. Called the "model of apparent mental causation," [Wegner and Wheatley 1999] it builds on evidence from a famous set of studies conducted by Libet in the 1980's. Libet's experiments were designed to test when a person becomes aware of their own "urge" to move (that ambiguous term first making an appearance in the literature). Participants were instructed to make a voluntary action by pressing a button at a time of their choosing, while also looking at a chronometer. They had to report the position of the dial on the chronometer when they "first felt the urge" to press the button [Libet 1985]. Libet found that participants felt what he called "conscious will" about 350ms after a readiness potential (RP), a kind of movement preparatory signal was fired [Haggard 2008, Libet 1985]. The results show that people feel the "urge" to move only after an RP (an unconscious process) had already been fired, which in turn could suggest that the "urge" to move might not be at all related to the actual mechanism causing the behavior.
Wegner used Libet's experiments to argue that volition cannot be a conscious process. He presents a model of volition wherein the conscious experience of choice is inserted retroactively into the stream of consciousness only after the brain has had some sensory feedback from an action [Wegner and Wheatley 1999, Wegner 2004]. It is a strong argument given the limitations of studies of volition, which rely on subjective report and make it very difficult to determine the cause of the conscious experience of volition. It is also a fairly plausible model when considering the qualities of voluntary actions—we usually only think about actions being willed after we make them—why wouldn't the feeling of conscious intention be retroactively inserted after sensory verification? Given that we also often mistake correlation for causation in the real world, couldn't our own experience of "causing" our actions be a case of the same problem?
However, there are a few possible points of weakness in Wegner's formulation of volition. Most notably, the model requires that there is some form of sensory feedback to produce the experience of conscious intention, as there is a causal connection between the two. Therefore, if we can show an instance of conscious volition (using Libet's phrase, an "urge to move") without sensory feedback, the model will no longer hold, as the proposed cause of conscious volition would be missing. Results from studies with epilepsy patients undergoing electrical stimulation of the supplementary motor area (SMA) present this evidence. With a certain amount of electrical stimulation (4-10 mA) to areas in the SMA, patients reported "a subjective ‘urge' to perform a movement or anticipation that a movement was going to occur," and crucially did not actually move [Fried et al. 2017, Fried et al. 1991, Haggard 2008]. Note that again the term "urge" is used without definition. It is not clear that Libet and Fried's versions of what exactly an "urge" is are the same, nor is it clear that the patients themselves were operating with a clear definition. It is likely an aspect of conscious experience that the experimenters assumed would be self-evident to the patients, which is certainly not a safe assumption.
Luckily, the authors quote the patients' reports, such as in one case an "urge to move right leg inward" [Fried et al. 1991]. Note a key part of this report is that it involves a specific goal: to make an action and move a limb in a certain direction. Previously, we defined volition to be both goal oriented and resulting in an intention to move, which coincides nicely with the reports given. Hence, it would appear as if we have a case of conscious volition without sensory feedback, suggesting that the experience of volition is not as illusory as Wegner makes it out to be.
Modifying Wegner's diagram to reflect the results of the SMA stimulation experiment and incorporating some of the phenomenal features of volition provides a new, more accurate model. It is still possible that the experience of volition is constructed in some part retroactively from sensory feedback, which supports the idea that pinning down the source of the feeling of volition is not always a trivial task. Another possibility is provided by Metzinger, who suggests that the feeling of volition is not "temporally segmented" and is instead composed of a set of conscious states across a range of time, which also contributes to the "thin" and "elusive" qualities [Metzinger 2006]. The key is that the first state of conscious volition is initiated before the action takes place, even if the actual experience of volition is comprised of a number of separate states constructed both before and after the action. There are now neural correlates to be accounted for, as well as the nature of their influence on action.
Neural correlates of volition
Before examining the literature surrounding the neuroscience of volition, let us briefly return to the question of the urge. It has become clear that the urge has been identified as a key component of any model of volition. Yet what an urge is has been left to the subject to introspect, resulting in an uncalibrated measure of this aspect of the experience of volition. Just as we reduced will to volition, could the urge be reduced to something better defined? Intention might begin to clear up the situation. Perhaps an urge to move is just an intention to move. This is a useful clarification because it uses language consistent with our working definition of volition; the phenomenon requires an intention to move. As has already been noted, intention is useful because it carries with it the concept of an intentional object, a goal state. The question now becomes more precise; instead of noting an urge, we have the intention to do x. Note that now time becomes a less important metric, as one can intend to do some action well before actually doing it, as opposed to a spontaneous urge. Additionally, intention can be confirmed with behavioral output: was the goal state resolved with movement?
One way of narrowing down the particular neural areas responsible for producing conscious volition, suggested by Fried and Haggard, is to examine the features of volition itself. For example, since volition leads to action, we would expect to see strong connections to motor control areas, and since it is goal-oriented, strong connections to monitoring and planning areas [Fried et al. 2017, Haggard 2008]. In the human brain, the areas known to be responsible for movement are the supplementary motor cortex (SMA) and pre-SMA. The prefrontal cortex is responsible for planning behavior and is known to have strong connections to motor areas such as the SMA. Based on the properties that we have assigned to volition, we would expect to see some sort of activations in these two areas of the cortex during the process of planning, deciding, and executing an action; i.e. intending to carry out some action. Both intention and control lie at the center of volition, so finding the neural areas responsible for both will begin to bring together a recognizable model of the experience of volition.
A study conducted by Lau et al. aimed to pin down these areas with fMRI. The researchers used Libet's task, asking participants to move their finger, but instead of focusing on the urge to move, the researchers asked participants to either "attend to their intention to move" or to the movement itself. Based on previous work, the researchers assumed that attending to a mental representation would increase the activity in the area responsible for the representation. In both cases they found activations in three areas: the right dorsal prefrontal cortex (DPFC), intraparietal sulcus (IPS), and pre-SMA. In the condition where the subjects attended to their intention, the BOLD signal increased in both the DPFC and pre-SMA [Lau et al. 2004].
There are two possible interpretations of the correlation between activity in the DPFC and pre-SMA. The first is that the DPFC caused the activity in the pre-SMA, and the second is that the DPFC monitors the increased activity in the pre-SMA. Both possibilities are plausible, as they would result in similar activations. The authors make the following conclusion:
Conscious awareness of visual stimuli has been demonstrated to be associated with the intensified BOLD signal in the relevant sensory area in the brain. Within the context of voluntary actions, it has been reported that awareness of the production of spontaneous movements is associated with an enhancement of the premovement slow potentials recorded over the medial frontal region. Taken together, if the intensity of activity in the pre-SMA correlates with the awareness of intention, our results suggest that attention to intention may be one mechanism by which effective conscious control of actions becomes possible [Lau et al. 2004].
The authors have made a connection between the increased BOLD signal and a conscious experience, essentially linking the neural process with the conscious one. Their view assigns the DPFC both a monitoring and control role to activity in the pre-SMA. If correct, this implies that the conscious experience of control over action is formed though this monitoring process, and that the pre-SMA's activity is a representation of intention. The authors do not define what exactly this is. Abstractly, we would expect a representation of intention to carry with it the intentional object; the goal state. It is unclear how to distinguish activity driven by a representation of intention versus the activity generated as a consequence of the execution of the intention's goal (preparing and executing the physical movement). One is the neural correlate of the conscious state and the other is the downstream neural activity driving the movement. Finally, a general activation of the pre-SMA does not necessarily indicate anything about the intention, as we would expect to see activity in the area preceding any movement, willed or not.
Lau et al. sidestep this issue by noting not the activation itself, but the increase in activation during the process of attending to the intention. They work under the assumption that attention increases BOLD signal level, which is precisely the relationship observed. Their task required subjects to introspect a conscious state, that is, draw attention to an internal process. It is worth reiterating just how potentially inaccurate this method is; we simply are not used to attending to the process of volition. Luckily in the case of this study, introspection was not used for a report, but rather as a second condition for observation of BOLD signal. The assumption the authors work with, though not completely safe for the aforementioned reasons, is necessary to neatly wrap up their argument and tie together the unconscious process to the conscious one by way of representation. With this said, what evidence can be taken from Lau et al. to provide support for a neural basis to the model of volition? Activity was observed in the pre-SMA regardless of the introspection condition, so further inquiry into this area seems justified. Their view of the DPFC's role also seems plausible given the function of the area, but additional exploration of the relationship between this area and the pre-SMA would be quite beneficial. Revising our model reflect these findings, we find a neural basis for the first aspect of the experience of volition: control.
Returning to stimulation studies, additional evidence for the role of the pre-SMA and SMA in voluntary movement has been developed by Desmurget et al. in a series of simulations to awake brain surgery patients. The researchers focused on two areas in particular, the pre-SMA and posterior parietal cortex (PPC). Under differing stimulation conditions, they noted three categories of results: the patient unconsciously moved, the patient reported an "urge" to move, or the patient thought they moved even though no such action took place. The first result, an unconscious movement, took place solely under stimulation to the pre-SMA. The other two results manifested under stimulation to the PPC. One patient reported an intention to move ("I felt a desire to lick my lips") and another reported having made a movement that never happened ("I moved my mouth, I talked, what did I say?") [Desmurget et al. 2009]. Additionally, in describing their experience, the patients used words such as will, desire, and want, all unprompted by the researchers.
What can be made of these reports? A particularly important aspect is that each contains a specific goal state, some perceived to have already been manifested. This is the intentional object. The simulations appear to have created intentions before, during, and after execution. The researchers have also avoided some of the problems of introspection by not prompting patients to report when they felt an urge, but rather allowing the patients to feely report what they were feeling without the constraint of time. It is also important that the researchers were able to generate the same actions that patients described intending to do through additional stimulation, as this gives additional credit to the subjective reports provided.
The authors provide a theory concerning the method by which an intention is created and developed. They suggest that the PPC stores a kind of "map of intentions," where stimulation at lower intensities activates nodes within this map leading to a conscious intention, and stimulation at higher intensities leads to a prediction about the movement made before the execution of that movement (a view based on forward modeling, the proposition that the brain predicts outcomes to events to improve possible responses). Their proposition is supported by the stimulation results, where the report of making a movement that never happened was prompted by higher intensities of stimulation to the PPC, and lower intensities lead to a conscious intention. Their view is further supported by the fact that stimulation to the pre-SMA resulted in an unconscious movement and no intention, suggesting that conscious volition is constructed less from connections from motor control areas and more from predictions made prior to the onset of movement.
The theory also aligns well with Libet's experiments, as the awareness of intention occurs only after the results of the movement have been predicted. Hence it would not be surprising to see a lag between neural activity preceding the movement and a report of awareness, even if that report was completely accurate. The relationship is noted in the diagram below, where time t1 marks the point at which conscious awareness of the intention begins. Note that at this point the preparation for the movement has already begun, and the prediction of the movement has just ended.
Further support for this model comes from a study by Fried et al. concerned with predicting movement from recordings of neurons in the pre-SMA and SMA. The authors were interested in reexamining Libet's task without measuring an RP, which has been called into question as a potentially unreliable neural marker of action, some suggesting it to be most influenced by random electrical activity [Schurger et al. 2012]. Instead, the researchers used an array of electrodes implanted in epilepsy patients. They repeated Libet's task, recording when the patient reported being aware of their intent to move as well as the actual movement time. They found similar results; the pre-SMA and SMA neurons begin to fire more rapidly a few hundred milliseconds before the subject reported being aware of their intention to move. Using an SVM classifier, the researchers were able to predict with 90% accuracy that a subject had begun to prepare for movement 500ms prior to the reported time of awareness, and 70% accuracy 1s before the reported awareness [Fried et al. 2011].
They also used the algorithm to predict when someone would become aware of their intention to move. The SVM predicted this time just 152ms prior to the actual reported time, on average. It was also able to predict that people had reported feeling an urge to move compared to controls with 98% accuracy. All of these predictions were made using just the recorded activity of the neurons in pre-SMA and SMA.
These results support two important points. First, we can be fairly certain that preparatory neural activity precedes an awareness of intention. This is an important part of sorting out the order in which unconscious neural process take place, though it can be confusing when attempting to reconcile the conscious experience of volition with its potential causes. Second, people's introspection of their own intention to act is inaccurate, but only relatively so. The authors report variance of around 200ms between trials, but the fact that the classifier was just 150ms off the true time on average suggests that people's reports are generally around a similar range, even if there is fluctuation within that range.
Finally, it is worth noting that although this model is illustrated with clearly defined boxes denoting each process, these categories are quite fuzzy. Metzinger's analysis of the experience of volition pointed to thin and elusive qualities not well defined by time. Although there was some agreement on when people experienced a feeling of intention in both Libet's and Fried et al.'s work, the task being performed was very simple and a much reduced version of volition in day-to-day actions. Though the experimental data points to a model with a delay between neural preparation and awareness of intention, further exploration is warranted for different tasks that lead to both murky and salient experiences of volition.
Pulling together a model of volition
Our model of volition begins with the selection of intention (1). This is a process started in the PPC, caused by upstream neural states. Desmurget et al. provide evidence that stimulation in this area could prompt an experience of intention or the experience of a goal having already resolved with movement, suggesting that the experience of intention arises after a prediction process has completed (2). The representation of the intentional object from the PPC (1) drives neural preparation for movement in the pre-SMA and SMA (3). Lau et al. propose that the DPFC serves as a way of monitoring this preparatory activity and generates a conscious awareness of the action before it is executed (4). After the intentional object has been resolved with a movement, sensory feedback provides information for a check to determine if the goal state and its resolution were in alignment, leading to a retroactive experience of agency over actions (5). Wegner's model provides support for the idea that the experience of volition is often constructed after the movement has taken place and is built off downstream feedback. Finally, we find the entire experience of volition to be comprised of a number of conscious states with varying intensities (6) that contribute to a thin and illusive experience, as suggested by Metzinger.
There are a few areas of this model that have yet to be explored in much detail that could help in building a much higher resolution view of volition at an experiential, computational, and neural level. In particular, the process involved in the prediction of movement is at the moment not very well defined or understood. Additionally, the theory concerning a "map of intentions" in the PPC c0uld be explored much more detail; for example what is the neural code that allows for the initiation of an intention, and how are the relevant nodes selected? How are intentions formed in contexts with novel constraints? How do upstream events such as thought play a role in the selection process, and what part of this process is done consciously? There is also the question of the relationship between the DPFC and pre-SMA that was not resolved with much certainty in the Lau et al. study; does the DPFC have the ability to make changes to activations in the pre-SMA after an intention has been selected? Presumably we can consciously choose to stop a movement before executing; how does such a choice perpetuate through this model? How are more complex movements than just a lifting a finger executed in this system; does the process work in a different manner for movements that are not premeditated?
The major hurdle to the study of the experience of volition is methodological. Too many studies use Libet's task, which as previously explored is limited in a number of ways. It was useful to show that awareness of intention follows an unconscious process, but problems associated with introspection of an often murky experience and assigning a time are quite formidable. Of course, given that the experience of volition is purely subjective, what other metric exists? There are not many other than behavioral output. Despite requiring some form of report, stimulation studies likely provide the best direction for future work in this area.
Finally, let us return to the notion of the experience of will. The model developed over the course of this inquiry concerns one very specific kind of will: the feeling of will experienced before, during, and after the execution of a movement. The systems proposed to compose this model apply only to physical movement. There are many more classes of will beyond movements, and this model cannot account for them. Many have used the Libet experiments to posit that "free will" does not exist, a conclusion that the experiments do not support for nearly any definition of "free" or "will." Though both Libet's work and this model conclude that neural preparatory activity precedes awareness of intention, neither says anything about the selection of intention, a process likely to be a key part of any conception of "free will." A model has been proposed just for the experience of volition, a process that presents significant challenges for future study in its own right.
Thanks to Professor Wayne Wu and anonymous peers in the Spring 2019 class of the Psychology and Neuroscience of Consciousness at Carnegie Mellon for feedback on earlier versions of this paper.
Desmurget, Michel, Karen T. Reilly, Nathalie Richard, Alexandru Szathmari, Carmine Mottolese, and Angela Sirigu. "Movement intention after parietal cortex stimulation in humans." science 324, no. 5928 (2009): 811-813. PDF.
Fried, Itzhak, Patrick Haggard, Biyu J. He, and Aaron Schurger. "Volition and action in the human brain: processes, pathologies, and reasons." Journal of Neuroscience 37, no. 45 (2017): 10842-10847. PDF.
Fried, Itzhak, Amiram Katz, Gregory McCarthy, Kimberlee J. Sass, Peter Williamson, Susan S. Spencer, and Dennis D. Spencer. "Functional organization of human supplementary motor cortex studied by electrical stimulation." Journal of Neuroscience 11, no. 11 (1991): 3656-3666. PDF.
Fried, Itzhak, Roy Mukamel, and Gabriel Kreiman. "Internally generated preactivation of single neurons in human medial frontal cortex predicts volition." Neuron 69, no. 3 (2011): 548-562. PDF.
Haggard, Patrick. "Human volition: towards a neuroscience of will." Nature Reviews Neuroscience 9, no. 12 (2008): 934-946. PDF.
Haggard, Patrick, and Martin Eimer. "On the relation between brain potentials and the awareness of voluntary movements." Experimental brain research 126, no. 1 (1999): 128-133. PDF.
Lau, Hakwan C., Robert D. Rogers, Patrick Haggard, and Richard E. Passingham. "Attention to intention." science 303, no. 5661 (2004): 1208-1210. PDF.
Libet, Benjamin. "Unconscious cerebral initiative and the role of conscious will in voluntary action." Behavioral and brain sciences 8, no. 4 (1985): 529-539. PDF.
Metzinger, Thomas. "Conscious volition and mental representation: Toward a more fine-grained analysis." (2006). PDF.
Schurger, Aaron, Jacobo D. Sitt, and Stanislas Dehaene. "An accumulator model for spontaneous neural activity prior to self-initiated movement." Proceedings of the National Academy of Sciences 109, no. 42 (2012): E2904-E2913. PDF.
Wegner, Daniel M. "Précis of the illusion of conscious will." Behavioral and Brain Sciences 27, no. 5 (2004): 649-659. PDF.
Wegner, Daniel M., and Thalia Wheatley. "Apparent mental causation: Sources of the experience of will." American psychologist 54, no. 7 (1999): 480. PDF.