Вы находитесь на странице: 1из 12

Neuroethics (2010) 3:161172 DOI 10.



Testing Free Will

Alfred R. Mele

Received: 2 September 2008 / Accepted: 15 October 2008 / Published online: 2 December 2008 # Springer Science + Business Media B.V. 2008

Abstract This article describes three experiments that would advance our understanding of the import of data already generated by scientific work on free will and related issues. All three can be conducted with existing technology. The first concerns how reliable a predictor of behavior a certain segment of type I and type II RPs is. The second focuses on the timing of conscious experiences in Libet-style studies. The third concerns the effectiveness of conscious implementation intentions. The discussion of first two experiments highlights some important problems with certain inferences made on the basis of existing data in scientific work on free will. The discussion of the third calls attention to powerful evidence that conscious intentions sometimes are among the causes of corresponding actions. This evidence has been largely ignored in the literature on free will. Keywords Consciousness . Decisions . Free will . Intentions . Libet . Neuroscience

Introduction Scientific work on free will has gained a lot of momentum in recent years. It features some striking
A. R. Mele (*) Department of Philosophy, Florida State University, Tallahassee, FL, USA e-mail: almele@fsu.edu

claims. Benjamin Libet [1, 2] contends that the brain decides to initiate actions about a third of a second before the person becomes conscious of the decision and that the remaining window of opportunity for free will to get involved is tinyabout 100 ms. Daniel Wegner [35] argues that conscious intentions are not among the causes of corresponding actions. If Wegner is right, then if only beings whose conscious intentions are sometimes among the causes of corresponding actions are capable of acting freely, even Libets tiny window of opportunity for free will is an illusion. Elsewhere, I have argued that these striking claims are not warranted by the data Libet, Wegner, and others offer in support of them [6, 7], chapter 2, [8 10]. Here I propose three experiments that would advance our understanding of the import of data that have already been gathered. I would conduct these experiments if I could. But I am a philosopher; and, with rare exceptions, philosophers do not have labs. I hope that the case I make here for the potential utility of these experiments will dispose others to conduct them. However, my discussion of the first two experiments is partly a rhetorical device for making salient some important problems with certain inferences made on the basis of existing data, and part of the aim of my discussion of the third experiment is to call attention to some powerful evidence that conscious intentions sometimes are among the causes of corresponding actionsevidence that has been largely ignored in the literature on free will.


A.R. Mele

Some Conceptual Background: Decisions, Intentions, Wanting, and Consciousness Some conceptual background is in order. I focus on the concept of deciding: that is, deciding to do somethingpractical decidingas opposed to deciding that something is true (as in Ann decided that Bob was lying). And I briefly discuss its connections to some related concepts. Like many philosophers, I take deciding to A to be an action.1 In my view, it is a momentary action of forming an intention to A [14], chapter 9). Deliberation about what to do is not momentary, but it must be distinguished from an act of deciding that is based on deliberation. Not all intentions are formed in acts of deciding. Consider the following: When I intentionally unlocked my office door this morning, I intended to unlock it. But since I am in the habit of unlocking my door in the morning and conditionswere normal, nothing called for a decision to unlock it [15], p. 231). If I had heard a fight in my office, I might have paused to consider whether to unlock the door or walk away, and I might have decided to unlock it. But given the routine nature of my conduct, there is no need to posit an action of intention formation in this case. My intention to unlock the door may have been acquired without having been actively formed. If, as I believe, all decisions about what to do are prompted partly by uncertainty about what to do [14], chapter 9), in situations in which there is no such uncertainty, no decisions will be made. This is not to say that, in such situations, no intentions will be acquired. Some decisions and intentions are about things to do straightaway. They are proximal decisions and intentions. Othersdistal decisions and intentionsare about things to do later. A shy student who has been thinking about when to raise his hand to attract his teachers attention decides to raise it now. This is a proximal decision. Later, after thinking about when to start writing a term paper, he decides to start it next Tuesday. This is a distal decision. Although Libet and Wegner focus on proximal decisions and intentions, others study their distal counterparts (see Experiment 3: Implementation Decisions). Deciding to do something should be distinguished from wanting (or having an urge) to do it. Sometimes people want to do things that they decide not to do. And often, when people want to do each of two incompat1

ible thingsfor example, meet some friends for dinner at 7:00 and attend a lecture at 7:00they settle matters by deciding which one to do. Just as deciding should be distinguished from wanting, so should intending. Intending to do something is more tightly connected to action than is merely wanting to do it. The account of practical deciding sketched here is not the only account of it. For critical discussion of alternative accounts, see [14], chapter 9. For the purposes of this article, a virtue of the account just sketched is that it is consonant with Libets apparent conception of deciding. A comment on consciousness is in order. The standard measure of subjects consciousness of their intentions in scientific studies is reports subjects make to the effect that they were conscious of certain intentions at certain times (the reports do not need to be oral. Manual reports are discussed in Experiment 2: On the Bearing of B-Time on C-Time). In the words of Richard Passingham and Hakwan Lau, the operational index of consciousness is the ability to report [16], p. 67); and notice that their assertion is about consciousness in general whereas my assertion in the preceding sentence is just about consciousness of ones intentions. (I should add that one who does not realize that the operational index of the ability to report is an actual report may misread the quoted assertion). Now, consciousness may be such that some occurrences or states that properly count as conscious for the ordinary adult human beings who are the loci of those occurrences or states are not reportable by those human beings. Unreportable consciousness is not my concern here. It is consciousness (or awareness) that is measurable by subjects reports that concerns me in this article; for it is consciousness of this kind, or in this sense, that is explicitly at issue in the scientific work that I discuss here. I call it report-level consciousness (or awareness). In the remainder of this article, I write simply in terms of consciousness (or awareness) and count on the reader to remember that report-level consciousness (or awareness) is at issue. Whether all decisions are consciously made is a matter I leave open.

Some Empirical Background: Libets Studies Libet contends both that the brain decides to initiate or, at least, prepare to initiate [certain actions] before there is

See [11], pp. 174-76, [12], pp. 25455, and [13], p. 94.

Testing free will


any reportable subjective awareness that such a decision has taken place [1], p. 536)2 and that If the act now process is initiated unconsciously, then conscious free will is not doing it [18], p. 62; see [2], p. 136). He also contends that once we become conscious of our proximal decisions, we can exercise free will in vetoing them [1, 2, 19], pp. 13749). Libet has many critics and many supporters. Some people follow him part of the way. They accept the thesis about when and how decisions are made but reject the window of opportunity for free will as an illusion [20]. In some of Libets studies, subjects are regularly encouraged to flex their right wrists whenever they wish. In such subjects who do not report any preplanning of flexings, electrical readings from the scalp (EEGs) averaged over at least 40 flexings for each subject show a shift in readiness potentials (RPs) that begins about 550 ms before the time at which an electromyogram (EMG) shows relevant muscular motion to begin [1], pp. 52930). These are type II RPs (p. 531). Subjects who are not regularly encouraged to act spontaneously or who report some preplanning produce RPs that begin about half a second earliertype I RPs. The same is true of subjects instructed to flex at a preset time [21], p. 325. According to a common use of readiness potential [RP], it is a measure of activity in the motor cortex that precedes voluntary muscle motion and, by definition, averaged EEGs generated in situations in which there is no muscle burst do not count as RPs. Libets use of the term is broader. For example, because there is no muscle burst in the veto experiment described shortly, some scientists would refer to what Libet calls the veto RP [p. 538] as an event-related brain potential [or ERP] rather than an RP). Subjects are also instructed to recallthe spatial clock position of a revolving spot at the time of [their] initial awareness ([1], p. 529) of something, x, that Libet variously describes as a decision, intention, urge, wanting, will, or wish to move.3 On average,

in the case of type II RPs, RP onset precedes what the subjects report to be the time of their initial awareness of x (time W) by 350 ms. Reported time W, then, precedes the beginning of muscle motion by about 200 ms. The results may be represented as follows: Libets results for type II RPs : 550 ms 200 ms 0 ms RP onset reported time W muscle begins to move Again, in Libets view, consciousness opens a tiny window of opportunity for free will in his subjects. If a subject, Wilma, becomes conscious of her intention at 150 ms, and if by 50 ms her condition is such that the act goes to completion with no possibility of its being stopped by the rest of the cerebral cortex [2], p. 138), her window is open for 100 ms. Libet writes: The role of conscious free will [is] not to initiate a voluntary act, but rather to control whether the act takes place. We may view the unconscious initiatives as bubbling up in the brain. The conscious-will then selects which of these initiatives may go forward to an action or which ones to veto and abort [19], p. 54). Libet mentions what he regards as two sources of evidence for veto power. The first is an experiment in which subjects are instructed to prepare to flex at a prearranged clock time and to veto the developing intention/preparation to act about 100 to 200 ms before [that] time [1], p. 538). Subjects receive both instructions at the same time. Libet writes: a ramplike pre-event potential was still recordedresembl[ing] the RP of self-initiated acts when preplanning is present.The form of the veto RP differed (in most but not all cases) from those preset RPs that were followed by actual movements [in another experiment]; the main negative potential tended to alter in direction (flattening or reversing) at about 150 250 ms before the preset time.This difference suggests that the conscious veto interfered with the final development of RP processes leading to action.The preparatory cerebral processes associated with an RP can and do develop even when intended motor action is vetoed at approx-

Elsewhere, Libet writes: the brain has begun the specific preparatory processes for the voluntary act well before the subject is even aware of any wish or intention to act ([17], p. 263).

Libet, Gleason et al. report that the subject was asked to note and later report the time of appearance of his conscious awareness of wanting to perform a given self-initiated movement. The experience was also described as an urge or intention or decision to move, though subjects usually settled for the words wanting or urge ([22], p. 627).


A.R. Mele

imately the time that conscious intention would normally appear before a voluntary act [1], p. 538).4 Subjects reports about unsolicited vetoing are Libets second source of evidence for veto power. Subjects encouraged to flex spontaneously (in nonveto experiments) reported that during some of the trials a recallable conscious urge to act appeared but was aborted or somehow suppressed before any actual movement occurred; in such cases the subject simply waited for another urge to appear, which, when consummated, constituted the actual event whose RP was recorded [1], p. 538). Libet says that his subjects were free not to act out any given urge or initial decision to act; and each subject indeed reported frequent instances of such aborted intentions (p. 530). Libet associates veto power with some pretty fancy metaphysics (see, for example, [19]). I set the metaphysical issues aside here and concentrate on the empirical ones.


The time the subject believes to be C-time when responding to the experimenter s question about C-time.

Experiment 1: A Stop-Signal Study The following labels facilitate discussion of Libets data and some of his inferences: E-time C-time The time at which a proximal decision is made or a proximal intention, urge, etc. is acquired. The time of the onset of the subjects consciousness of an item of the kind just specified.

For a more thorough discussion of the experiment, see Libet et al. [22, 23]. In [7] (p. 34), I explain that Libet implausibly describes what is vetoed here as intended motor action. The subjects were instructed in advance not to flex, but to prepare to flex at the prearranged time and to veto this; and they intentionally complied with the request. They intended from the beginning not to flex at the appointed time. So what is indicated by the segment of what Libet refers to as the veto RP that precedes the change of direction? Presumably, not the presence of an intention to flex; for then, at some point in time, the subjects would have both an intention to flex at the prearranged time and an intention not to flex at that time. And how can a normal agent be in this condition? If you were to intend now to pick up a tempting donut two seconds from now while also intending now not to pick up the donut two seconds from now, what would you do? Would you soon start reaching for it with one hand and quickly grab that hand with your other hand to halt its progress toward the donut? This is far from normal behavior.

Libet contends that average E-time is 550 ms for subjects who are regularly encouraged to flex spontaneously and report no preplanning. And he arrives at an average C-time of 150 ms by adding 50 ms to his average B-time (200 ms) to correct for what he believes to be a 50 ms bias in subjects reports (for alleged evidence of the existence of this bias, see [1], pp. 53435 and [2], p. 128). Regarding E-time in subjects who display type II RPs, Libets apparent reasoning is straightforward: (1) the act now process is initiated unconsciously at 550 ms on average [18], p. 62); (2) proximal decisions or intentions are what initiate act now processes; (3) so average E-time in these subjects is 550 ms.5 Elsewhere, on the basis of various data, I have argued that claim 3 is less plausible than the claim that an early segment of the type II RP is associated with a potential cause of a subsequent proximal intention to flex [7], chapter 2, [10], chapters 3 and 4). Here, I take up a related issue. Daniel Dennett echoes a common judgment when he asserts that the type II RP is a highly reliable predictor of flexing [24], p. 229). Even if this is so, is the brain activity associated with, say, the first 300 ms of this RPcall it type 300 activitya highly reliable predictor of a flexing action or even a muscle burst? In fact, this is not known. In the experiments that yield Libets type II RPs, it is the muscle burst that triggers a computer to make a record of the preceding brain activity. In the absence of a muscle burst, there is no record of that activity. So, for all anyone knows, there were many occasions on which type 300 activity occurred in Libets subjects and there was no associated muscle burst. Recall the subjects who reported spontaneously vetoing conscious urges to flex. Libet points out that in the absence of the muscles electrical signal when being activated, there was no trigger to initiate the computer s recording of any RP that may have preceded the veto [2], p. 141). For all anyone knows, type
5 Libet is inclined to generalize from his findings. He writes: our overall findings do suggest some fundamental characteristics of the simpler acts that may be applicable to all consciously intended acts and even to responsibility and free will ([1], p. 563).

Testing free will


300 activity was present before the urges were suppressed. William Banks reports that although the common view, which he propounds, is that the unconscious process as measured by the RP is sufficient in and of itself to initiate action and the conscious decision is an ineffective latecomer, along for the ride, this view should be tested [25], p. 239). He asserts that for the common view to be verified, we need an absence of cases in which RP potentials are found but do not eventuate in the response, and, he adds, It is possible that robust RPs crop up frequently during the critical interval, but only when conscious decision is reported is there a response. In this case the unconscious process would clearly not be a sufficient cause of the action. We do not detect the ineffective RPs because recording RPs requires back-averaging from the response. The RPs that do not eventuate in a response would therefore not be discovered. Two points should be made straightaway. First, in the sense in which Banks is using RP here (according to which it is not true by definition that there are no RPs without muscle bursts), it is false that recording RPs requires back-averaging from a muscle burst. (Back-averaging from muscle bursts involves using muscle bursts to trigger the making of a record of preceding brain activity.) Recall that EEGs (what Libet calls the veto RP) were recorded for subjects instructed to prepare to flex at t but not to flex then. The EEGs were back-averaged from t, and they resembled type I RPs until about 150250 ms before t [1], p. 538)until time v, for short. Second, this is evidence that the brain events indicated by the segment of the type I RPs that precedes time v are not sufficient for producing a flexingand, more precisely, that they are not sufficient for producing events that are sufficient for producing a flexing (that is, less distant sufficient causes). If (1) until time v, the averaged veto EEGs and the averaged EEGs for type I RPs are produced by neural events of the same kind, then (2) the occurrence of events of that kind is not sufficient for producing (events that are sufficient for producing) a flexing. For if 1 is true and 2 were false, the subjects in the veto experiment would have flexed.

I return to the shorter (550 ms), type II RPs exhibited by the subjects in Libets main study and to type 300 activity. Some scientists are trying to develop techniques for identifying in single trials the EEG components of ERPs (an ERP, by definition, is computed by averaging over a series of trials). If and when a reliable technique of this kind is developed, a stop-signal task can be added to Libet-style studies to test for type 300 activity that is not followed by a muscle burst (for a primer on stop signal experiments, see [26]). Subjects can be given a version of Libets instructions that include the additional instruction not to flex if they detect a stop signal (for example, the white face of a Libet clock turning red). A computer can be programmed to issue the stop signal in, say, 25% of trials shortly after it detects what looks like a segment of type 300 activity.6 The time at which the signal is issued can be varied: for example, 100 ms into what looks like type 300 activity, 200 ms into it, and 300 ms into it. If subjects refrain from flexing on a statistically significant number of stop signal trials at one or more of these intervals, we will have learned something about type 300 activitynamely, that it lacks the sufficiency at issue. And if an externally issued stop signal can be effective in this experimental situation, the same may be true of an internally generated onefor example, a conscious decision to veto an urge to flex. Because the experiment just described cannot yet be conducted, I mention it only to help set the stage for a proposal of a currently feasible experiment. The remaining background is the following trio of points. First, the main difference between type I and type II RPs, in Patrick Haggards words, is that the former have earlier onsets than the latter [27], p. 49). Second, much of the type II RP (all but an early segment of it) resembles a late segment of the type I RP. Third, type I RPs are observed, for example, in studies in which subjects watching a Libet clock are instructed to flex at a preset time. In the proposed experimentExperiment 1 subjects who have already produced type I and type II RPs are instructed to flex when the dot hits a certain point, p, on a Libet clock unless they detect a stop signal. EEGs are back-averaged from point p.
6 In conversation, Walter Sinnott-Armstrong suggested an imaginary experiment that involved detection of roughly this kind.


A.R. Mele

On 25% of the trials, the stop signal is issued (a change of clock color from white to red). The times at which it is issued are divided into three equal groups: p-450 ms, p-350 ms, and p-250 ms. Experimenters will already have checked each subjects graphs for type I and type II RPs and identified the longest similar looking segments of the two graphs. Call this the common segment. They can look for an averaged EEG that resembles, say, the first half of the common segmenta type H EEG, for shortin each of the three main conditions: (1) no stop signal, (2) effective stop signal (i.e., a stop signal and no muscle burst), and (3) ineffective stop signal. Possibly, it will be discovered that, in each of these three conditions, there is a type H EEG. If a type H EEG is discovered in condition 2, we can conclude that brain activity associated with a type H EEG is not sufficient for a subsequent muscle burst and that a type H EEG is not, in Dennetts words, a highly reliable predictor of flexing [24], p. 229).7 Again, if an externally issued stop signal can be effective in this experimental situation, the same may be true of an internally generated one. I use may be for a reason: to indicate that the suggestion is speculative. For discussion of some problems with a study in which subjects were asked, in effect, to generate internal stop signals [28], see [29]. The purpose of Experiment 1 is to produce evidence about whether a certain segment of type I and type II RPs is correlated with brain activity that is a highly reliable predictor of flexing. Even if an entire type I or type II RP is a highly reliable predictor of flexing, a substantial segment of either type of RP might not be. For example, the existence of that segment might be compatible with successful vetoing. Recall the hypothesis that an early segment of the type II RP is associated with a potential cause of a proximal intention. A related hypothesis is that this segment is associated with a potential cause of an overt action. To the extent to which a type H EEG resembles a segment of a type II RP, the discovery that a type H EEG is present in effective stop signal scenarios would provide support for the hypothesis that the brain activity associated
Obviously, researchers would test for an averaged EEG associated simply with detecting the change in clock color so it can be subtracted from the averaged EEG exhibited by the same subject in stop-signal trials.

with a type H EEG is at most a potential cause of action. Consider the following two propositions: P1. Type 300 activity is sufficient to produce (events that are sufficient for producing) a muscle burst around 0 ms. P2. P1 would be true if not for the possibility of a conscious veto. Those who believe that one or the other of these propositions has been shown to be true either do not realize that, in the experiments that yield Libets type II RPs, the muscles electrical signal when being activated is what triggers the computer to make a record of the preceding brain activity for the purposes of averaging [2], p. 141) or do not recognize the implications of this. How can we, on the basis of the data, be justified in believing that type 300 activity has the actual or counterfactual sufficiency at issue, if no one has looked to see whether type 300 activity is ever present in cases in which there is no muscle burst around 0 ms? The answer is simple: we cannot.8 V. S. Ramachandrans discussion of the following thought experiment is interesting in connection with Experiment 1: Im monitoring your EEG while you wiggle your finger .I will see a readiness potential a second before you act. But suppose I display the signal on a screen in front of you so that you can see your free will. Every time you are about to wiggle your finger, supposedly using your own free will, the machine will tell you a second in advance! [30], p. 87). Ramachandran asks what you would experience, and he offers the following answer: There are three logical possibilities. (1) You might experience a sudden loss of will, feeling that the machine is controlling you, that you are a mere puppet and that free will is just an illusion.(2) You might think that it does not change your sense of free will one iota, preferring to believe that the machine has some sort of spooky paranormal precognition

One who deems a segment of what Libet calls the veto RP ([1], p. 538) to match averaged EEGs for type 300 activity may regard the matching as evidence that type 300 activity is not sufficient to produce (events that are sufficient for producing) a muscle burst around 0 ms.

Testing free will


by which it is able to predict your movements accurately. (3) You mightdeny the evidence of your eyes and maintain that your sensation of will preceded the machines signal (p. 87).

This list of possibilities is not exhaustive. Here is another. You might experience an urge to test the machines powers; and you might wonder whether you can watch for the signal to appear on the screen and intentionally refrain from wiggling your finger for a minute or two after you see it. Libets data definitely leave it open that you can do this. You might even display an EEG that resembles the EEG displayed by Libets subjects in the veto experiment. Perhaps you hit on the imagined test because it occurs to you that (1) Whenever you wiggle your finger, signal S appears a second before you wiggle it does not entail (2) Whenever signal S appears, you wiggle your finger a second later. (A brain event signified by signal S may be causally necessary for your wiggling your finger without causally ensuring that you will wiggle it. Incidentally, whenever Lydia wins a lottery prize, she buys a lottery ticket before she wins; but, to her dismay, it is false that whenever she buys a lottery ticket, she wins a lottery prize). If you succeed in your watch-and-refrain attempt, you might have the further thought that S is a sign of the presence of a potential cause of a proximal intention or decision to wiggle your finger and that, even when that potential cause is present, you may decide not to wiggle your finger and behave accordingly. But if this is how you are thinking, then, provided that you are thinking clearly, you will not see the machine as controlling you. And, clear thinker that you are, you will neither be tempted to believe that the machine has paranormal predictive powers nor moved to deny the evidence of your eyes. I do not propose that Ramachandrans thought experiment be turned into an actual experiment. My purpose in discussing it was to reveal a relevant confusion. From (1) the fact that when we back-average from the muscle burst, we see a distinctive EEG, it does not follow that (2) whenever an early segment of that EEG is present, there is a subsequent muscle burst. And, of course, it is 2 that guides Ramachandrans thinking about free will in the passages just discussed. Instead, I propose Experiment 1. This experiment would

generate evidence that bears on the question how the type H EEG is to be interpreted. The finding that this averaged EEG is not a reliable predictor of a muscle burst would contribute to the case for the thesis that the brain activity associated with the first half or so of the type II RP is at most a potential cause of overt actions. For all anyone knows, that activity may be a potential cause of proximal intentions that are more proximal causes of the actions. The proximal intentions may emerge much later than Libets 550 ms E-time (see [7], pp. 4245 for independent evidence for this). They may even emerge, on average, around reported time W in subjects instructed to make W judgments.

Experiment 2: On the Bearing of B-Time on C-Time C-time, again, is the time of the onset of the subjects consciousness of a proximal decision, intention, urge, etc.; and B-time is the time the subject believes to be C-time when responding to the experimenter s question about C-time. Subjects reports of B-time are used as evidence about C-time. One connection in which C-time is important to Libet is his position on veto power. Whether subjects in Libets studies are ever conscious of relevant proximal urges or intentions early enough to veto them, as he claims, depends partly on what their C-times are. There is a lively literature on how accurate B-times are likely to bethat is, on how likely it is that they closely approximate C-times (for a review, see [31]. This is not surprising. Reading the position of a rapidly revolving dot at a given time is a difficult task, as Wim van de Grind observes [31], p. 251). The same is true of relating the position of the dot to such an event as the onset of ones consciousness of a proximal intention to flex a wrist. Patrick Haggard notes that the large number of biases inherent in cross-modal synchronization tasks means that the perceived time of a stimulus may differ dramatically from its actual onset time. There is every reason to believe that purely internal events, such as conscious intentions, are at least as subject to this bias as perceptions of external events [32], p. 82). One fact that has not received sufficient attention in the literature on accuracy is that individuals


A.R. Mele

display great variability of B-times across trials. Patrick Haggard and Martin Eimer [33] provide some relevant data. For each of their eight subjects, they locate the median B-time and then calculate the mean of the premedian (i.e., early) B-times and the mean of the postmedian (i.e., late) B-times. At the low end of variability by this measure, one subject had mean early and late B-times of 231 ms and 80 ms and another had means of 542 ms and 351 ms (p. 132). At the high end, one subjects figures were 940 ms and 4 ms and another s were 984 ms and 253 ms. Bear in mind that these figures are for means, not extremes. These results do not inspire confidence that B-time closely approximates C-time. If there were good reason to believe that C-times vary enormously across trials for the same subject, we might not find enormous variability in a subjects B-times worrisome in this connection. But there is good reason to believe this only if there is good reason to believe that B-times closely approximate C-times; and given the points made about cross-modal synchronization tasks in general and the cross-modal task of subjects in Libet-style experiments, there is not. Another factor that may make it difficult for subjects to provide B-times that closely approximate C-times is their uncertainty about exactly what they are experiencing. As Haggard observes, subjects reports about their intentions are easily mediated by cognitive strategies, by the subjects understanding of the experimental situation, and by their folk psychological beliefs about intentions [32], p. 81). He also remarks that the conscious experience of intending is quite thin and evasive [33], p. 291). Even if the latter claim is an overstatement and some conscious experiences of intending are robust, the claim may be true of many of the experiences at issue in Libet-style studies. One can well imagine subjects wondering occasionally whether, for example, what they are experiencing is an intention (or urge) to act or merely a thought about when to act or an anticipation of acting soon. Hakwan Lau and coauthors say that they require their subjects to move a cursor to where they believed the dot on a Libet clock was when they first felt their intention to press the button [34], p. 82; emphasis mine). One should not be surprised if some subjects given such an instruction were occasionally to wonder whether they were experi-

encing an intention to press or just an urge to press, for example. (Presumably, at least some lay folk treat intentions and urges as conceptually distinct, as dictionaries do.) Subjects may also wonder occasionally whether they are actually feeling an intention to press or are mistakenly thinking that they feel such an intention. One way to seek to reduce variability in a subjects B-times is to give him or her a way of conceiving of, for example, making a conscious proximal decision that is easily grasped and applied. Subjects in a Libet-style experiment may be given the following instructions: One way to think of deciding to flex your right wrist now is as consciously saying now! to yourself silently in order to command yourself to flex at once. Consciously say now! silently to yourself whenever you feel like it and then immediately flex. Look at the clock and try to determine as closely as possible where the dot is when you say now! Youll report that location to us after you flex. (see [29], p. 10) Subjects can also be regularly reminded to make their decisions spontaneouslythat is, to make them without thinking in advance about when to flex. Experiment 2 is simply a Libet-style study with these instructions. RPs and muscle bursts are measured in the standard way. After flexing, subjects report B-times by moving the dot to the location on the clock where they believe the dot was when they consciously said now!9 If, as I predict, subjects given these instructions individually show much less variability in B-times than subjects given typical Libet-style instructions, we would have grounds for believing that their reports about when they consciously said now! involve less guesswork and, accordingly, additional grounds for skepticism about the reliability of B-times in typical studies. If such Btimes as have actually been gathered are unreliable

9 My notion of B-time is broad enough to apply here even if, in saying now!, subjects are not actually deciding to flex. Btime is the time the subject believes to be C-time when responding to the experimenter s question about the onset of the subjects consciousness of a proximal decision, proximal intention, proximal urge, etc. Now!-saying can fall under the etc.

Testing free will


indicators of C-times, little weight can be put on them in arguments about whether or not there is time enough to veto conscious proximal urges and the like; and the same is true of arguments about whether or not C-time is too late for conscious proximal intentions and the like to play a role in producing corresponding overt actions.10 I am not insisting that, in this experiment, decisions are made when subjects say now! (see n. 10). Even if subjects do not make decisions then, the finding that they individually show much less variability in B-times than subjects in typical studies would cast further doubt on the reliability of B-times in those studies. Greater variability in this context is suggestive of a greater role for guesswork. EMG signals can be recorded from speech muscles even in silent speech [36, 37]. Such recordings may be made in variants of Experiment 2. Subjects afterthe-fact reports are evidence about when it was that they consciously silently said now!; and EMG recordings from, for example, the larynx in an experiment of the kind at issue may be another form of evidence about this. It would be interesting to see how the results of the two different measures are related. A relatively simple experiment would leave overt action out. Subjects would be instructed to watch a Libet clock, to consciously and silently say now! to themselves whenever they feel like it, and to be prepared to report after the silent speech act on where the dot was on the clock when they said now! The times specified in the reports can be compared to the times of the EMG activity.

Experiment 3: Implementation Decisions Roy Baumeister writes: if there are any genuine phenomena associated with the concept of free will, they most likely involve conscious choice. Such a view has to contend with the now widespread belief that consciousness is a useless, feckless epiphenomenon, and that all behavior is guided by nonconscious processes [38], p. 76). If all behavior were produced only by nonconscious processes, and if conscious choices or intentions and their neural correlates were to play no role at all in producing any corresponding actions, free will would be in dire straits. Do any experiments yield evidence that the fact that an agent consciously made a decision (or choice) to do something or had a conscious intention to do something sometimes has a place in a causal explanation of a corresponding overt intentional action?11 Work on implementation intentions seems to yield such evidence [3942]. Implementation intentions, as Peter Gollwitzer conceives of them, are subordinate to goal intentions and specify the when, where, and how of responses leading to goal attainment [41], p. 494). They serve the purpose of promoting the attainment of the goal specified in the goal intention. In forming an implementation intention, the person commits himself or herself to respond to a certain situation in a certain manner.12 In one study of subjects who had reported strong goal intentions to perform a BSE [breast selfexamination] during the next month, 100% did so if they had been induced to form additional implementation intentions [41], p. 496). In a control group of people who also reported strong goal intentions to do this but were not induced to form implementation intentions, only 53% performed a BSE. Subjects in the former group were asked to state in writing where and when they would perform a BSE during the next month. The intentions they consciously expressed in writing are implementation intentions. If, in response to the request, these subjects actively formed relevant implementation intentions, they decided in advance on a place and time for a BSE.
11 12


Would subjects conscious, silent now!s actually express proximal decisions? Perhaps not. To see why, consider an imaginary experiment in which subjects are instructed to count consciously and silentlyfrom 1 to 3 and to flex just after they consciously say 3 to themselves. Presumably, these instructions would be no less effective at eliciting flexings than the now! instructions. In this experiment, the subjects are treating a conscious eventthe conscious 3-sayingas a go signal (when they say 3, they are not at all uncertain about what to do, and they make no decision then to flex). Possibly, in a study in which subjects are given the now! instructions, they would not actually make proximal decisions to flex but would instead consciously simulate deciding and use the conscious simulation event as a go signal. However, the possibility of simulation is not a special problem for Experiment 2. In Libets own studies, subjects may be treating a conscious experiencefor example, their initial consciousness of an urge to flexas a go signal (see [35], p. 352).

I treat the terms choice and decision as synonyms.

Although a proximal intention can specify the when, where, and how ([41], p. 494) of a response leading to the attainment of a goal one already has, the implementation intentions that concern Gollwitzer are distal intentions.


A.R. Mele

Another study featured the task of vigorous exercise for 20 minutes during the next week [41], p. 496). A motivational intervention that focused on increasing self-efficacy to exercise, the perceived severity of and vulnerability to coronary heart disease, and the expectation that exercise will reduce the risk of coronary heart disease raised compliance from 29% to only 39%. When this intervention was paired with the instruction to form relevant implementation intentions, the compliance rate rose to 91%. In a third study reviewed in Gollwitzer [41], drug addicts who showed symptoms of withdrawal were divided into two groups. One group was asked in the morning to form the goal intention to write a short curriculum vitae before 5:00 P.M. and to add implementation intentions that specified when and where they would write it (p. 496). The other subjects were asked to form the same goal intention but with irrelevant implementation intentions (i.e., they were asked to specify when they would eat lunch and where they would sit). Once again, the results are striking: although none of the participants in the second group completed the task, 80% of the subjects in the first group completed it. Many studies of this kind are reviewed in [41], and Gollwitzer and Paschal Sheeran report that findings from 94 independent tests showed that implementation intentions had a positive effect of medium-tolarge magnitudeon goal attainment [42], p. 69). Collectively, the results provide evidence that the presence of relevant distal implementation intentions significantly increases the probability that agents will execute associated distal goal intentions in a broad range of circumstances. In the experimental studies that Gollwitzer reviews, subjects are explicitly asked to form relevant implementation intentionsthat is to decide in advance on a place and time to execute the pertinent goal intentionand the intentions at issue are consciously expressed [41], p. 501).13 Apparently, the results Gollwitzer reports are bad news for any argument for the nonexistence of free will that depends on the premise that conscious

intentions (or their neural correlates) never play a role in producing corresponding overt actions.14 However, someone (a behaviorist, for example) might object that neither the making of conscious implementation decisions nor the having of conscious implementation intentions accounts for the impressive success rates of the implementationintention groups and that, instead, it is the sincere reporting of implementation decisions or intentions that accounts for this. It might be objected that the most that has been shown is that reporting both goal intentions and implementation decisions or intentions is more effective than reporting goal intentions alone. This contention is testable, and a proper test would yield evidence about the relative effectiveness of the sincere reporting of conscious implementation decisions or intentions, on the one hand, and the conscious decisions or intentions themselves, on the other. A testExperiment 3is easy to construct. The key, obviously, is to omit the instruction to report the implementation decision or intention. For example, in a new version of the BSE study, subjects who report strong goal intentions to perform a BSE next month may be instructed to consciously decide on a place and time next month to conduct the BSE before they leave the room and to make a mental note of their decision. They can be asked whether they have made such a decision and have made a mental note of it before they are permitted to leavewithout also being asked to report what they decided. I would not be surprised if it were discovered that the reporting has some effect. But then the issue is whether the effect is great enough to account for the difference between the two groups in the original study. My prediction is that the effect will be too small to do this.

Conclusion Elsewhere, I have argued against some inferences about free will made on the basis of various data,

14 13

It should not be assumed that all members of all of the control groups lack conscious implementation intentions. Indeed, for all anyone knows, most members of the control groups who executed their goal intentions consciously made relevant distal implementation decisions.

An epiphenomenalist about conscious intentions may claim that the fact that subjects had conscious implementation intentions plays no role in accounting for their behavior and, perhaps, that unconscious implementation intentions would have been just as effective. For various problems with this claim, see [10], chapter 7, section 3.

Testing free will

171 11. Frankfurt, H. 1988. The importance of what we care about. Cambridge: Cambridge University Press. 12. McCann, H. 1986. Intrinsic intentionality. Theory and Decision 20: 247273. 13. Searle, J. 2001. Rationality in action. Cambridge, MA: MIT Press. 14. AA. 2003. 15. AA. 1992. 16. Passingham, R., and H. Lau. 2006. Free choice and the human brain. In Does consciousness cause behavior? An investigation of the nature of volition, eds. S. Pockett, W. Banks, and S. GallagherCambridge, MA: MIT Press. 17. Libet, B. 1992. The neural time-factor in perception, volition and free will. Revue de Mtaphysique et de Morale 2: 25572. 18. Libet, B. 2001. Consciousness, free action and the brain. Journal of Consciousness Studies 8: 5965. 19. Libet, B. 1999. Do we have free will? Journal of Consciousness Studies 6: 4757. 20. Hallett, M. 2007. Volitional control of movement: The physiology of free will. Clinical Neurophysiology 118: 11791192. 21. Libet, B., E. Wright, and C. Gleason. 1982. Readiness potentials preceding unrestricted spontaneous vs. preplanned voluntary acts. Electroencephalography and Clinical Neurophysiology 54: 322335. 22. Libet, B., C. Gleason, E. Wright, and D. Pearl. 1983. Time of unconscious intention to act in relation to onset of cerebral activity (readiness-potential). Brain 106: 623642. 23. Libet, B., E. Wright, and A. Curtis. 1983. Preparation- or intention-to-act, in relation to pre-event potentials recorded at the vertex. Electroencephalography and Clinical Neurophysiology 56: 367372. 24. Dennett, D. 2003. Freedom evolves. New York: Viking. 25. Banks, W. 2006. Does consciousness cause misbehavior. In Does consciousness cause behavior? An investigation of the nature of volition, eds. S. Pockett, W. Banks, and S. Gallagher, 235256. Cambridge, MA.: MIT Press. 26. Logan, G. 1994. On the ability to inhibit thought and action: A users guide to the stop signal paradigm. In Inhibitory Processes in Attention, Memory, and Language, eds. E. Dagenbach, and T. Carr, 189239. San Diego: Academic. 27. Haggard, P., and B. Libet. 2001. Conscious intention and brain activity. Journal of Consciousness Studies 8: 47 63. 28. Brass, Marcel, and P. Haggard. 2007. To do or not to do: The neural signature of self-control. Journal of Neuroscience 27: 91419145. 29. AA. 2008a. 30. Ramachandran, V. 2004. A brief tour of human consciousness. New York: Pi Press. 31. van de Grind, W. 2002. Physical, neural, and mental timing. Consciousness and Cognition 11: 241264. 32. Haggard, P. 2006. Conscious intention and the sense of agency. In Disorders of volition, eds. N. Sebanz, and W. Prinz, 6985. Cambridge, Mass.: MIT Press. 33. Haggard, P. 2005. Conscious intention and motor cognition. Trends in Cognitive Sciences 9: 290295. 34. Lau, H., R. Rogers, and R. Passingham. 2007. Manipulating the experienced onset of intention after action execution. Journal of Cognitive Neuroscience 19: 8190.

including Libet s data [7], chapter 2, [8 10]. My primary aim here was to sketch a trio of experiments that would advance our understanding of the import of data that have already been gathered. Experiment 1 concerns how reliable a predictor of behavior a certain segment of type I and type II RPs is. The experiment would shed light on how to interpret a segment of the averaged EEG that proceeds muscle bursts in certain Libet-style studies and on when the point of no return is reached in such studies. Experiment 2 addresses another issue about reliability: Are B-times, as gathered in typical experiments, reliable indicators of C-times? If not, such Btimes as have been gathered should carry little weight in arguments about whether or not there is time enough to veto conscious proximal urges or intentions and about whether or not C-time is too late for conscious proximal intentions and the like to play a role in the production of corresponding overt actions. Finally, Experiment 3 would generate useful data about the effectiveness of conscious implementation decisions or intentions.

Acknowledgment A draft of article was completed during my tenure of 2007-08 NEH Fellowship. (Any views, findings, conclusions, or recommendations expressed in this article do not necessarily reflect those of the National Endowment for the Humanities.) I am grateful to Neil Levy and Adina Roskies for comments on that draft.

1. Libet, B. 1985. Unconscious cerebral initiative and the role of conscious will in voluntary action. Behavioral and Brain Sciences 8: 52966. 2. Libet, B. 2004. Mind time. Cambridge, MA: Harvard University Press. 3. Wegner, D. 2002. The illusion of conscious will. Cambridge, MA: MIT Press. 4. Wegner, D. 2004. Frequently asked questions about conscious will. Behavioral and Brain Sciences 27: 679 688. 5. Wegner, D. 2008. Self is magic. In Are we free? Psychology and free will, eds. J. Baer, J. Kaufman, and R. Baumeister, 226247. New York: Oxford University Press. 6. AA. 2004. 7. AA. 2006a. 8. AA. 2006b. 9. AA. 2008b. 10. AA. n.d.

172 35. Keller, I., and H. Heckhausen. 1990. Readiness potentials preceding spontaneous motor acts: Voluntary vs. involuntary control. Electroencephalography and Clinical Neurophysiology 76: 351361. 36. Cacioppo, J., and R. Petty. 1981. Electromyographic specificity during covert information processing. Psychophysiology 18: 518523. 37. Jorgensen, C., and K. Binsted. 2005. Web browser control using EMG based subvocal speech recognition. Proceedings of the 38th Hawaii International Conference on System Sciences, vol. 38, pp. 18. 38. Baumeister, R. 2008. Free will, consciousness, and cultural animals. In Are we free? Psychology and free will, eds. J. Baer, J. Kaufman, and R. Baumeister, 6585. New York: Oxford University Press.

A.R. Mele 39. Gollwitzer, P. 1993. Goal achievement: The role of intentions. European Review of Social Psychology 4: 14185. 40. Gollwitzer, P. 1996. The volitional benefits of planning. In The psychology of action, eds. P. Gollwitzer, and J. Bargh, 287312. New York: Guilford. 41. Gollwitzer, P. 1999. Implementation intentions. American Psychologist 54: 493503. 42. Gollwitzer, P., and P. Sheeran. 2006. Implementation intentions and goal achievement: A meta-analysis of effects and processes. Advances in Experimental Social Psychology 38: 69119. 43. Haggard, P., and E. Magno. 1999. Localising awareness of action with transcranial magnetic stimulation. Experimental Brain Research 127: 102107.