Does the brain “initiate” freely willed processes? A philosophy of science critique of Libet-type experiments and their interpretation

Abstract

In the extensive, recent debates on free will, the pioneering experiments by Benjamin Libet continue to play a significant role. It is often claimed that these experiments demonstrate the illusory nature of freely willed actions. In this article, we provide a detailed analysis and evaluation of Libet’s experiments from a philosophy of science perspective. Our analysis focuses on Libet’s central notion of the “initiation” of freely willed processes by the brain. We examine four interpretations of the notion of initiation: in terms of a cause, a necessary condition, a correlation, and a regular succession. We argue that none of these four interpretations can be supported by the design and results of Libet’s experiments. In addition, we analyze two recent Libet-type experiments. Our general conclusion is that neither Libet’s original experiments nor later Libet-type experiments can justify the claim that allegedly freely willed processes are in fact initiated by the brain.

The recent literature in neuroscience and philosophy of mind shows an increasing interest in the relation between specific brain processes and free will. The pioneering experiment by Benjamin Libet and co-workers (see Libet, Gleason,Wright, & Pearl, 1983) is still being discussed, related experiments have been performed or are being proposed, and strong conclusions concerning the existence of a free will are drawn on the basis of these experiments.

In this article, we scrutinize Libet’s central experiment and some more recent Libet-type experiments from a philosophy of science perspective. In particular, we examine several claims about the “initiation” of the conscious free will to act by an earlier brain process (the rise of a so-called “readiness potential”). Our analysis exploits insights from philosophical accounts of scientific experimentation and causality. The overall conclusion is that the claim that a conscious free will does not exist cannot be based on the results of Libet-type experiments.

In the second section, we describe the setup and results of Libet’s experiment in some detail. The third section constitutes the core of this article. We provide four interpretations of the claim that the readiness potential initiates the (allegedly ¹ ) conscious free will to act: “initiation by the readiness potential” as a cause, as a necessary condition, as a correlated event, and as a regularly preceding event. We assess these interpretations from a philosophy of science perspective and show that none of them can be supported by the actual results of the Libet experiment. In the fourth section, we analyze and evaluate the analogous relation between the readiness potential and the voluntary motor act along the same lines as in the third section. Our conclusion from the third and fourth sections is that the Libet experiment does not show that the readiness potential initiates either the conscious free will or the voluntary motor act itself. Two recent experiments, an already performed one by Soon and co-workers (Soon, Brass, Heinze, & Haynes, 2008) and a merely proposed one by Mele (2010), are discussed in the fifth section. We conclude that neither of these newer studies can confirm the claim that free will is illusory. The final section briefly puts forward some observations about what has been, and has not been, achieved in the research under discussion, and what may, and may not, be expected from further research.

The Libet experiment and its impact on the debate on free will

The basis of Libet’s experiment was provided by a finding by Kornhuber and Deecke (1965). They observed that voluntary motor acts were preceded by a characteristic type of negative electrical signal, which became known as the readiness potential (RP). This electrical signal could be detected by an electroencephalogram (EEG) at the vertex of the skull. A remarkable discovery was that the RP is already recordable at about 800 ms before the actual motor action occurs. This long period of time suggested to Libet that there might be a “discrepancy” between the moment at which the RP starts and the moment at which a conscious will to perform the voluntary action arises. Therefore, according to Libet (2004),

[E]stablishing the time of the conscious will relative to the onset of brain activity (the RP) was clearly important. If conscious will were to follow the onset of RP, that would have a fundamental impact on how we view free will. (p. 125)

To determine the moment at which a person becomes aware of the “voluntary urge to act” relative to the RP, Libet and his co-workers designed an ingenious experiment. ² In this experiment, the participants were asked to make a specific voluntary movement: flexing the fingers or wrist of their right arm. The actual experiment consisted of three time measurements: first, measuring the start of the RP; second, establishing the moment at which the participants become aware of the urge (or intention or will) to flex their wrist or fingers; third, determining the moment of the contraction of the muscles leading to the flexing of the wrist or fingers. Next, these three measurements were compared in order to establish the temporal order of the events. Of these measurements, the second constituted the new and the most challenging part of the experiment. After conceiving the experiment in general terms, Libet was puzzled about how a concrete measurement of the time of a person’s awareness could be performed. The problem is that the experience of an urge is not detectable from the outside, and that any attempt to communicate such an experience would itself generate another RP, thus disturbing the plan of the experiment. As Libet (2004) recollects, at some point he realized that the solution to the “apparently intractable problem” (pp. 125–126) was that the time at which the subject becomes aware of the urge or intention could—somehow—be “noted silently,” and then reported after the flexing of the wrist or fingers. In this way, an RP accompanying the report of the awareness, which would interfere with the RP of the movement, does not occur.

The eventual design was as follows (Libet, Gleason, et al., 1983, p. 625). Participants were asked to sit down in a chair and to flex their wrist or fingers at a freely chosen moment in time, while the experimenters were performing the three time measurements. At the beginning of each trial, the participants were to relax and fix their gaze on the center of a screen of a cathode ray oscilloscope positioned in front of them. The oscilloscope produced a spot of light that revolved in a clockwise circle, completing each revolution in 2.56 seconds. The “clock position” of the revolving spot of light was to be observed by the participant for later recall and report (after the flexing of the fingers or wrist had occurred). The participants were instructed to report the time of the earliest occurrence of the awareness (W) of wanting to perform a given self-initiated movement, by recalling the spatial image of the clock at the relevant moment. They were also instructed to let the urge to flex appear on its own, without any preplanning.

The results of the experiment were as follows. If participants preplanned their motor act (although they were instructed not to do this), the RP appeared already about 1000 ms before the start of muscular activity. If the participants did not preplan the motor act, the RP appeared about 500 ms before the muscle contraction. The participants reported to have become aware of their urge or intention to act about 200 ms before the contraction of the muscles, but well (300–800 ms) after the onset of the RP.

Although the experiment concerned only one, rather elementary, type of voluntary motor action, Libet, Gleason, et al. (1983, pp. 640–641) consider their results applicable to a much wider class of human acts. The studied motor act, they claim, is often seen as an ideal example of a freely willed action. They also point to the fact that the absence of any larger meaning in the prescribed movement and the possibility to execute this movement “with capriciously whimsical timings” make it a good candidate for extrapolation to other spontaneous voluntary acts.

Notably, in his own interpretation of the experiment, Libet (see in particular 1999) puts much weight on a somewhat specific observation: Sometimes the research participants reported that an urge to act appeared, but that they suppressed or “vetoed” this urge. Because no muscle contraction followed in these cases, there was no trigger to initiate the computer storage of an EEG (and RP) signal (the experiment was designed in such a way that the contraction was needed to store the results of the EEG measurements). Therefore, no RP recordings were made of trials in which a participant reported to have vetoed the urge to act. However, Libet, Wright, and Gleason (1983) claimed to have shown in another experiment that if participants did veto their intention to act, a (large) RP was still recordable. Although Libet (1999) himself states that “the existence of a veto possibility is not in doubt” (p. 51), such a phenomenon has been deemed highly questionable by many commentators (e.g., Mele, 2008; Nelson, 1985; Spence, 1996). However that may be, in our analysis and assessment of the Libet experiment and its interpretation, we will leave aside this veto part.

The Libet experiment has had, and is still having, a major impact on neuroscientific and philosophical debates on free will, volition, and the role of consciousness in human behavior. The experiment has led to further experimentation by neuroscientists (e.g., Haggard & Eimer, 1999; Soon et al., 2008), and it has stimulated extensive debate among philosophers (e.g., Mele, 2010) and psychologists (e.g., Wegner, 2002), thus bridging the different disciplines concerned with questions about volition, free will, and related matters.

Quite a few commentators have drawn the strong conclusion from the Libet experiment that the idea of a free will is an illusion. A representative author is Spence (1996), who states that “[i]f the findings of Libet and co-workers are replicated, then we must conclude that ‘decisions’ to act arise prior to our conscious awareness of them. Thus our ‘decision’ or ‘freedom’ is illusory (if by these terms we mean conscious phenomena)” (pp. 83–84).

But is this conclusion justified? Is it really true that Libet’s experiment (if replicated) shows that “freedom is illusory”? In the ensuing sections, we will argue that this is not at all the case. Thus, the target of this article are not the Libet-type experiments per se, but rather those interpretations that infer the nonexistence of free will from these experiments.

Why the readiness potential does not “initiate” the conscious free will

As we have seen, the three main events of the experimental process are: (a) the onset of the readiness potential at time t₁ (RP-t₁ ), (b) the earliest awareness of the conscious free will to act at t₂ (W-t₂ ), ³ and (c) the contraction of the muscles that start the actual movement of the fingers or wrist at time t₃ (M-t₃ ). A central claim of the experiment is that these three events succeed each other, hence t₁ < t₂ < t₃ . It is important to note that these experimental events and their temporal order are by no means “directly observed.” As is usual in experimental science (Radder, 2003, pp. 161–172), the claim that these events have successively occurred in the course of the experimental process depends on substantial theoretical interpretation, especially in the case of RP-t₁ and W-t₂ . Part of the scientific debate, then, focuses on the plausibility of the implied interpretations. For example, quite a few of the commentators on Libet (1985) question the reliability and validity of the experimental procedures on which the reported events and their times of occurrence are based. This applies in particular, though not exclusively, to W-t₂ . In this article, however, the focus is on the precise nature of the relation between the three types of events. Thus, we leave aside all kinds of other problems that may be, and have been, raised about the experiment. ⁴

The crucial claim of the experiment is that RP-t₁ precedes W-t₂ by a substantial amount of time. The main conclusion based on this claim is phrased by Libet as follows.

What we found, in short, was that the brain exhibited an initiating process [i.e., RP-t₁ ] beginning 550 msec before the freely voluntary act [i.e., M-t₃ ]; but awareness of the conscious will to perform the act [i.e., W-t₂ ] appeared only 150–200 msec before the act. The voluntary process is therefore initiated unconsciously, some 400 msec before the subject becomes aware of her will or intention to perform the act. (Libet, 2004, pp. 123–124) ⁵

Moreover, quite a few commentators have drawn the “big” conclusion that, if this finding can be generalized to all human acts, free will does not exist because our allegedly conscious free will to act is no more than an epiphenomenon, the result of a purely physical process that starts with a much earlier readiness potential and terminates in the relevant bodily movement. It is, of course, this big conclusion that has fueled the strong interest in, and the extended discussion of, the Libet experiment.

In Libet’s publications, the relations between RP-t₁ and W-t₂ and between RP-t₁ and M-t₃ are almost invariably specified by the notion of “initiation.” This notion features, for instance, in the subtitle of Libet, Gleason, et al. (1983), while the title of Libet (1985) includes the closely related term “initiative.” The principal aim of our article is to question the appropriateness of this notion and of the conclusions based on it.

We start by considering the claim that, on average, RP-t₁ precedes W-t₂ and M-t₃ by the times reported by Libet, and we ask ourselves which conclusions this fact entails. More precisely, our two central questions are: (a) what can be meant by the notion of initiation and (b) can the use of this notion be justified by the experiment? We discuss four possible answers to these questions, which interpret the “initiating” event, successively, as a cause, as a necessary condition, as a correlated event, and as a regularly preceding event. Thus, our discussion proceeds from a rather strong to a quite weak interpretation and examines whether these interpretations make sense given the actual experimental results. In this way, we analyze and evaluate the relation between RP-t₁ and W-t₂ (in this section) and between RP-t₁ and M-t₃ (in the next one).

As a first interpretation, consider the claim that “RP-t₁ is the cause of W-t₂ .” ⁶ This claim implies that, for each research participant, an actual causal relation obtains between the two particular events. Are this claim and its implication supported by the experimental results? ⁷ To answer this question we will discuss it on the basis of two distinct, well-developed theories of causality: the interpretation of causation in terms of so-called “INUS conditions” and the manipulability interpretation of causal inference. What is important for our purposes is that both theories can be easily related to causal reasoning in actual scientific practices, in particular in experimental or observational contexts. Furthermore, both take up a prominent position in philosophical debates on causality (cf. Psillos, 2002). Although it is also the case that both theories have been criticized, we will see below that the major points of criticism do not apply to the more limited uses we make of the theories in this article.

Let us first look at the account of causation in terms of INUS conditions (Mackie, 1974, chap. 3). The INUS account is a refinement of theories that analyze causation in terms of single sufficient and necessary conditions between the cause and its effect (e.g., the theory of J. S. Mill). Philosophers of science realized that, in most cases, an appropriate account of real-world phenomena requires the more sophisticated INUS theory. This certainly applies to the mind/brain phenomena we are addressing in this article.

Suppose that C (a putative causal factor) and E (its putative effect) are successive, actual events and that X is a set of actual events, such that the conjunction of C and X (C ∧ X) is a sufficient condition for E. That is to say, in general it will be the case that C alone is not sufficient for E (because of the complexity of the circumstances). Furthermore, if C is to be a causal factor, it should of course not be a redundant part of C ∧ X. Finally, C ∧ X need not be necessary for E (because E may have other sufficient conditions). In a standard example, C is a short-circuit in an electricity network in a particular house, E the burning down of the house, while X may include the failure of the fuses and the proximity of inflammable materials. In this case, the short-circuit, which should not be redundant, is sufficient for the burning down only in combination with the further conditions X (failure of the fuses etc.). In addition, many other conditions may lead to such a fire, for instance smoking in bed and falling asleep. For this reason, C ∧ X is not a necessary condition for E, because the alternative set of events C′ ∧ X′, which includes smoking in bed (C′) and falling asleep (X′), may have started the fire. More generally, events C_i and E instantiate a causal relation if and only if (given appropriate background circumstances) C_i is an INUS condition for E. That is to say, C_i is an Insufficient, but Non-redundant part of an Unnecessary but Sufficient condition (namely C_i ∧ X_i ) for the occurrence of event E. We will call the INUS condition C_i a causal factor for E and the minimally sufficient condition C_i ∧ X_i a cause of E. Finally, according to the INUS theory, the full cause of E is the disjunction of all possible causes of the type C_i ∧ X_i .

In applying the INUS theory to the Libet experiment, we use it as an epistemic criterion. More precisely, we merely use it as providing a plausible necessary condition for the claimed existence of a causal factor. What we do not suppose is that being an INUS condition is sufficient for inferring a causal relation. For this reason, a substantial criticism of the INUS theory (that being an INUS condition is not always sufficient for the existence of a causal relation ⁸ ) is irrelevant to our analysis. More generally, the worries about the correctness of the INUS theory as a complete metaphysical account of causation (Psillos, 2002, chaps. 2, 3, and 4) need not detain us either.

As we have seen, the idea of the theory is that INUS conditions cover the most complex cases, although simpler conditions may obtain in less complex circumstances. In our case, we are interested in one particular set of conditions (namely the set that includes the onset of the RP) and not in other possibly sufficient conditions C_i ∧ X_i of the conscious free will to act. Hence, we do not have to address the full cause of W-t₂. In other words, we do not need to investigate the U-component of the INUS approach.

The other components (I, N, and S) are relevant, though. First, in discussions of the Libet experiment, it is often acknowledged (see n. 5) that, in addition to RP-t₁ , other brain events (say, the set X_br ) need to be taken into account. Thus, the question is whether the experiment shows that, in the experimental circumstances, RP-t₁ ∧ X_br is sufficient for W-t₂ . Here we have to be careful, because the crucial aspect of W-t₂ is not the occurrence of the conscious free will to act as such but the specific time at which this will arises. After all, as several commentators have rightly remarked (e.g., Brass & Haggard, 2008; Näätänen, 1985), it is not the nature of the will to act itself (which was prescribed by the instructions of the experimenters and practiced in the preliminary runs of the experiment), but the timing of this will which constitutes the element of freedom in the Libet experiment. ⁹ In fact, however, the experiment has produced no data at all to support the claim that the occurrence of a specific set of events RP-t₁ ∧ X_br is sufficient for inferring the specific timing t₂ of W-t₂ . Put differently, the experimental results do not explain why, in a particular trial, the conscious free will occurred at specific time t ₂ , rather than somewhat earlier or later (see also n. 12, and the corresponding discussion in the main text, below). Furthermore, since the possible role of X_br has not been concretely investigated, Libet’s experimental design has not provided any evidence for or against the claim that X_br alone is sufficient for W-t₂ . For this reason, we do not know whether RP-t₁ is non-redundant.

In sum, the Libet experiment allows for the insufficiency of RP-t₁ , at least in principle, by referring to other brain states X_br (the I-component of the INUS criterion); it is compatible with the possibility of sufficient conditions other than RP-t₁ ∧ X_br (the U-component); but it does not provide any evidence for the crucial claim that RP-t₁ ∧ X_br is sufficient for W-t₂ (the S-component); and it does not show that RP-t₁ is a non-redundant element of RP-t₁ ∧ X_br (the N-component). Hence, if being an INUS condition is necessary for inferring a causal relation (which is a plausible assumption), the Libet experiment does not allow us to infer that RP-t₁ is a causal factor for W-t₂ or that RP-t₁ ∧ X_br is a cause of W-t₂ .

But perhaps, one might object, the problem derives from the INUS theory rather than from the Libet experiment: perhaps the INUS theory is not a suitable theory for examining the causal relations in the Libet experiment. To counter this objection, we will also analyze the experiment with the help of a quite different theory, namely the manipulability theory of causation. ¹⁰ The basic idea of this theory is that a cause is a means for controlling its effect through a suitable intervention. As Woodward (2003a) explains, “if C causes E, then there is some way of intervening on C … that in some appropriate background circumstances will change E, and if there is such an intervention, then C causes E” (p. 92). ¹¹ An important additional condition is that the changes in E should occur exclusively in virtue of E’s relationship to C, and not through some other causal route to E. Suppose, for example, that we can set on fire some inflammable material through actively producing a short-circuit in an electricity network and suppose that the setting on fire is not due to some other factor independent of the short-circuit. In this case, the short-circuit is the cause of the fire.

As in the case of the INUS theory, we use the manipulability theory for our relatively limited purposes. That is to say, we do not use it as an explication of the meaning of causality, but merely as a criterion for evaluating causal inferences concerning the results of the Libet experiment. Therefore, we need not be concerned by the problem that the explication of the meaning of causality by the manipulability theory is circular (Lange, 2003, pp. 124–128). Furthermore, the additional condition entails that the criterion provided by the theory is fallible. To make the claim that C causes E with certainty, we have to be sure that any other, known or unknown, account of the relevant change of E can be excluded. Since, in practice, such certainty cannot be achieved, positive causal claims will remain fallible in principle, even if they may be reliable enough for all practical purposes. Because we merely employ the intervention condition of the theory and argue for a negative claim only (to wit, that there is no causal connection in the case under discussion), we do not need to address this issue any further.

In the Libet experiment, the claim that RP-t₁ is the cause of W-t₂ requires that specific changes in the timing of the conscious free will to act can be brought about by certain interventions that change the readiness potential. In the manipulability theory, such interventions are not limited to changes brought about by human experimenters (Woodward, 2003a, p. 94). In principle, their occurrence may also be the result of some natural process, what Woodward calls a “natural experiment.” In Libet’s experiment, however, such interventions (be they human or natural) have not taken place. Moreover, according to some authors, the relevant human interventions (which would effect a change of the timing of W by modifying RP) cannot be performed. As Haggard concludes, “the problem of describing and classifying the brain processes of action initiation” is that “we cannot intervene on the processes of action initiation and see how RPs and conscious experience are altered in consequence” (Haggard & Libet, 2001, p. 62). Such an intervention or manipulation, however, is exactly what is called for in order to satisfy Woodward’s manipulation requirement. Furthermore, in the Libet experiment other factors that might account for changes in the timings of W have not been considered either. Hence, we have to conclude once more that Libet’s experimental results cannot be used to validate the claim that RP-t₁ is the cause of W-t₂ .

Remarkably enough, this conclusion can be supported by a statement of Libet himself. From a somewhat different replication of Libet’s experiment, Haggard and Eimer (1999) conclude that their finding “rules out the RP as the unconscious cause of the conscious state upon which W judgement depends” (p. 132), because their RPs do not co-vary with their early versus late Ws. ¹² Libet, however, responds by questioning the similarity of the Haggard and Eimer replication to his own experiment, since their measurements pertain to a different kind of readiness potential. For this reason, he concludes that, concerning his own experiment, “no conclusion about causation on these grounds can be presently drawn” (Libet, 2004, p. 135, our emphasis; see also Haggard & Libet, 2001, pp. 58–60). This quotation also shows that in these publications from 2001 and 2004 Libet takes the interpretation of “initiation” in terms of “cause” to be a natural one. In spite of this, the result of our analysis so far is that the claim that RP-t₁ causes W-t₂ is implausible, since it fails to meet both the INUS condition and the manipulability criterion.

A second (and considerably weaker) interpretation of the claim that RP-t₁ initiates W-t₂ is that the former constitutes a necessary condition for the latter. In this case, W-t₂ cannot come about without RP-t₁ , but the occurrence of RP-t₁ does not tell us anything about the occurrence of W-t₂ . Unfortunately, this second interpretation cannot be supported by the experimental results either. After all, a regular succession of the onset of a readiness potential and the earliest awareness of the conscious free will to act is certainly not enough to substantiate a necessary condition between the two (types of) events. ¹³ Establishing the necessity of this regularity would require additional information about the mechanism of the connection between RP-t₁ and W-t₂ , information that the experiment does not provide. Consider our earlier example of a house on fire (owing to a short-circuit), and the claim that the presence of oxygen is a necessary condition for the burning down of the house. In contrast to the Libet experiment, in this case there are reasons (derived from our experimental and theoretical knowledge of the chemistry of combustion) which justify us to go beyond a regularity claim to a necessity claim. Although there are, in the Libet case, some further data concerning the connection between RP and W (e.g., Sirigu et al., 2004), as yet these data are not sufficient to infer a robust necessity relation.

Furthermore, even if RP-t₁ could be shown to be necessary for W-t₂ , other problems for this interpretation of the notion of initiation would remain. For a start, in general there may be a large number of other factors that are necessary for W-t₂ . The presence of oxygen in the blood of the participant’s brain is just one of those. This fact gives rise to a difficult question: Why attach such a great significance to RP-t₁ instead of, for instance, the presence of oxygen in the blood or one of the many other necessary conditions? Without adding further knowledge of the relevant brain processes, this question seems unanswerable.

In spite of this, suppose that we could single out one particular necessary condition as uniquely salient. Even so, it would not be enough to confirm the claim that RP-t₁ initiates W-t₂ , rather than, say, W-t₂ ′ (willing to act at the different time t₂ ′) or W′-t₂ (willing the different act W′ at time t₂ ). By way of comparison, consider this analogy: Switch on your computer, select the digital version of this article, and go to http://www.youtube.com/music; then choose to play, at some freely willed subsequent time, a specific piece of music by a mouse-click on the YouTube site. In this experiment, the switching-on and surfing to YouTube will surely generate some kind of electrical “readiness potential” in the computer, while the act of playing a specific piece of music can be realized with the mouse through a “spontaneous flexion of a finger or wrist.” Obviously, this computer-RP is a necessary condition for playing any piece of music (available at YouTube) at any subsequent time. It will be equally obvious, however, that the computer-RP in no way determines the specific time of your conscious free will to play a particular piece of music: It could take seconds, days, or it might never happen. A natural interpretation of this experiment is that the computer-RP enables us to play a variety of pieces of music at a variety of times. Similarly, if the readiness potential could be interpreted as a necessary condition, a natural interpretation would be to see it as enabling, rather than initiating, a variety of freely willed actions at a variety of freely willed times. ¹⁴

Given the failure of the first two interpretations (in terms of a cause and in terms of a necessary condition), let us try a third (and even weaker) one: The initiation by the readiness potential means, or implies, that RP-t₁ is (fully or strongly) correlated to W-t₂ . This interpretation is advocated by Merikle and Cheesman (1985), who state that “in Libet’s experiments, RPs, conscious awareness, and voluntary action are perfectly correlated” (p. 548). Again, however, the trouble with this claim is that it cannot be inferred from the experimental evidence. There are two weighty problems.

The first is that Libet’s RP values do not refer to individual events. Instead, they have been calculated, for each particular participant, by “averaging” the results of a series of 40 different trials. In fact, even today it proves to be impossible to measure the single trial RP values for each participant, since “[e]ach RP is very small and virtually buried among the other electrical activities of the resting brain. Therefore, many such acts had to be performed to produce a computer-averaged tracing that summated the small RPs” (Libet, 2004, p. 124; see also Kornhuber & Deecke, 1965, pp. 4, 12–13). The problem is that the amplitude of the noise (the “other electrical activities”) is roughly 10 to 100 times larger than the sought-for RP signal. However, since the noise is both positive and negative, it approximately cancels out when averaged over a large number of trials. Still, because the single trial values of the RPs are unknown, it is obviously impossible to calculate a correlation coefficient between the individual events RP-t₁ and W-t₂ of a particular participant. An important consequence is that a prediction of single Ws on the basis of individual RPs is impossible as well. For reasons of presentation, we discuss this problem in the context of the correlation interpretation. But of course, as we already anticipated in our earlier discussion, it also constitutes a crucial problem for the causal and necessary condition interpretations.

The second problem with this interpretation is more straightforward. It derives from a general feature of all correlations. Even if we could establish a strong correlation between individual RPs and Ws, this fact alone would not be enough for drawing the relevant conclusions. After all, the singing of early birds (who start before sunrise) is strongly correlated with, but definitely does not initiate, the rising of the sun. A similar problem plagues the leap from a (potential) correlation between RP-t₁ and W-t₂ to the claim that RP-t₁ initiates W-t₂ . More generally, the correlation relation is symmetric (if event A is correlated with event B, B is also correlated with A), but the notion of initiation is clearly asymmetric (if A initiates a process that includes B, B cannot simultaneously initiate A). This evident fact shows that initiation cannot be explained on the basis of correlation alone. Of course, if we could add the claim that A always precedes B, we would have a temporally asymmetric relation. However, as we will argue now, in the Libet experiment we cannot even make such a minimal claim.

The fourth and final interpretation brings us back to the bottom line: “initiation in the sense of temporal precedence.” However, we have just seen that the Libet experiment does not deliver the quantitative shape of single-trial RPs. What is more, the results of the so-called “averaging” process are, both logically and empirically, fully compatible with the fact that in several of the 40 trials no RP-related signal has been present at all. In the case of voluntary motor acts, there definitely is a residual signal, but this signal does not represent an average RP in the usual sense of “average,” in which one first separately measures each single RP value and then calculates the average value. Instead, the final RP signal is the cumulative effect of the 40 trials, divided by 40. ¹⁵ An important consequence of this “averaging” technique—which seems to have gone unnoticed thus far in discussions of the Libet experiment—is that we do not know whether every single trial has contributed a certain portion to this cumulative RP. That is to say, it is not only that the single-trial values of the RPs are unknown, but Libet’s experiment does not even allow us to conclude that there is some non-zero RP value in each single trial. By way of comparison, suppose that in a certain room there are 40 philosophers and 120 children, each of whom has one of these philosophers as a parent. In this case, it would be equally mistaken to conclude that each of the 40 philosophers has one or more children. Similarly, we do not know whether an RP always precedes the conscious free will to act in each separate trial. ¹⁶

In spite of this, suppose that an RP of some (unknown) shape does occur in each separate trial and that this particular readiness potential always precedes a particular conscious free will to act. If this were the case, the claim that RP-t₁ always precedes W-t₂ would be justified. It will be obvious, however, that this does not constitute an acceptable interpretation of the notion of initiation. After all, the mere fact that an event B (e.g., W-t₂ ) is regularly preceded by another event A (e.g., RP-t₁ ) does not tell us anything about the ontological status of event B. No definite conclusions about the existence or nonexistence of a free will can be drawn from this fact. Furthermore, event W-t₂ is probably regularly preceded by a countless number of other events, both inside and outside the brain. The mere notion of precedence cannot explain why the event RP-t₁ should be more significant for W-t₂ than anyone of the other regularly preceding events.

In sum, if Libet’s experiment is to make a substantial contribution to the debate on free will, the notion of initiation cannot merely mean “precedence.” An initiating event must somehow “mark” the ensuing process in a specific way. As we have shown in this section, however, stronger interpretations of the notion of initiation fail to be supported by the results of Libet’s experiment. Hence, this experiment does not license any conclusions about a possible “initiation” of our conscious free will to act by unconscious physical brain processes.

Why the readiness potential does not “initiate” the motor process either

The second important relation is the one between the readiness potential RP-t₁ and the action M-t₃ , the start of the actual muscular movement. Separated from the context of the Libet experiment, this relation would merely specify a connection between a brain and a motor process, and as such our critical assessment below would only apply to the neuroscientific claims concerning this connection. However, in the context of the Libet experiment, this relation is also relevant to the debate on free will. After all, in this context it is used to argue that the conscious free will does not initiate the motor process because this process has already been initiated by the earlier readiness potential.

Remarkably enough, the precise nature of the relation between RP and M is equally unclear. Thus, Banks (2006) states that

it is surprising that, with so much interest in the findings of Libet et al. in both scientific and philosophical quarters, the lack of evidence on the necessity and the sufficiency of unconscious activity measured by the RP has not been seen as an issue. In the absence of the needed research, I will go along with the crowd and act as though the RP were necessary and sufficient for the action. (p. 240)

From what we have said thus far, it will be clear that we will not be “going along with the crowd.” Hence, in this section we address the question of whether the event RP-t₁ can be said to “initiate” the process that results in the motor action M-t₃ . Moreover, we will do so in complete analogy to our discussion in the previous section. A fortunate consequence of this analogy is that our present discussion can be quite brief, since our earlier arguments can be carried over almost completely. The only difference has to do with the fact that the critical event in the previous section was not the mere occurrence of W-t₂ but the time of its occurrence. In the present context, however, it is the occurrence of M-t₃ that is the relevant event.

Considering the interpretation of “initiation” as a cause, the latter point requires a slight change of our argument in terms of INUS conditions. Again, the Libet experiment allows for the insufficiency of RP-t₁ (for M-t₃ ) through the inclusion of other brain states X′_br (the I-component of the INUS criterion); it is similarly compatible with the possibility of sufficient conditions other than RP-t₁ ∧ X′_br (the U-component); again, however, the experiment does not provide any evidence for the claim that RP-t₁ is a non-redundant element of RP-t₁ ∧ X′_br (the N-component). In contrast, at first sight the claim that RP-t₁ ∧ X′_br is sufficient for M-t₃ (the S-component) does seem to be plausible (at least within the context of our first interpretation). In fact, however, we do not know whether this claim is true, as is shown by an argument of Mele, which we will discuss in the next section. Since Mele convincingly shows that Libet’s experiment does not imply that RP-t₁ ∧ X′_br is sufficient for M-t₃ , we may conclude that RP-t₁ cannot be interpreted as a causal factor for M-t₃ and RP-t₁ ∧ X′_br not as a cause of M-t₃ . Furthermore, the interpretation of “initiation” by means of the manipulability theory of causal inference requires the presence or absence of an intervention on RP-t₁ , and hence our argument concerning the relation between RP-t₁ and W-t₂ applies just as well to the relation between RP-t₁ and M-t₃ .

Finally, our earlier arguments concerning the other three interpretations are equally straightforward in the present case. As Banks already noted in the above quotation, there is no evidence that RP-t₁ is necessary for M-t₃ . Furthermore, exactly the same problems that plague the correlation and the precedence interpretation in the case of W-t₂ apply to the case of M-t₃ . In sum, Libet’s experiment supports neither the claim that the readiness potential initiates the conscious free will nor the claim that it initiates the process that results in the actual motor event.

Why more recent experiments do not solve the problems

Given this conclusion, one might ask whether more recent experiments are able to solve, or avoid, these problems. We will address this question in the present section by discussing two “Libet-type” experiments. Such experiments are characterized by measurements of W and M that are identical or similar to the ones performed by Libet himself; they may include RP-measurements, but they may also be based on measurements of other brain states. In assessing these experiments, we will both draw on our analyses in the previous sections and add some additional arguments.

The first experiment, by Soon and colleagues (2008), presents the results of an fMRI (functional magnetic resonance imaging) study. The experiment involves a new design that seems to avoid several shortcomings of Libet’s experimental setup. First, the authors point out that the RP is generated by the SMA (the supplementary motor area) and hence it merely provides information about the later stages of planning an action. In contrast, they claim that, on the basis of the brain activity of the prefrontal and parietal cortex, a specific decision can be predicted up to 10 seconds before it enters awareness (Soon et al., 2008, p. 543). ¹⁷ Thus, the moment at which a free decision is claimed to be encoded in the brain is much earlier than in Libet’s experiment. Second, by basing their results on different brain measurements, this study avoids the problems of the measurement of single RP-values discussed in the preceding sections. ¹⁸ Third, the experiment involves free decisions about more than one option, namely a choice between pressing one of two buttons. Therefore, the measured brain activity does not merely reflect some unspecific preparatory activation, and hence the experiment tries to avoid the interpretation of the relevant brain activity as merely a necessary condition. Fourth, the authors claim that their experiment enables the prediction of which option will be chosen, which suggests that they have discovered a robust correlation between brain activity and voluntary motor acts. More precisely, it proved to be possible to predict the decision made in about 60% of the cases, instead of the normal 50% chance level.

In spite of these novel features, this study does not allow the conclusion that the conscious free will is an illusion. The reason is that it suffers from two difficult problems of its own. First, a significant aspect of Libet’s experiment is that, at least in its usual interpretation (which we, however, challenged), each action is invariably preceded by an RP. The Soon et al. (2008) experiment, in contrast, merely concludes that there is a (slightly) increased chance that a certain motor action will be performed. This probabilistic result is clearly not sufficient for making a causal claim in the sense of the INUS theory. Yet, it might call for some sort of probabilistic causality. In this case, we should note that a correct prediction of at most 10% above chance level is not really a strong result. More important, however, is that it is unclear how to interpret this result. To put it bluntly: Is each participant free for 40% of his or her actions and not free for the other 60%?

A further drawback concerns the claim that “free” decisions are neurologically determined even 10 seconds before we become aware of them. The problem is that, in this case, something like a veto argument becomes extremely plausible: It is very hard to imagine that it would not be possible for a participant to decide, during the almost 10 seconds after the allegedly initiating brain processes have taken place, not to perform any action. Moreover, this claim is testable if we can design a new experiment that includes the additional option of refraining from action during the relevant time interval.

A second, recent Libet-type experiment is proposed by Mele. It aims to tell us something more about a causal relationship between RP activity and the motor action. Mele (2010, p. 164) points out that, in fact, Libet’s experiment does not permit the conclusion that the RP is a sufficient condition for the occurrence of the voluntary motor acts. As we noted in our second section, in Libet’s experimental design it is the muscle contraction that triggers the computer to store the previously measured brain activity. Without muscle contraction there is no storing, and without storing there is no (measured) contribution to the RP. This implies that it may well be that there were many moments during the experiment when there was RP activity but no muscle contraction. If this were the case, the RP-related brain activity would not be sufficient for the action. The point is that at present we simply do not know whether or not this is the case, and yet this knowledge is essential to justifying claims about causality.

To remedy the problem Mele (2010, p. 165) proposes an experiment to test for any RP activity that is not followed by a muscle contraction. If the latter would happen at a sufficient rate, the brain activity would not be a sufficient condition (and hence not a cause) of the motor action. ¹⁹ Since this experiment requires recording an RP for each single action, it is still unfeasible at present.

As we did in the previous sections, Mele explains that certain statements concerning causality and predictability are not supported by Libet’s experiment and he claims that, if feasible, his experiment will be able to solve this problem. However, this claim is questionable, for two reasons. Firstly, Mele points to the lack of knowledge of the presence of single-trial RPs in the specific case of the absence of muscular action, and he suggests a procedure to overcome this problem. This procedure, however, does not solve the more general problem concerning the presence of single-trial RPs (also in the case of motor activity), which results from the specific nature of the “averaging” technique. Secondly, Mele interprets causation as simply a sufficient condition. However, if one advocates an analysis of causality in terms of conditions (instead of the manipulability or some other account), the complexities of brain processes and motor control require the more comprehensive INUS approach. The reason is that this approach, derived from philosophy of science and experimentation, takes into account three relevant additional questions: (a) whether there might be other brain conditions X such that the RP is only sufficient for M in conjunction with X, (b) whether these conditions X as such might be sufficient to initiate M (in which case the RP would be redundant), and (c) whether there might be other conditions apart from RP and X (conditions somehow related to free will) which could also be sufficient for M. Hence, a plausible causal analysis of Libet-type experiments cannot be limited to the sufficiency issue of the RPs, but has to take into account the full complexity of the situation, as the INUS approach does. ²⁰

The conclusion of this section is that neither the already performed experiment by Soon and his co-authors nor the proposed experiment by Mele can be employed to confirm the claim that free will is illusory.

Concluding observations

Our detailed methodological analysis has demonstrated that the claim that the existence of a conscious free will is illusory finds no support in the results of Libet’s experiment. This illustrates the relevance of philosophy of science regarding neuroscientific methodology and philosophy of mind. In this respect, it is instructive to compare the current debate on neuroscience and free will with that on the issue of nonlocality in quantum mechanics. The latter debate also pertains to a fundamental ontological question: Is the (quantum) world essentially nonlocal, which would imply the occurrence of a specific kind of action at a distance, or do the interactions between (quantum) objects always require some form of local contact? In view of the fundamental nature of this question, philosophers of physics have meticulously scrutinized any premises and loopholes of the arguments for a nonlocal interaction, and experimental physicists have performed several dozens of ingenious test experiments (Shimony, 2007). Compared to this, we cannot but conclude that the debate on the significance of Libet-type experiments for the phenomenon of free will is still in its early stages.

Yet, even if further experiments could be made more robust (having greater reliability and validity), the experimental demonstration of the non-existence of free will can be expected to remain controversial. The reason is that, in the case of experiments aimed at examining essentially contested concepts, such as free will, the phenomenon of the “experimenters’ regress” (Collins, 1985, chaps. 4 and 5), especially in its basic form of the “knowers’ regress” (Radder, 1996, chap. 2), is likely to occur. ²¹ This knowers’ regress says, quite simply, that in a case where both the existence of a phenomenon and the method for testing its existence are controversial, no resolution can be expected. The existence of the phenomenon could only be successfully tested if there were an agreed-upon test method, but an agreed-upon method could only be successfully developed if there were sufficient agreement about the nature of the hypothesized phenomenon. In the case of the Libet experiment, the critique by Bennett and Hacker (2003, pp. 224–231) illustrates the occurrence of this regress. Based on an alternative view of the nature of human freedom, these authors essentially contest Libet’s claim that (normal) voluntary actions are characterized by an urge to act. Because of this profound disagreement about the phenomenon under investigation, Libet’s experimental test of this phenomenon is deemed inappropriate as well. It can of course not be logically excluded that such fundamental disagreements among scientists and philosophers may disappear in the future. Yet, considering the persistence of widely diverging views on many basic concepts in the history of humankind, assuming such a future disappearance of this diversity constitutes an implausible historical claim.

From our analyses in this article, we may conclude that the results of Libet’s study and more recent Libet-type experiments do not permit strong (in particular, causal) claims about the relation between neurological processes, conscious free will, and motor activity. Instead, these experiments are better characterized as “exploratory experiments” (cf. Steinle, 1997), which help to chart a developing area of intriguing neuroscientific research. We are claiming, therefore, not that the experiments do not contribute at all to the debate on neuroscience and free will, but rather that at present this contribution is still exploratory instead of providing a convincing test for or against the existence of free will.