Featured Video Play Icon

Libet-Style Experiments, Neuroscience, and Libertarian Free Will

Marcelo Fischborn (Federal University of Santa Maria)

[Published version of the paper]

[PDF of Marcelo Fischborn’s paper]

[Jump to Eddy Nahmias’ comment]

[Jump to Adina Roskies’ comment]

[Jump to Marcelo Fischborn’s response]

 

Abstract: People have disagreed on the significance of Libet-style experiments for discussions about free will. In what specifically concerns free will in a libertarian sense, some argue that Libet-style experiments pose a threat to its existence by providing support to the claim that decisions are determined by unconscious brain events. Others disagree by claiming that determinism, in a sense that conflicts with libertarian free will, cannot be established by sciences other than fundamental physics. This paper rejects both positions. First, it is argued that neuroscience and psychology could in principle provide support for milder determinist claims that would equally conflict with libertarian free will. Second, it is argued that Libet-style experiments—due to some of their peculiar features, ones that need not be shared by neuroscience as a whole—currently do not (but possibly could) establish such less demanding determinist claims. The general result is that neuroscience and psychology could in principle undermine libertarian free will, but that Libet-style experiments have not done that so far.

 

Introduction

Recent discussions about free will and cognitive science (especially neuroscience) were largely influenced by some intriguing and controversial experiments conducted by Benjamin Libet and others in the 1980s (see Libet, Wright, Feinstein & Pearl, 1982; Libet, Gleason, Wright & Pearl, 1983; and Libet, 1999). It was known at the time that a specific sort of neural activity called ‘readiness potential’ (RP) preceded voluntary movements (Kornhuber & Deecke 1965). Libet sought to investigate the temporal relation between RPs, movements, and the moment when subjects become conscious of wanting to move. He found that RPs start on average approximately 350 milliseconds before the subjects become conscious of wanting to flex a finger, and approximately 500 milliseconds before actual movement (Libet et al. 1983, Libet 1999).1 Libet concluded that the voluntary acts under examination are initiated unconsciously in the brain. More recently, Soon, Brass, Heize & Heynes (2008) found neural activity that predicts which of two buttons a subject will push 7 seconds (or even 10 seconds) before the subject has consciously decided between the options. Although the accuracy is of no more than approximately 60%, the authors conclude that conscious decisions are determined by unconscious neural activity.

There has been considerable disagreement about the significance of this kind of result for debates about the existence of free will. The aim of this paper is to assess these divergences with regard to a particular conception of free will, namely, libertarian free will. For the purposes of this paper, let us understand as ‘libertarian’ any conception that holds that free will is incompatible with determinism. (‘Determinism’ will be characterized in Section 2.) Below, I start by framing current disputes on the impact of Libet-style experiments on libertarian free will (Section 1), and then I argue for two theses. The first is that, contrary to what some have defended, neuroscience and psychology can, in principle, establish modest determinist claims that might threaten libertarian free will (Section 2). The second is that Libet-style experiments have not so far established that sort of claim, though they could in principle (Section 3). Neuroscience and psychology could in principle undermine libertarian free will, but Libet-style experiments have not yet done that.

 

1. Disputes on the Impact of Libet-Style Experiments on Libertarian Free Will

Some people have interpreted results from Libet-style experiments as a straightforward case against free will. Haynes, one of the authors in Soon et al. (2008), for example, describes the challenge as follows:

…our and Libet’s findings do address one specific intuition regarding free will, that is the naïve folk-psychological intuition that at the time when we make a decision the outcome of this decision is free and not fully determined by brain activity. (Haynes, 2011, p. 92)

Similarly, Misirlisoy and Haggard describe a

…personal experience [that] provides a powerful impetus for the folk concept of free will. We consciously decide on a course of action and only then we do carry out the relevant actions to fulfill it. When presented with a choice of two options, we may think about them, and then we perform a conscious selection between them by exercising our will. In this sense, our will is experienced as free. (Misirlisoy & Haggard, 2014, p. 37)

And they add—partly on the basis of the results in Soon et al. (2008)—that neuroscience has “called this intuition into question, by showing that unconscious activity in the brain preceding our intention—activity that we are never aware of—predicts the emergence of that specific intention to act” (Misirlisoy & Haggard, 2014, p. 38).

The reasoning in these passages seems to be as follows. First, our intuitive conception of free will is said to require that our decisions are not determined by previous (allegedly unconscious) activity in the brain; in other words, a libertarian view of free will is ascribed to common thought. But, second, Libet-style experiments are said to undermine this intuition. As a consequence, our intuitive, libertarian notion of free will is an illusion.

Such confidence in the implications of neuroscience for the free will debate has been challenged by others, remarkably in philosophy. Nahmias offers the following argument schema as a means of clarifying how Libet-style experiments and other results from cognitive science can have an impact on the debate:

  1. Free will requires that X is not the case.
  2. Science is showing that X is the case (for humans).
  3. Thus, science is showing that humans lack free will. (Nahmias, 2014, p. 5)

He then analyzes a group of candidates for ‘X’, the first of which is ‘determinism’. He gets the following argument (see Nahmias, 2014, p. 5):

  1. Free will requires that determinism is not the case.
  2. Science is showing that determinism is the case (for humans).
  3. Thus, science is showing that humans lack free will.

Premise D1 states a form of incompatibilism, and given premise D2, the argument as a whole is a form of hard determinism: free will requires determinism to be false, but since determinism is true, there is no free will.

Nahmias denies, first, that Libet-style experiments can support premise D2 because they would not be in a position to establish determinism such as it is understood by incompatibilists:

In incompatibilist arguments, determinism is defined as the thesis that a complete description of a system (e.g., the universe) at one time and of all the laws that govern that system logically entails a complete description of that system at any future time. (Nahmias, 2014, p. 6)

Nahmias says that this sort of determinism “requires a closed system”, and then objects that the brains and behaviors studied by cognitive scientists are not closed systems. He adds that results such as those in Soon et al. (2008) “do not show that, given prior events […] certain decisions or behavior necessarily occur” (Nahmias, 2014, p. 6, author’s emphasis).

Roskies (2006) offers a similar argument for the claim that neuroscience cannot tell whether the universe is, at a fundamental level, determinist. She argues that observed determinism or indeterminism at one level of description cannot be taken as evidence that another level is deterministic or indeterministic. For example, neuroscientists could come to the conclusion that brains are indeterministic. But, due to the possibility of determinist chaos, she says, “apparent indeterminism in one level of description is entirely compatible with determinism at the fundamental physical level” (2006, pp. 420–421). In this way, Roskies accepts that “neuroscience can indicate […] that, regardless of whether or not the universe is deterministic, the brain effectively is” (p. 421), but insists that it is determinism at the fundamental physical level that is critical for the traditional debate about free will.

Before going ahead, I should mention that Nahmias and Roskies also doubt premise D1 in the argument above. Nahmias argues that cognitive scientists cannot simply assume that premise D1 accurately represents philosophers’ and laypersons’ views. According to him, most philosophers as well as most laypersons seem to be compatibilists. Regarding philosophers’ beliefs, we have evidence from Bourget and Chalmers’ (2013) online survey. And, regarding laypersons’ beliefs, Nahmias mentions results in experimental philosophy by himself and colleagues (Nahmias, Morris, Nadelhoffer & Turner, 2006; Nahmias, Coates & Kvaran, 2007; see Nichols & Knobe, 2007, for a contrary view). And Roskies (2006, p. 422), partly drawing on the same experimental data, also doubts that neuroscience could have an impact on ordinary practices of responsibility, even if it could affect ordinary conceptions about free will.

In the following sections, I do not focus on the question whether compatibilism is conceptually stronger, nor on whether it represents common thought more accurately than incompatibilism. Instead, the focus is on whether Libet-style experiments (and neuroscience, more generally) are, or can be, a threat to free will if incompatibilism is correct, or, as we may put it, if Libet-style experiments (and neuroscience) do, or could, undermine a libertarian conception of free will. This is precisely what is at issue: The scientists mentioned above claim that such experiments actually exclude libertarian free will; the philosophers mentioned claim that neuroscience could not do that in principle.

 

2. Neuroscience, Determinism, and Libertarian Free Will

Let us begin by assessing the claim that neuroscience cannot establish a sort of determinism that is incompatible with libertarian free will. It is true, as Nahmias says, that in discussions among compatibilists and incompatibilists, determinism is often characterized as a thesis concerning the workings of the universe as a whole. In that sense (let us label it ‘universal determinism’), the thesis says, roughly, that all events in the universe—including, of course, human decisions and actions—are caused by previous events in accordance with laws of nature. For the purposes of this paper, I will ignore if neuroscience can support determinism so defined. I want to ask instead if there is a more modest form of determinism that is both (a) sufficient for undermining libertarian free will, and (b) supportable, at least in principle, by neuroscience. I claim that there is, and in order to develop my argument I focus first on why incompatibilists take universal determinism to threaten free will.

In general, libertarians reject universal determinism because, for them, free will requires that we do have (at least sometimes) alternative possibilities for what we do and choose. Chisholm (1964), for example, claims that someone acts freely only if she could have done otherwise. But he rejects a (compatibilist) conditional analysis of ‘could have done otherwise’, that is, an interpretation in which ‘she could have done otherwise’ means that ‘she would have done otherwise if she had chosen otherwise’. Instead of such an analysis—which is consistent with the possibility that, given prior events and the laws of nature, she could not choose otherwise—Chisholm holds that ‘she could have done otherwise’ requires ‘she could have chosen otherwise’:

Suppose, after all, that our murderer could not have chosen, or could not have decided, to do otherwise. Then the fact that he happens also to be a man such that, if he had chosen not to shoot he would not have shot, would make no difference. For if he could not have chosen not to shoot, then he could not have done anything other than just what it was that he did do. (Chisholm, 1964, p. 175-176, author’s emphasis)

In a similar way, Kane says that

when we wonder about whether the wills of agents are free, it is not merely whether they could have done otherwise that concerns us […] What concerns us is whether they could have done otherwise voluntarily (or willingly)… (Kane, 2009, p. 275, author’s emphasis)

In order to be able to do otherwise voluntarily, as Kane says, one must be able to choose otherwise. We have again the requirement of alternative choice possibilities. In contrast with Chisholm, however, Kane does not think it generalizes to every action. For him, libertarian free will requires alternative possibilities only for some actions, those which he labels ‘self-forming actions’ (SFAs). It is because universal determinism entails that (given what happened in the past and the laws of nature) we never have alternative possibilities that libertarians regard it as incompatible with free will. For if everything (including actions and decisions) is determined according on the basis of past events and the laws of nature, then no one can ever choose otherwise.

But now it should become clear that even less demanding forms of determinism can conflict with libertarian free will. Instead of a single statement that every possible event is determined, for example, we might have claims that particular sorts of events are determined—I will refer to these as ‘statements of local determination’. Consider the following schema for generating statements of this sort:

LD: For any event x, if x is P, then x causes another event, y, that is Q.2

LD says that whenever there is an event of sort P, that event causes a second event of sort Q, i.e., events of sort P deterministically cause events of sort Q. We can imagine a similar law that would prevent an individual to choose otherwise given some previous event whose occurrence was not within her control:

LD1: For any event x, and any subject s, if x is a pattern of neural activity of type B in s’s brain, then x causes s to decide to push a given button.

Here, whenever a specific pattern of neural activity happens in a subject’s brain, a specific decision results, namely, to push a given button. It should be clear that we could generate a potentially infinite number of statements of local determination like LD1.

Statements like LD1 are such that, were they true, they could undermine the sort of libertarian free will that we have been examining. Consider Chisholm’s case. If an action is to be free in his libertarian sense, then the agent has to be able to do and choose otherwise. By this criterion, and given LD1, if a pattern of neural activity of type B occurs in a subject’s brain, then, in this particular situation, this subject would be unable to choose otherwise.3 Consequently, an action resulting from such a decision would not be free in Chisholm’s sense. Additionally, the more decisions happened to be determined according to that sort of law, the less would be the space for choices and actions that are free in a libertarian sense.

The situation is more complex in Kane’s account, but not radically distinct. As we have seen, he only requires SFAs to be such that the agent could have done and chosen them otherwise. On his account, if a pattern of neural activity of type B happens in a subject’s brain, causing him to behave according to LD1, this does not entail that the action is not free in a libertarian sense, but merely that it is not a SFA. A free action or choice can be deterministically caused on Kanes’s account, provided that the causal chain ends in a SFA (see Kane, 2009, p. 271-272). Thus, the truth of LD1 would not directly shrink the number of actions resulting from libertarian free will, but only the number of SFAs. But this still allows that the discovery of more and more laws similar to LD1 could decrease our confidence in the existence of SFAs, and, indirectly, decrease our confidence in the existence of libertarian free will.

The result from the discussion so far is that, contrary to the suggestions by Nahmias and Roskies, determinist statements less demanding than universal determinism can equally threaten libertarian free will. And it seems clear that sciences other than fundamental physics, such as neuroscience and psychology, could in principle support statements of local determination similar to LD1. LD1 itself suggests this, since I have deliberately designed it to resemble the results reported by Soon et al. (2008). And there is no reason to think that neuroscience and psychology could not, in principle, find evidence supporting claims of that sort. We can now consider whether Libet-style experiments have already, as a matter of fact, established some statement of local determination.

 

3. Libet-Style Experiments and Statements of Local Determination

The question now is whether results from Libet-style experiments support some determinist claim that potentially threatens libertarian free will. I will argue that they do not. The results currently available are insufficient to establish even such weaker determinist statements as LD1.

If we try to interpret Libet’s original results in the light of LD1, we get something like this:

LDL: For any event x, and any subject s, if x is a RP-II in s’s brain, then x causes s to decide to flex his/her finger “now” and to move his/her finger.

If LDL is true, we can say that readiness potentials of type II determine a peculiar sort of choice, namely, choices to “move now” that are accompanied by actual movement. However, the results fall short of definitely establishing the truth of LDL. Libet measured the time lapse between voluntary movement and RP onset by averaging the EEG signal recorded from 1.4 seconds before finger movements (Libet et al. 1982, p. 324). Only data within this time interval was actually stored and analyzed. That means that, due to its very design, Libet’s original experiments could not find an RP-II that is not followed by a decision to “flex now”, and by actual movement. But this is critical for assessing the truth of LDL. The only way to falsify it is by finding an RP-II that is not followed by a decision to “flex now”. Therefore, Libet’s results support in fact the claim that some RPs of type II are followed by decisions to “flex now”, rather than the stronger LDL. In other words, Libet’s results leave it open whether RP-II is a deterministic cause of decisions to “flex now”, or if it is just something that precedes the sort of action investigated, but that could also precede other sorts of actions and states.4

Consider now the experiments by Soon et al. (2008). Here subjects were asked to choose between a left and a right button, press it immediately after deciding for one of them, and then report the time of the decision. During this process their brain activity was scanned with fMRI. Using advanced decoding techniques, the authors were able to show that the spacial pattern of activation is some brain regions (e.g. BA10 in frontopolar cortex) contained predictive information about which button the subject would choose and actually press. This information was available in the brain at about 7-10 seconds before the time subjects reported to have consciously decided, and it predicted the result with nearly 60 % accuracy (see Soon et al., 2008, p. 544, Fig. 2). In more precise terms, the authors were able to identify some patterns of neural activity whose occurrence indicated that a particular decision would follow with a probability of approximately 60 %—when the chance probability is 50%.

We could also try to infer something similar to LD1 here. We would get a statement of local determination whose antecedent specifies some pattern of neural activity, and whose consequent specifies a particular choice accompanied by behavior (pressing a right or a left button). As in Libet’s case, the study excludes from the start the possibility of identifying those same patterns of neural activity in situations that are not followed by decisions and movements of the types under investigation. But here the possibility of inferring a determinist statement is even smaller (indeed null). Since the accuracy is of just 60 %, it follows that in approximately 40 % of cases those patterns of neural activity were followed by a different decision than the one to be expected. For the sake of argument, name ‘XYR’ a pattern of neural activity whose occurrence rises the probability of a decision to push the right button to 60 %. Given that the right-button and left-button options are mutually exclusive, it follows that we should expect XYR neural activity to be followed by decisions to press the left button in approximately 40 % of cases. That means that in some occasions XYR neural activity is not followed by decisions to push a right button. Therefore, a statement of local determination with the occurrence of XYR as its antecedent and the occurrence of a decision to press the right button as its consequent would be surely false. Of course, it remains an open question whether future studies could improve accuracy, that is, whether we are facing determinist processes still poorly known, or processes that are intrinsically stochastic (see Haynes, 2011, p. 93). Either way—and this is the important point here—we are far from having established a determinist claim that could conflict with libertarian conceptions of free will.

The previous arguments suggest that Libet-style experiments have so far provided no result that could undermine libertarian free will, although neuroscience and psychology more generally could in principle do that. Could Libet-style experiments themselves some day affect libertarian free will? In order to answer this question, we need to flesh out what possible result from a Libet-style experiment would lend support to a statement of local determination. The key difficulty, as we have seen, is to establish that some sort of neural activity happens exclusively in situations that are followed by a particular sort of decision. There are technical difficulties here. In the case of type-II RPs, one needs a reference point on the basis of which EEG recordings from many trials can be averaged. This makes it difficult, practically, to investigate if RPs that are candidates for deterministic causes of specific decisions can appear without the expected effect (see Libet, 1985, p. 538; Gomes, 1999, p. 64). But practical difficulty does not mean impossibility. One possibility would be to add some form of intervention to Libet-style experiments that induced RPs whose effects could then be analyzed. Additionally, one could have a comparison between intervention and control conditions—a methodology widely used in attempts to infer causal connections. I think the same strategy could possibly be employed with the experiments by Soon et al. (2008). As we have seen, however, we face here the preliminary issue that we do not have at present a plausible candidate for determinist cause of decisions that could be tested. Again, this only suggests difficulties of a practical order and the absence of established threats to libertarian free will, but not a principled impossibility.

 

Conclusion

There has been divergence about the significance of Libet-style experiments for discussions about free will. In what concerns specifically libertarian free will, my conclusion is that parties have drawn exaggerated conclusions. Contrary to what one side has defended (e.g. Roskies, 2006; Nahmias, 2014; Sinnott-Armstrong, 2011), experiments in neuroscience and psychology could, in principle, support determinist statements that could undermine libertarian free will. But, contrary to what those in the opposite side have insisted (e.g. Misirlisoy & Haggard 2014; Haynes, 2011), results so far obtained fall short of actually supporting even those weaker statements of local determination. Assumptions involving libertarian free will are often in place in discussions about free will and neuroscience. First, because libertarianism is a more demanding view, both metaphysically and empirically, some have assumed that if science leaves space for free will at all, then it must be for some weaker, compatibilist sort of free will (see, e.g., Koch, 2012, p. 111; Schlosser 2012). This, together with a second assumption that libertarianism is the correct view (or that it better represents laypersons’ views), has also lead some to conclude that neuroscience shows that free will (in itself, or in the way it is commonly understood) is an illusion (e.g., Haynes, 2011; Harris, 2012, p. 16; Misirlisoy & Haggard, 2014, p. 37).5 If the present results are correct, data from Libet-style experiments lend support to none of those assumptions, although they (as well as other studies in neuroscience and psychology) could in principle do that.

 

References

Bourget, D. and Chalmers, D. (2013). What do philosophers believe? Philosophical Studies.

Chisholm, R. (1964). Human freedom and the self. In Pereboom (2009), pp. 172–184.

Davidson, D. (1967). Causal relations. In D. Davidson (2001). Essays on actions and events (pp. 149–162). 2nd edition. Oxford: Oxford University Press.

Gomes, G. (1998). The timing of conscious experience: A critical review and reinterpretation of Libet’s research. Consciousness and Cognition, 7, 559–595.

____. (1999). Volition and the readiness potential. Journal of Consciousness Studies, 6.8-9, 59-76.

Harris, S. (2012). Free will. New York: Free Press.

Haynes, J. (2011). Beyond Libet: Long-term prediction of free choices from neuroimaging signals. In W, Sinnott-Armstrong & L. Nadel (2011), 85–96.

Kane, R. (2009). Free will: New foundations for an ancient problem. In D. Pereboom (2009), pp. 268–288.

Koch, C. (2012). Consciousness: Confessions of a romantic reductionist. Cambridge, MA: MIT Press.

Kornhuber, H, & Deecke, L. (1965). Hirnpotentialänderungen bei Willkürbewegungen und passiven Bewegungen des Menschen: Bereitschaftspotential und reafferente Potentiale. Pflüger’s Archiv für die Ges. Phys, 284.1, 1–17.

Libet, B. (1985). Unconscious cerebral initiative and the role of conscious will in voluntary action. The Behavioral and Brain Sciences, 8, 529-566.

____. (1999). Do we have free will? In W. Sinnott-Armstrong & L. Nadel (2011), pp. 1–10.

Libet, B., Gleason, C., Wright, E. e Pearl, D. (1983). Time of conscious intention to act in relation to onset of cerebral activity (readiness-potential): The unconscious initiation of a freely voluntary act. Brain, 106, 623–642.

Libet, B.; Wright, E; Feinstein, B. & Pearl, D. (1982). Readiness potentials preceding unrestricted ‘spontaneous’ vs pre-planned voluntary acts. Electroencephalography and Clinical Neurophysiology, 54, 322–335.

Mele, A. (2009). Effective intentions: The power of conscious will. New York: Oxford University Press.

____. (2006). Free will: Theories, analysis, and data. In S. Pockett et al. (2006), pp. 187–206.

Misirlisoy, E. and Haggard, P. (2014). A neuroscientific account of the human will. In Sinnott-Armstrong (2014), pp. 37–42.

Nahmias, E. (2014). Is free will an illusion? Confronting challenges from the modern mind sciences. In W. Sinnott-Armstrong (2014), pp. 1–25.

Nahmias, E., Coates, J. and Kvaran, T. (2007). Free will, moral responsibility, and mechanism: Experiments on folk intuitions. Midwest Studies in Philosophy, 31, 214–232.

Nahmias, E., Morris, S., Nadelhoffer, T., and Turner, J. (2006). Is incompatibilism intuitive? Philosophy and Phenomenological Research, 73, 28–53.

Nichols, S. & Knobe, J. (2007). Moral responsibility and determinism: The cognitive science of folk intuitions. Noûs, 41:4, 663–685.

Pereboom (ed.) (2009). Free will. Indianopolis: Hackett.

Pockett, S., Banks, W. & Gallagher, S. (eds.) (2006). Does consciousness cause behavior? Cambridge, MA: MIT Press, 2006.

Pockett, S. & Purdy, S. (2011). Are voluntary movements initiated preconsciously? The relationships between readiness potentials, urges, and decisions. In Sinnott-Armstron & Nadel (2011), pp. 34–46.

Roskies, A. (2006). Neuroscientific challenges to free will and responsibility. Trends in Cognitive Science, 10:9, 419–423.

Schlosser, M. (2012). Free will and the unconscious precursors of choice. Philosophical Psychology, 25:3, 365–384.

Sinnott-Armstrong, W. (2011). Lessons from Libet. In Sinnott-Armstrong & Nadel (2011), pp. 235–246.

____. (ed.) (2014). Moral psychology (vol. 4): Free will and moral responsibility. Cambridge, MA: MIT Press.

Sinnott-Armstrong, W. & Nadel L. (eds.) (2011) Conscious will and responsibility. Oxford: Oxford University Press.

Soon, C., Brass, M., Heize, H, & Heynes, J. (2008). Unconscious determinants of free decisions in the human brain. Nature Neuroscience, 11:5, 543–545.

 


Notes

  1. This information refers only to what Libet calls ‘type II’ RP, i.e., RPs preceding movements for which subjects reported no previous planning of the moment to move. For other conditions, see Libet et al. (1982, 1983). It is worth noting that both the specific measurements and the implications for free will of Libet’s results are a matter of dispute. On the former, mentioned difficulties include the effects of instructions and training during the experiments, and subjects’ ability to accurately report the time of decisions (see, e.g., Gomes, 1998). Questions related to the latter point include the representativeness and significance of finger flexings for free will, the precise nature of the mental phenomena investigated, and various others (see, e.g., Mele, 2006, 2009, the essays in Sinnott-Armstrong & Nadel, 2011, and in Part II of Pockett, Banks & Gallagher, 2006, as well as most of what is discussed below).
  2. This is a modified and simplified version of an analysis of causal laws developed by Davidson (1967, p. 158).
  3. Strictly speaking, the subject would be unable not to choose to push a given button. The logical possibility (whatever its empirical plausibility) remains that the subject could make simultaneously other, unrelated decisions. What is usually taken to be relevant in the free will debate, however, is the possibility of not choosing in a particular way.
  4. Pockett and Purdy (2011, p. 36–37) say that “Waveforms that look like RPs have been known for decades to occur before a variety of expected events that are not movements”. This suggests that RPs in fact are not uniquely related to decisions to “flex now”. It should also be mentioned that Libet’s experiments on ‘veto’ conditions—when subjects were instructed to prepare to move at a prearranged time and, shortly before, block that preparation—indicated that a great initial portion of an RP of type I may not be followed by actual movement (see Libet, 1985, p. 537-538, especially Figure 2).
  5. As I have noted earlier, controversies remain in the philosophical debate on compatibilism versus incompatibilism, as well as in the experimental research on laypersons’ beliefs about free will. On the latter, see, for example, Nahmias, Morris, Nadelhoffer & Turner (2006), Nahmias, Coates & Kvaran (2007), and Nichols & Knobe (2007).

7 thoughts on “Libet-Style Experiments, Neuroscience, and Libertarian Free Will”

  1. What is ‘neuro-determinism’, does it support incompatibilism, and might it be true?

    Marcelo Fischborn has clearly articulated why existing neuroscience studies do not establish a type of determinism which threatens free will, a point with which I agree. He also raises important challenges for those, like Adina Roskies and I, who have questioned whether neuroscience or other mind sciences could establish a type of determinism relevant to debates about free will. Fischborn argues that neuroscience could establish ‘local determinism’ such that neural activity of type A (necessarily) causes neural activity (or behavior) of type B, presumably in a law-like way in all relevant organisms. I will begin by explaining why the local determinism Fischborn describes does not support arguments for incompatibilism, and then I will argue that it is unlikely to be established by neuroscience in any form that should trouble compatibilists or libertarians—that is, anyone who thinks we might have free will.

    In my chapter (2014) that Fischborn discusses, I argue that neuroscience and other cognitive sciences cannot establish determinism as it is used in incompatibilist arguments. What I had in mind were the two most influential arguments for the conclusion that free will is inconsistent with universal determinism—namely, the Consequence Argument (e.g., van Inwagen 1983) and the Manipulation Argument (e.g., Pereboom 2001, Mele 2013). The Manipulation Argument works by stipulating a powerful manipulator who has complete knowledge of the state of the universe, say, 30 years ago, and of all the laws of nature, and she uses that knowledge to intervene on an agent (before or soon after his birth) to ensure that he does exactly what she wants 30 years later (while satisfying compatibilist conditions). The argument then challenges the compatibilist to explain what difference there is between that agent and a ‘twin’ agent whose un-manipulated life follows the same trajectory in a deterministic universe, or to bite the bullet of saying that such a manipulated agent has free will and is morally responsible for the manipulated action.

    But if indeterministic events occur, even if they occur only outside the agent, the manipulator would be unable to ensure that the agent does what she wants—for instance, a few quantum events, magnified by non-linear dynamics, could radically alter the state of the universe in which the agent found himself 30 years down the line. Hence, the Manipulation Argument seems to require universal determinism to work; local determinism—for instance, in the human brain—won’t do the trick.1

    Similarly, the Consequence Argument relies on universal determinism to reach its conclusion. The argument relies on a premise that no one has a choice about (or control over) the state of the universe in the distant past (or about the laws of nature), and (universal) determinism entails that the distant past and laws entail one’s present actions. The argument will not work if there are any indeterministic events anywhere that occur after the distant past and influence one’s (much later) actions. And the premise that no one has a choice about states in the past becomes much less intuitive if those past states are only a few moments ago, such that local determinism might apply.

    So, both of these incompatibilist arguments rely on a definition of universal determinism standardly used in the free will debates (see, e.g., Hoeffer 2010; van Inwagen 1983), namely:

    Universal Determinism: A universe is deterministic iff a complete statement P of the state of the universe at a time T, together with a complete statement of the laws L that govern that universe, entails a statement about any event E in that universe [at any time later than T]. That is, Necessarily((P&L)  E). (Some definitions leave out the bracketed phrase to indicate a time-symmetric form of determinism.)

    Though some eschew defining determinism in terms of causation, here is another possible definition that I think could be used in modified versions of the incompatibilist arguments:

    Universal Causal Determinism: A universe is deterministic iff for every event E, there is some set S of prior events, such that, as governed by the laws L, S is a sufficient cause of E. That is, S causes E with probability 1. (Note that S is going to get larger and larger as the time between S and E increases and more background causes must be in place to ensure that E occurs.)

    Now, it might be that incompatibilists could mount an argument that does not focus on universal determinism, but instead on local determinism of the sort Fischborn describes. But it is unclear what such an argument would look like, or whether it could be mounted without relying on premises and principles that rule out free will in an indeterministic universe as well, in which case it seems local determinism be irrelevant. (For instance, Derk Pereboom, Galen Strawson, and others argue that free will is impossible whether determinism is true or false; it is not determinism, universal or local, that poses the problem, but rather the impossibility of agents’ being originating causes, or sources, of their actions.)

    Robert Kane and some other libertarians (e.g., Laura Ekstrom, Mark Balaguer) certainly worry about local determinism in the brain, since their theories of free will require that indeterministic events occur in the brain (at the right time and place). However, they arrive at the conclusion that such events are required for free will only by first developing (or accepting) incompatibilist arguments using universal determinism (such as those described above) and then positing indeterministic events in the time and place they think makes the most sense for free will (and they disagree amongst themselves where best to ‘place’ the indeterminism). They do not present incompatibilist arguments using local determinism.

    So, Fischborn is right that local determinism might pose a threat to such libertarian theories, if neuroscience could actually establish it (but see below). But why should we accept the incompatibilism that motivates these libertarians unless we are considering the sort of universal determinism that neuroscience cannot establish, since only physics ‘covers’ the entire universe. Indeterminism anywhere, not just in the brain, falsifies the sort of determinism used in incompatibilist arguments.

    As far as I know, there are no noteworthy incompatibilist arguments using local determinism.2 As I point out in my chapter, scientists who argue that neuroscience challenges free will sometimes talk as if local determinism at the level of neural activity is what poses the problem. But they don’t mean it. Instead, they focus on either the causal irrelevance of conscious mental states—for instance, because they think Libet’s and others’ studies show that such states occur too late to make a difference—or the causal irrelevance of non-physical mental states, because they think neuroscience establishes that dualism is false and our behavior is caused only by physical states (or they think these two worries are the same).

    But neither epiphenomenalism nor physicalism require the truth of local (or universal) determinism—that is, indeterministic quantum events in the brain would not make a difference to whether epiphenomenalism or physicalism were true (though some have tried to posit that non-physical mental events somehow ‘use’ quantum indeterminism as a place to make a difference—e.g., Popper & Eccles (1977) and perhaps Libet). Conversely, local determinism in the brain would not entail epiphenomenalism or physicalism. Conscious mental states can make a causal difference if the neural processes on which they supervene (or with which they are identical) make a causal difference, even if the causal relations are deterministic.3 Even dualism is consistent with local determinism if the intervening activity of a non-physical mind serves as a necessary conjoint cause with the neural activity that causes the relevant outcomes—that is, when neural activity of type X causes behavior of type Y, it does so only because ‘soul’ activity of type S is a cause of Y as well (of course, being non-physical, such activity will be undetectable by neuroscientific research, such that the evidence will suggest that the neural activity alone is sufficient).

    So far, I have argued that local determinism does not raise a challenge to free will without relying on universal determinism (e.g., because one has already bought into a libertarian view inspired by arguments that require universal determinism), or that local determinism is not the real source of the challenge (e.g., because the real challenge is epiphenomenalism or physicalism or lack of causal sourcehood). Yet we might still wonder whether there is something intuitively threatening about local determinism in the brain.4 If so, we’d want to know if neuroscience could establish the truth of local ‘neuro-determinism’. I think it’s unlikely.

    Here is Fischborn’s definition of ‘local determination’ or deterministic causation as it applies to human neural activity and a specific type of behavior:

    LD1: For any event x, and any subject s, if x is a pattern of neural activity of type B in s’s brain, then x causes s to decide to push a given button.

    Now, no one should be particularly worried about whether specific brain activity deterministically causes us to push a button, in part because it’s a trivial behavior, and in part because the determining neural activity (of type B) is probably not a “previous event whose occurrence was not within [the subject’s] control” (the phrase Fischborn uses to set up his definition).5 After all, if the subject just agreed to be in a neuroscience experiment that involves pushing a button whenever she feels like it (e.g., as in Libet’s and Soon et al.’s paradigm), then her decision to agree to do so (or the neural activity underlying her decision) was both a significant cause for the existence of neural activity of type B, and it was under her control—unless, of course, this earlier decision was also not under her control. But now it looks like we need an argument to show that all of our decisions and behaviors are not under our control, or are such that we have no choice about them. Here, we might go back to a standard incompatibilist argument using universal determinism, but if we’re shifting our focus to local determinism in the brain, we will instead want to see if neuroscience might show that claims like LD1 can be generalized to all decisions and behaviors. Fischborn suggests just this when he writes, “It should be clear that we could generate a potentially infinite number of statements of local determination like LD1.”

    But it is not at all clear that we could generate infinitely many statements like LD1. That is, it is not clear that, even if future neuroscience might discover some deterministic ‘little laws’ between trivial or automatic behaviors and immediately prior neural states, it will also discover deterministic causal relations between complex behaviors or decisions and specific prior neural activity. There are (at least) three reasons for doubting this sort of ‘neuro-determinism’: multiple realizability, complexity and chaos, and embodied and extended cognition. For lack of space, I’ll just toss these problems out for discussion without much explanation.

    First, if multiple realizability is true, then there will not be LD statements that apply across all brains or even within the same brain over time. Instead, many of the law-like relations, should they exist, will be among events at a higher-level of description, and these events will be realized in a variety of neural processes. If there are any deterministic relations at the neural level, they would involve the mechanistic components (e.g., among synapses) that realize the higher-level psychological events, such as perceptions and decisions, but these components will be organized in different ways within and between subjects as the realizers of the same psychological events. Hence, many of the causal generalizations involved in human decisions and actions would not be deterministic at the neural level.6 Note that if such law-like causal relations exist at the psychological level, it is much less intuitively threatening to free will than determinism at the mechanistic level, likely because the latter suggests that the causal work is being done in a way that leaves no causal work for our conscious deliberations and decision-making (see Nahmias, Coates & Kvaran 2007).

    Second and relatedly, the complexity of the human brain suggests that any complex decision will involve such a wide disjunction of neural processes that there will be no pattern of neural activity of a given type that can be picked out as the cause of that decision. Assuming physicalism and universal determinism, there will be a token set of neural activity just prior to the decision that ensures that decision occurs. But that exact (token) set of neural activity will likely never happen again, nor will it ever happen in another agent. Throw in chaos theory (non-linear dynamics) and seemingly insignificant changes to any token pattern of neural activity might have very significant effects on the ensuing decision or behavior. Hence, the specificity of neural activity that determines later neural activity (such as the activity that realizes specific decisions) becomes increasingly fine-grained. And the likelihood of there being law-like deterministic relations among complex neural states becomes vanishingly small.7

    Finally, if theories of extended or embodied cognition are on the right track, there will be events and processes outside the brain, and perhaps outside the body, that are components of some of our mental states, including deliberations and decisions, and hence part of the cause of some of our behaviors. If so, neural activity alone will not be sufficient to cause some decisions or behaviors. Hence, local determinism in the brain could not be generalized to some decisions and behaviors. For instance, if the letters on the screen as I type are an integral component (and not just cause) of my decisions about how to write this sentence, then there will be no law-like relation between neural states alone and my decisions.

    If physics, contra the currently dominant interpretation of quantum physics, ends up discovering that the fundamental laws that govern causal relations among the physical constituents of the universe are deterministic, then universal determinism will be true. Because humans are a part of the physical universe, that would mean that our decisions and behavior would be causally determined. Incompatibilists worry about this possibility (with some worrying that quantum indeterminism would not help). Compatibilists like me need to address their arguments. We also need to address challenges from the modern mind sciences that suggest epiphenomenalism or related threats. But neither the incompatibilist arguments nor the scientific challenges arise because of the (limited if any) deterministic relations neuroscience might discover.

    Seeing why neuroscience is unlikely to discover such deterministic relations helps us see why compatibilism makes sense. Our complex decisions cannot be causally explained (or predicted) by reference only to types of neural activity, much less by reference to types of states of the universe in the distant past (whatever that might mean). Instead, they often will be causally explained best by reference to our prior mental activity, which of course, is realized in (token) neural activity (or perhaps neural activity plus states of the body or beyond). Our individual mental states as we deliberate and make decisions, realized in our complex and unique brains, thus serve as the causal source of many of our decisions and behaviors, even if our universe turns out to be deterministic.8

    At long last, let me conclude with two apologies. First, I’m sorry that my response has taken us so deep into the swamps of the free will debate and murky questions about how to understand determinism. I didn’t mean for this to happen, but as I engaged with Fischborn’s interesting arguments, I found myself pulled into these swamps and this commentary maps my attempt to find my way out. It turns out that just defining ‘determinism’ is a tricky task (e.g., see here), and unchartered territory remains on my map of the relationships between determinism, causation, and levels of explanation (and the sciences that study those different levels).

    Second, I’m sorry if my response to Fischborn might seem uncharitable. A charitable reading might suggest that he aimed to show that neuroscience could raise problems for libertarian theories of free will, such as Kane’s, which require indeterministic events in the brain. And a charitable—and correct—response is that Fischborn is right about that (though I wonder whether, instead of posing these problems for Kane by discovering lots of deterministic relations like LD1, neuroscience is more likely to pose problems by obtaining enough evidence to conclude that quantum indeterminism never has influences on neural interactions, or it gets ‘washed out’ at that level, a discovery that would surprise me).9 Fischborn has indeed put pressure on these libertarian accounts, but I (charitably?) took him to be raising some of the broader issues I’ve dragged us into the swamps to discuss here. In any case, I appreciate his paper, I am happy to be a part of this discussion, and I look forward to everyone’s comments!10

     

    References

    Hoefer, C. 2010. “Causal Determinism.” The Stanford Encyclopedia of Philosophy (Fall 2015 Edition), Edward N. Zalta (ed.), forthcoming URL = http://plato.stanford.edu/archives/fall2015/entries/determinism-causal/.

    Mele, A. 2013. “Manipulation, Moral Responsibility, and Bullet Biting,” Journal of Ethics, 17(3): 167–84.

    Nahmias, E. 2014. “Is free will an illusion? Confronting challenges from the modern mind sciences.” In W. Sinnott-Armstrong (ed.), Moral Psychology, vol. 4: Freedom and Responsibility (Cambridge: MIT Press), 1-25.

    Nahmias, E., Shepard, J. & Reuter, S. 2014. “It’s OK if ‘My Brain Made Me Do It’: People’s Intuitions about Free Will and Neuroscientific Prediction.” Cognition 133(2): 502-513

    Nahmias, E., Coates, J. & Kvaran, T. 2007. “Free Will, Moral Responsibility, and Mechanism: Experiments on Folk Intuitions.” Midwest Studies in Philosophy 31: 214-242, 2007.

    Pereboom, D. 2001. Living Without Free Will. Cambridge: Cambridge University Press.

    Popper, K. & Eccles, J. 1977. The Self and its Brain. New York: Springer-Verlag.

    Van Inwagen, P. 1983. An Essay on Free Will. Oxford: Oxford University Press.

    Notes

    1. Oisin Deery and I (manuscript) argue that the manipulator could not pull off the trick even in a deterministic universe, since at the time she does her calculations, she would not have access to events that are outside her past light cone—too far away to reach her even at the speed of light—but those unknowable events might be causally relevant to the agent’s actions in 30 years, since they are in the past light cone of those actions.
    2. Perhaps one could be developed if one was aiming for the conclusion Pereboom aims for: that free will is incompatible with one’s actions “being causally determined by factors beyond one’s control.” But Pereboom never defines exactly what he means by such causal determination, and it’s not clear how the cases required for the Manipulation Argument works if universal determinism does not apply. (See note 5 below.)
    3. Recognizing this point is crucial, I believe, in avoiding the misleading intuition that if our brains cause our decisions, we don’t. Of course, we need to deal with the causal exclusion argument here, but that argument does not require deterministic causal relations at the lower level.
    4. In Nahmias, Shepard, and Reuter (2014) we present experimental results suggesting that most people do not find free will to be undermined by the stipulation of perfect prediction of decisions based on preceding neural activity—which might suggest local ‘neuro-determinism’.
    5. It may be that Fischborn is gesturing here towards an argument that is similar to van Inwagen’s Mind Argument (1983), something like: (1) No one has a choice about the neural activity that causes later decisions; (2) if no one has a choice about what causes X, then no one has a choice about X; (3) so, no one has a choice about their decisions. I think both of these premises are implausible.
    6. By analogy, interactions among molecules might be deterministic (putting aside quantum effects for now), but any law-like causal relations among colliding billiard balls will be multiply realizable by different molecules and will be described in terms of the higher-level structures and interactions among the balls, not in terms of the laws that describe the molecules.
    7. Note that Kane relies on non-linear dynamics to allow quantum indeterminism to ‘percolate up’ to influence which among competing neural networks ‘wins out’ to cause the outcome of a close-call decision (or SFA). But the chaos without the quantum indeterminism still gets us a lack of deterministic relations between specifiable types of neural activity and complex decisions or behaviors.
    8. Oisin Deery and I (manuscript) use causal modeling to develop this point in more detail, providing a compatibilist account of causal sourcehood. We apply this account to Manipulation Arguments to elucidate a principled difference, relevant to free will and moral responsibility, between a manipulated agent and a (merely) determined agent. The manipulated agent is not the causal source of his actions because there is another source (the manipulator) that better predicts and explains his decision. The same is not true for the determined agent.
    9. It’s always struck me as a reductio of Kane’s view (and others like it) that it seems to entail that future neuroscientists could endow humans with free will for the first time. How? First, they discover that there are no indeterministic interactions that influence causal interactions in the human brain. Then, they figure out how to implant a device in the brain (just where Kane needs indeterminism) that uses quantum indeterministic events to influence the outcome of those neural processes essential to Kane’s SFAs (where before our close-call decisions were only apparently indeterministic). The neuroscientists implant these devices in two babies, Adam and Eve, so they grow up with them as an integral part of their decision-making process, such that the indeterminism properly ‘belongs’ to them, as Kane would say. Adam and Eve would, of course, experience their choices no differently than we do, but according to Kane, they would be the very first humans with free will and the first to be morally responsible for their actions—i.e., those actions that properly derive from SFAs. Perhaps Kane does not see this scenario as a tough bullet to bite, but it seems absurd to me. (I heard an objection to Kane along these lines at a conference long ago, but I cannot remember who raised it.)
    10. Thank you to Brett Castellanos for organizing this session and inviting me to participate, and to Adina Roskies and Oisin Deery for helpful comments on an earlier version of this paper.
  2. Fischborn discusses recent philosophical arguments concerning the significance of neuroscience for discussions of free will. He argues that Libet-style experiments have failed to threaten Libertarian free will, but that in principle, neuroscience and psychology could do so, by supporting a “milder determinism.” In what follows I will argue that he is correct in his assessment of the extant Libet-style experiments, but that he is wrong that other experiments in this style could succeed where these fail. More importantly, however, I disagree that a “milder determinism” is a coherent concept that can impact the free will debate.

    I agree with the argument made about the Libet cases. In the original Libet experiments the data was collected only in cases in which movement occurred, making it impossible to determine whether there were RP-like events that did not co-occur with movement. Recent work by Schlegel and colleagues have shown that RPs can occur in the absence of motor activity, and thus the presence of an RP does not unfailingly indicate the occurrence of a choice to move (Schlegel et al. 2013). Further studies show that RPs occur during anticipation of a timed event, or during a cognitive decision task (Schlegel, in prep). RPs even precede hypnotically induced actions which the subject may not be aware of making (Schlegel et al. 2015). All these results are consistent with the claim that the RP is a reflection of brain events of which we are unaware, but this is not enough to secure the claim that the brain events which the RPs reflect are unconscious decisions. Nor is the fact that we are unaware of these brain events evidence for our brain making up our mind before we do.
     
    Indeed, recent work by Schurger et al suggests that RPs could be a natural consequence of experiments that probe the events preceding action by averaging over spontaneous fluctuations in baseline activity over many trials time-locked to action initiation, if an accumulator model of decision is presupposed (Schurger, Sitt, and Dehaene 2012). Time-locking to movement results in curves that appear as gradual exponential increases in neural activity prior to action, just like the RP. Independent work in neuroscience provides evidence for an accumulator model. Thus, RPs are likely an artifact of the experimental design, given a popular neural model for decision-making.

    More importantly, the fact that we are unaware of the brain events underlying decisions tells us nothing about their role in cognition. After all, we are unaware of the brain events which constitute our conscious perceptual experience, too! What made Libet’s experiments seem compelling was evidence that we had made our decisions without being conscious of having yet decided. The way in which Libet timed the awareness of decision has been widely criticized as problematic. Not frequently recognized is the fact that the way in which he measured that time ensured that he was not probing the time of conscious decision, but rather another state: the awareness of our conscious decision, a metastate which is dependent on that decision, and thus must occur after the decision was made (Roskies 2010). Thus, it is unsurprising that that state occurs late in the game, after decision-making neural processes are evident.

    The experiments by Soon et al are a different kettle of fish. They found mild predictive power in brain imaging data, up to 10 seconds prior to decision (Soon et al. 2008). To interpret the 57% accuracy in these experiments (only 7% above chance) as evidence of determination is ludicrous. What the data shows is that some information about future choices exists in the brain, based on its previous experience. This is unremarkable. When choosing randomly, for instance, one retains a record of prior choices, and future choices can be influenced by the nature of that record. If I have chosen one way three times in a row, I will have encoded that information, and may be more likely to choose the other option the next time. A weak predictive power is unsurprising in these situations, and it indicates only that our decisions are sensitive to temporal factors such as prior history, not that they are determined.

    Fischborn focuses on the potential of these experiments to undermine Libertarian free will, which he characterizes as the view that our decisions are not determined by previous neural activity. This is imprecise: many Libertarians accept that our decisions are dependent on neural activity, but that the neural activity itself is a source of indeterminism, the indeterminism which grounds free will. Thus, merely showing that decisions depend upon brain activity is insufficient to undermine Libertarian free will. One must show that this activity is deterministic. I agree with Fischborn that none of the Libet-style experiments show this.

    Fishborn’s larger point is that some could. He thinks neuroscience can support a milder determinism, or “local determinism”. This is a scope-limited determinism: the claim that for all events of a certain type, they are followed by another event, without exception. I first discuss the plausibility of such a law. I then contend that either this type of local determinism is not a threat to free will, because the event-types that it identifies are so limited as to leave open the possibility of doing otherwise, or they are so broad that neuroscience could never provide compelling evidence for their existence. Let me explain.

    At first glance, Fischborn’s claims seem plausible. Perhaps it is the way in which he couches his principles. For example, he refers to LD1 as a law. LD1 says “ For any event x, and any subject s, if x is a pattern of neural activity of type B in s’s brain, then x causes s to decide to push a given button.” Once we accept LD1 as a law, the rest of the argument goes down more easily. But I see no reason to take it to be a law, without begging the question. First, I don’t see what in neuroscience would give us license to call LD1 a law, rather than a generalization. How do we know that for any event x and subject s such a principle will hold? (For that matter, how do we generalize over subjects – what is the same pattern of neural activity across two different brains?) For exceptionless laws, we look to physics, not neuroscience. I am of course not saying that local determinism cannot exist at all: I do think that the brain evolves according to the laws of nature. But I doubt that neuroscience could give us principles like LD1 that can do the job. Secondly, LD1 says that a pattern of neural activity x of type B causes s to decide to y. Surely some brain pattern could cause one to make a certain decision. For example, if you blasted s’s left motor cortex with enough electric current, s’s right finger would move and push a button, if there was one under his finger. But we would not call his finger movement the result of deciding to push the button. And there may in fact be normal patterns of brain activity that result in a right-sided button push, such as the patterns that constitute such a decision, or perhaps immediately precede reaching the decision threshold coupled with other activity that is sufficient to surpass it. Perhaps, once this pattern is instantiated the subject cannot choose otherwise. But this is not enough to show that he could not have, since earlier in the evolution of the pattern things could have been different. Merely passing a point of no return is not equivalent to lack of free will: I approach a traffic light in my car, and it turns yellow. I must decide whether to step on the gas or brake. I decide. I step on the gas and accelerate before the intersection. Given how fast my car is going, I cannot now stop if the light turns red. That does not mean I did not freely decide to drive through the intersection rather than stop. Just because a decision may be unchangeable at a certain point in time does not mean it was not made freely.

    Now let us consider the kind of neural pattern event-types that Fischborn invokes in LD1. Suppose that the event-type at issue is the firing of some connected neurons: whenever neurons ABC fire, then neuron D fires. I suppose neuroscience in principle could show that D, when monitored, always fires as a consequence of ABC (but, notice, not that it always must). But nothing resembling a human choice is encoded in such a small and restricted group of neurons, nor is any choice manifest as action in a simple 2-level hierarchy. We could not interpret such local determinism as being a psychological state. When we take into account that each neuron in the brain makes approximately 10,000 connections to other neurons, and is affected by as many, that there are multiple stages of processing, interactions with inhibitory interneurons, and crosstalk with other circuits for any action we might take and decisions we might consider, we see that local determinism would have to be fairly global to correspond to decisions, choices, actions, beliefs, or what have you. So even if local determinism can be pinpointed, it will be of a neural subsystem so impoverished that the neurons identified will not correspond to decisions or actions.

    Conversely, any neural system rich enough to subserve the process of decision-making and issuance in action will be too complex and interconnected to be the seat of local determinism. Alterations at any of multiple levels of processing could change the decisions; other thoughts, sensory inputs, or even noise could lead to a person’s doing otherwise. There will be a requirement of global knowledge to establish determinism in a system, and global knowledge of that sort is unattainable. Moreover, even if it were, some of the more promising Libertarian theories are consistent with determinism in many cases of action, and thus do not depend upon indeterminism for all free actions. Kane’s theory, for example, predicates freedom on indeterminism for self-forming actions (SFAs); these may occur infrequently and only under certain conditions. Subsequent actions are free in virtue of following from a character shaped by SFAs, yet entirely determined. Local determinism would fail to be evidence for the claim that SFAs do not exist. Thus, even evidence of local or even more broad determinism would be insufficient to falsify a theory like Kane’s.

    Fischborn suggests that finding enough local determinisms will decrease our confidence in the existence of Kane’s SFAs, and thus in Libertarian free will. Probably not. If one expects SFAs to be rare, and brain patterns almost infinite, then a good Bayesian will agree that seeing some local determinisms will not decrease credence in SFAs very much. Moreover, from what I know about stochastic activity in the brain, the likelihood of finding many local determinisms is small, and I suspect, they would be quite local. And what should one conclude when one observes seemingly stochastic activity if one widens one’s scope beyond that local network? That the variations in activity are merely effects of different but deterministic networks? Or that they reflect indeterministic activity? The answer matters greatly for how one updates one’s credences, but it is unclear what reasons one has to choose one or the other option. Thus, I maintain that to truly undermine Libertarianism, you will need more than just evidence for instances of law-governed activity. You will need reasons to think that indeterministic events do not affect brain activity, and thus for a global determinism (extending at least through the brain, if not beyond). Merely seeing evidence consistent with determinism will not suffice.

    Where does this leave things? I’ve argued that a “milder determinism” will not provide evidence that bears signficantly on the plausibility of Libertarian theories of free will. I agree with Fischborn that Libet-style experiments fail to threaten free will, but for deep structural reasons, and not just the idiosyncratic shortcomings that he points to here.

  3. In my paper, I insisted in two central theses. First, that neuroscience could in principle provide results that would pose a threat to libertarian free will. Second, that Libet-style experiments have not found that kind of result so far. In their comments, Adina Roskies and Eddy Nahmias agree with the second point, but both make reservations regarding the first one. Roskies says that I am “wrong that other experiments in this style could succeed where these fail. More importantly, however, I disagree that a ‘milder determinism’ is a coherent concept that can impact the free will debate”. More subtly, Nahmias argues that those deterministic statements by means of which I claim neuroscience could impact on libertarian free will are “unlikely to be established by neuroscience in any form that should trouble compatibilists or libertarians—that is, anyone who thinks we might have free will.” Here I focus on these apparent divergences; I am not sure they are so deep as they might seem.

    Why do I think neuroscience could have an impact on libertarian free will? My paper’s point was that this could occur if results from neuroscience showed that particular choices are not free in a libertarian sense. And this could in principle be done by showing that, in such choices, subjects could not choose otherwise because some previous neural activity determined that only a particular choice could result. Determination of the kind at issue would obtain if at least two claims of the following kind were true:

    1. Statement of local determination (LD): For any event x, and any subject s, if x is a pattern of neural activity of type B in s’s brain, then x causes s to decide to push a right button.
    2. Statement about brain activity: Neural activity of type B occurred in John’s brain at t.

    Given these statements, it follows that John decides (or will decide) to push a right button at some time later than t. John’s decision is thus determined by the occurrence of neural activity of type B—a different choice could not occur given its occurrence and the truth of 1. An additional condition was that such previous neural activity was not under the agent’s control in any relevant sense. Without that, space for free will could remain available—for example, if the determination relation could only occur given a previous decision to participate in a particular experiment.

    How could these deterministic relations impact on libertarian free will? I have described two libertarian views that require choices that are not so determined. First, Chisholm’s view requires that, for every action, it can qualify as free only if it results from a choice that could have been different. On this view, choices determined by previous neural activity (one which, presumably, the agent could not have chosen not to have) in the specified sense would not be free. So, as it seems, neuroscience could have an impact on libertarian free will by showing that some actions and choices are not free in a libertarian sense. A second, more nuanced view is Kane’s. On this view, a determined action (in the sense at issue) may be free, provided that the causal chain that leads to it traces back to an action that originates in a choice that was not determined and, thus, could have been different – those are called ‘self-forming actions’ (SFAs). Here, evidence for determination according to statements of local determination would not directly show that a choice is not free in a libertarian sense, but only that it does not originate an SFA.

    At this point, both Nahmias and Roskies contended that statements of local determination would not be enough to affect positions such as Kane’s. My original suggestion was that “the discovery of more and more [statements of local determination] could decrease our confidence in the existence of SFAs, and, indirectly, decrease our confidence in the existence of libertarian free will.” Here is what they say:

    “I wonder whether neuroscience is more likely to pose problems for Kane by discovering lots of deterministic relations like LD1 or instead by inductively concluding that quantum indeterminism never has influences on neural interactions, or gets ‘washed out’ at that level, a discovery that would surprise me” (Nahmias)

    “Fischborn suggests that finding enough local determinisms will decrease our confidence in the existence of Kane’s SFAs, and thus in Libertarian free will. Probably not. If one expects SFAs to be rare, and brain patterns almost infinite, then a good Bayesian will agree that seeing some local determinisms will not decrease credence in SFAs very much. […] Thus, I maintain that to truly undermine Libertarianism, you will need more than just evidence for instances of law-governed activity. You will need reasons to think that indeterministic events do not affect brain activity, and thus for a global determinism (extending at least through the brain, if not beyond).” (Roskies)

    I must agree with these observations. In order to have a stronger impact on the existence of choices that are free in Kane’s libertarian sense, neuroscience would need to support some stronger deterministic hypothesis. Here is a new attempt to capture what would be enough:

    Hypothesis of the deterministic nature of choices (DNC): For any subject s, any choice x, and any course of action X, if s chooses to do X, then there is a previous event y of a type Y in s’s brain, such that whenever an event of type Y occurs in someone’s brain, then this subject chooses to do X.

    DNC is a generalization over the idea of statements of local determination. What it says is basically that, for every choice, there is a statement of local determination that is true of that choice and, thus, that every choice is determined. Here, again, the determining factor must not be under the agent’s control. And now I think DNC—if true—would show that no one is free in a libertarian sense in both Chisholm’s and Kane’s view. In addition, I think that relevant statements of local determination—should they be discovered—could plausibly provide some inductive (although fallible) support for DNC.

    As I argued, available results from Libet-style experiments fall short of establishing statements of local determination, and should also fail to support DNC. I think Nahmias and Roskies will agree with that, and also that no compelling evidence for something like DNC is so far available. My question now is whether I can plausibly maintain my claim that neuroscience could in principle have an impact on libertarian free will, given the new focus on the more demanding DNC. In the remainder of this reply, I argue that commentators’ contentions only appear to conflict with my claim, but in fact are compatible with it.

    Let me start with a clarification. When I ask if neuroscience could in principle provide support for DNC or simpler statements of local determination, I am not asking primarily if it is plausible or likely that neuroscience will do that some day. What interests me is whether neuroscience, given their nature, is a suitable source of evidence (or lack thereof) on these issues. (As I note below, neuroscience may remain relevant on these issues even if not exclusively, but, say, in combination with other sciences). If that is the case, as I think it is, then even the absence of evidence for the relevant determination gives us some reason to think that our choices are undetermined. (Compare: astronomy does not give us evidence that our choices are determined. But that does not support the claim that our choices are undetermined because astronomy is in principle unsuited for deciding that issue.) Nahmias and Roskies present reasons suggesting that neuroscience would never support the kind of determination I previously characterized. I think these reasons are compatible with the claim that neuroscience could in principle support that.

    Complexity of the brain system underlying choices: Both Nahmias and Roskies point to the complexity of the brain processes underlying choices as a reason for doubting the possibility of establishing LD statements:

    “the complexity of the human brain suggests that any complex decision will involve such a wide disjunction of neural processes that there will be no pattern of neural activity of a given type that can be picked out as the cause of that decision.” (Nahmias)

    “When we take into account that each neuron in the brain makes approximately 10,000 connections to other neurons, and is affected by as many, that there are multiple stages of processing, interactions with inhibitory interneurons, and crosstalk with other circuits for any action we might take and decisions we might consider, we see that local determinism would have to be fairly global to correspond to decisions, choices, actions, beliefs, or what have you. […] any neural system rich enough to subserve the process of decision-making and issuance in action will be too complex and interconnected to be the seat of local determinism. Alterations at any of multiple levels of processing could change the decisions; other thoughts, sensory inputs, or even noise could lead to a person’s doing otherwise. There will be a requirement of global knowledge to establish determinism in a system, and global knowledge of that sort is unattainable.” (Roskies)

    Ok, the neural basis for choices is quite complex. But how do we know about that? Certainly this knowledge partly derives from what neuroscience itself tells us. If that makes LD statements unlikely, it also shows that neuroscience has a bearing on the question whether choices are determined or not. The alleged unlikelihood that those statements are true is not proof that they are in principle unsupportable by neuroscience. On the contrary, it derives from the way reality is according to neuroscience. If reality turned out to be different, then neuroscience could tell us that too – because, in principle, it could do that.

    Multiple-realizability: Nahmias argues that “it is not at all clear that we could generate infinitely many statements [of local determination]”. He says that “if multiple realizability is true, then there will not be LD statements that apply across all brains or even within the same brain over time”. I think Nahmias’ remark is correct about statements of local determination as previously characterized. However, in order to know if the antecedent of Nahmias’ claim is true, we turn again, at least in part, to neuroscience. How would we know if brain states of the same type realize different types of psychological states in different individuals or within the same individual at different times? That multiple-realizability is true (and that this makes LD statements unlikely) is consistent with statements of local determination being in principle supportable by neuroscience. (I am inclined to think that statements of local determination relativized to individuals, or even to individuals at different periods of their life, could also preclude their ability to choose otherwise and, thus, conflict with the libertarian views here considered.)

    Chaos: Relatedly to the point about the complexity of brains, Nahmias mentions chaos as a further component that would make the truth of LD-statements unlikely: “Throw in chaos theory (non-linear dynamics) and seemingly insignificant changes to any token pattern of neural activity might have very significant effects on the ensuing decision or behavior.” In such a case, he says, “the likelihood of there being law-like deterministic relations among complex neural states becomes vanishingly small.” First of all, I think the challenge posed by chaos is mostly epistemic because the very idea of chaos is deterministic. It says that minimal differences in the initial conditions of a system may lead to big differences on its subsequent states; but if the initial state is exactly the same, then the subsequent states will be exactly the same too. In practice, it probably will be difficult to distinguish (deterministic) chaotic systems from indeterministic ones. Anyway, if neuroscience can support the idea that choices happen according to a chaotic dynamics, then, by the same token, it will also have something to say about the (in)deterministic nature of choices.

    Extended/embodied mind: Finally, Nahmias mentions theories of extended/embodied mind as a further impediment for the discovery by neuroscience of LD-like statements. If these theories are correct, “neural activity alone will not be sufficient to cause some decisions or behaviors. Hence, local determinism in the brain could not be generalized to some decisions and behaviors.” (Nahmias) If these theories are true, then neuroscience alone would not be able to establish LD-like statements. But perhaps neuroscience plus some other sciences will still have something relevant to say about whether choices are deterministic or not. I believe that such knowledge would still be relevant for assessing the existence of choices that are free in the libertarian senses considered.

    Nahmias’s mentions in addition that libertarians such as Kane only are incompatibilists because they previously endorse arguments that employ the idea of universal determinism (namely, Manipulation Arguments and The Consequence Argument). Statements of local determination by themselves, then, would be insufficient to motivate the belief that they are incompatible with free will. In other words, if there is an argument for incompatibilism that employs statements of local determination, then this argument is still to be presented. I agree with that, and for now have no argument to offer.

    To sum up: I agree that LD statements seem unlikely now, but I contend that this derives (at least in part) from the fact that neuroscience has in principle something relevant to say about their truth, and even about the stronger DNC. What neuroscience says now indicates that they are unlikely to be true.

    I thank Adina Roskies and Eddy Nahmias for their illuminating comments, and the organizers for making this opportunity to discuss these issues possible. I’m looking forward to continuing the discussion.

  4. Marcelo, thanks for your responses! Your new statement of ‘neuro-determinism’ is helpful:

    “Hypothesis of the deterministic nature of choices (DNC): For any subject s, any choice x, and any course of action X, if s chooses to do X, then there is a previous event y of a type Y in s’s brain, such that whenever an event of type Y occurs in someone’s brain, then this subject chooses to do X.”

    However, I’m still unclear how DNC could be true if multiple realizability is true. If you mean “type Y” to refer to whatever complex of individual neural events that causes X, then why mention ‘types’ at all–wouldn’t we just want to say: for every choice x there is some immediately prior physical state (largely or entirely neural) that is causally sufficient for x? Or do you mean “type Y” to refer to a disjunct of neural events (y1…yn), any one of which will cause X? But as Adina and I argued, if X is a choice or action, it is likely that this disjunct will be quite large, especially if you mean for ‘s’ to refer to any human who can make choices of type x.

    I guess I’m not sure how DNC responds to the concerns we raised about LD.

    I’d like to think more about how, if DNC were true (perhaps because multiple realizability, etc. are false), it might pose a challenge to free will–that is, why should we be incompatibilists about ‘neuro-determinism’ and free will if we were not already incompatibilists? Perhaps others can provide some ideas about this.

    1. Thanks again for the comments.

      I should have been explicit that I agree that both LD-statements and DNC cannot be true if multiple-realizability is true. On this point, my central question is: Has neuroscience anything to say about the truth of multiple-realizability? If it has, then I think it still has something to tell about the truth of LD-statements and DNC.

      I also think that we can proceed without mentioning ‘types’. I think the crucial question is whether the occurrence of some set of neural conditions/processes (at least in principle repeatable, and not under the agent’s control) can be a sufficient condition for the occurrence of one specific type of choice (say, to do A rather than not-A). This kind of determination, moreover, seems compatible with multiple-realizability. Let me try to present an example:

      N1: set of neural conditions.
      N2: set of neural conditions.
      C1: choice to A.
      C2: choice not to A.
      t1: some specific time
      t2: some specific time after t1.

      Suppose that, due to multiple realizability, mental events can be realized by different neural realizer in different subjects and at different periods of their like. Now suppose that these statements are true:

      For every subject at any instant of time, the occurrence of N1 entails that N2 will also occur.
      Within t1-t2, N2 realizes C1 in John’s brain.
      Within t1-t2, N1 occurs in John’s brain.
      Therefore, C1 will occur in John (that is, John will decide to A).

      If these conditions obtain, I would say that John’s decision to A was determined by previous neural activity N1, that is, that a different choice could not have resulted from the occurrence of N1, given 1-3.

      Again, I don’t think we have good reasons to believe in this kind of determination, but I think that neuroscience has in principle something to say about these issues. If if hadn’t, perhaps we would not find so many reasons to doubt the possibility of establishing LD-statements, DNC, and 1-3.

  5. Thanks Marcelo and Eddy for your comments.

    I’d like to make a clarification about chaos. Chaotic behavior can be deterministic or indeterministic: that is, it can be underlaid by deterministic events, or some indeterministic ones. But that means that if neuroscience gives us evidence of chaotic processes in the brain, we cannot from that infer that the underlying processes are deterministic. So if that is our question, merely identifying chaos in neural systems will not serve to answer that question.

    On other matters, I am unsure, given what Marcelo has added, how to take his response to Eddy’s worry about types. Certainly, for some event token C1 there will be some sufficient cause, N2, and there will be a sufficient cause of that, N1. But if we are giving up on the notion of laws governing types, what we are dealing with now is just causation. So now the determination worry seems to boil down to the simple question of whether there are indeterministic events in the brain, which I have argued is not a question neuroscience is suited to answer.

    Of course, what “suited to answer” means can vary among us. I took Marcelo to be arguing that neuroscience could in principle give us evidence for local determination that would be compelling enough to provide reason to doubt some Libertarian view. But in Marcelo’s response he seems to have something much weaker in mind: if neuroscience tells us about the brain such that we have reason to believe it cannot give us evidence for local determination, then it is still in principle able to provide evidence of local determination. I suppose he says this because that evidence is negative. But the evidence is not really negative – it is just not really evidence. What he seems to suggest is that anything that told us about the brain (as opposed to, say, the stars, given Marcelo’s example of astronomy) is relevant to understanding the brain. But while true this is trivially true, it doesn’t help us make progress on the question we began with. I think it is clearer to say that neuroscience cannot provide evidence that bears on the truth of Libertarianism.

  6. Thanks Adina for your comment.

    I remain unsure about how to understand some of your points. The key question, it seems to me, is whether neuroscience (broadly constructed, so as to include cognitive neuroscience) can provide some sort of support for claims of the following sort:

    Under specifiable neural conditions, subjects are not able to choose to do A rather than not-A.

    I don’t think supporting this kind of claim necessarily involves knowing whether processes that underlie brain processes are deterministic or not. And I agree that merely supporting this kind of claim may not motivate the thought that it conflicts with free will. But the fact remains that some libertarian views of free will assume the falsity of some such claims, and I wonder what kind of scientific investigation might be able support or falsify them. I remain in doubt that this is a matter (just) for physicists.

    Anyway, I surely will continue to reflect on the points that both Adina and Eddy have raised, and I thank again for the opportunity to discuss these matters here.

Comments are closed.