Assimilations and Rollbacks: Two Arguments Against Libertarianism Defended

Abstract

The Assimilation Argument purports to show that libertarians cannot plausibly distinguish supposed exercises of free will from random outcomes that nobody would count as exercises of free will. If this argument is sound, libertarians should either abandon their position or else concede that free will is a mystery. Drawing on a parallel with the Manipulation Argument against compatibilism, Christopher Franklin has recently contended that the Assimilation Argument is unsound. Here I defend the Assimilation Argument and the Rollback Argument, a second challenge to libertarianism that Franklin rejects. My aim in doing so is to underscore the force of these challenges, and thereby to resist what appears to be an emerging trend in the literature. By not coming to grips with the kind of power that libertarians must secure, many writers have recently downplayed or dismissed the pressing worry for libertarianism.

Libertarians believe that free will is incompatible with causal determinism and that we sometimes exercise free will. An influential challenge to libertarianism has come to be known as ‘the Luck Objection.’ This objection (or family of objections) targets the libertarian’s core view that there are causally undetermined exercises of free will. If prior to an action’s occurrence, it is causally undetermined whether or not the agent will perform that action, then it isn’t up to her whether it occurs; its occurrence is a piece of luck, good or bad as the case may be, for which no praise or blame is merited.¹ Or so the objection goes.

A recent argument in the spirit of the Luck Objection has been dubbed ‘the Assimilation Argument’ (Shabo 2011, 2013). The leading idea is that libertarians cannot plausibly distinguish supposed exercises of free will from randomized outcomes that nobody would count as up to us. If libertarians cannot plausibly resist the assimilation of the undetermined to the random, they cannot credibly maintain that some causally undetermined actions are exercises of free will, even supposing that some actions are causally undetermined.

While the Assimilation Argument has affinities with the Luck Objection,² it notably eschews the words ‘luck,’ ‘chance,’ and their cognates, and so bypasses questions about what these words mean and how the associated concept(s) should be analyzed (Shabo 2011, 106).

Here I defend the Assimilation Argument against a recent challenge. Christopher Evan Franklin (2012) contends that the argument is easily “dispatched,” as is the Rollback Argument, a more traditional version of the Luck Objection on which I rely. As I hope to show, both challenges to libertarianism emerge unscathed.

A survey of the recent literature reveals a growing tendency to downplay or dismiss the problem for libertarians embodied in the Luck Objection.³ By defending the Assimilation Argument and the Rollback Argument, I hope to reinforce the reasons for taking this family of objections seriously.

The Assimilation Argument and the Rollback Argument

As I shall understand ‘free will,’ to say that someone exercises free will in acting as she does is to say that it’s up to her whether she voluntarily and intentionally performs that action or voluntarily and intentionally does something else instead (or perhaps refrains from performing any “positive” action then). According to incompatibilists, only causally undetermined actions meet this condition. The Assimilation Argument purports to show that no causally undetermined actions do meet it (assuming that such actions exist), and that we therefore have good reason to reject libertarianism.

More specifically, the Assimilation Argument pertains to every action such that, immediately before it occurs, it’s causally undetermined whether it or some other action (or an intentional omission) will ensue. For convenience, unless otherwise indicated, in speaking of causally undetermined actions, I shall have in mind actions whose occurrence is undetermined by their immediate causal antecedents, where the agent’s voluntarily and intentionally “doing otherwise” is consistent with these antecedents together with the laws of nature.

Put simply, the Assimilation Argument goes this way:

Truly random outcomes are possible in principle. If it’s truly random which of two possible outcomes occurs, the actual outcome isn’t up to anyone. If the libertarian is to plausibly maintain that some causally undetermined actions are up to us, she must explain how such actions are crucially different from random outcomes. Since, however, the libertarian cannot plausibly explain this, she cannot plausibly maintain that people sometimes exercise free will.

As I have formulated it, this argument rests on an important assumption: that a causally undetermined human action could in principle be a truly random outcome and hence not up to the agent who performs it, just as other types of events, such as a particle’s swerving one way when another way was possible, or some radioactive material’s decaying at one rate when another rate was possible, could be random outcomes. Libertarians who take the Luck Objection seriously do not dispute this assumption; instead, they accept that it’s incumbent on them to explain how a causally undetermined action could be up to the agent and thus unlike such random outcomes. In short, the libertarian must explain how it’s within the agent’s power to settle which of the two causally possible futures is realized. On the assumption that an undetermined action could be a random outcome, we can consistently maintain that an undetermined outcome is an action while denying that it’s up to the agent whether or not that action occurs. Even if it were free from external and internal compulsion, such an action wouldn’t be an exercise of free will. Why this assumption matters will soon become clear.

My main contention is that if there are causally undetermined actions, they resemble uncontroversially random outcomes in that it isn’t up to anyone whether or not they occur. To support this claim, I provide a series of cases involving causally undetermined outcomes. The first case doesn’t feature an agent at all and the outcome clearly isn’t up to anyone. The last case features an action that libertarians should be able to count as an ordinary exercise of free will. If the argument succeeds, libertarians cannot suitably differentiate the last case from some or all of its predecessors. That is, they cannot explain how the outcomes in this case could be up to the agent when the outcomes in some or all of the other cases aren’t.

Here are the cases (2013, 301–302; cited in Franklin 2012, 399–400):

• Case 1: A particle in a remote region of spacetime has an objective, .5 probability of swerving one way, and an equal probability of swerving another way. No other event in the universe influences this probability distribution or which way the particle swerves, given this distribution. The particle ends up swerving in the first of the two possible ways.

• Case 2: A similar particle in a non-remote region of spacetime will swerve in one direction or another, with each outcome having an equal, objective probability. Scientists are using a device to detect the particle’s behavior. If the particle swerves one way, the device transmits one signal to a receiver implanted in Alice’s brain; the receiver then causally determines Alice’s brain to go into a state that is intrinsically indistinguishable from an ordinary, executive intention to tell the truth in response to the question she has been asked. If the particle swerves the other way, the device transmits a different signal to the receiver, which causally determines Alice’s brain to go into a state intrinsically indistinguishable from an ordinary, executive intention to lie. In the event, the particle swerves in the first direction, with the result that Alice exhibits truth-telling behavior, behavior that is caused in a normal way by the relevant neural state.

• Case 3: This case is exactly like Case 2, except that the device has been miniaturized (swerving particle and all) and placed in Alice’s brain, where it induces one neural state or the other, eliminating the need for a receiver.

• Case 4: The scientists introduce an upgraded device. The upgraded device harnesses indeterminacy in Alice’s own neural pathways, indeterminacy that must be present if her action is to meet the libertarian’s conditions for free will. The upgraded device can target pathways that are causally inert, pathways that are involved in reflex-type behavior, or pathways that are involved in the formation of intentions. When the pathways in the last of these regions are targeted, the upgraded device charges a particle (or group of particles) in that pathway, causing that particle(s) to behave like the particle in Case 3. That is, the particle(s) will swerve one way or another, with each outcome having an equal, objective probability. In this case, however, the outcome is an intention to lie or an intention to tell the truth—or at least a neural state that is intrinsically indistinguishable from one of these intentions or the other.

• Case 5: There is no input from any external device. Due to its ordinary, endogenous workings, Alice’s brain state as she mulls her options is such that lying and telling the truth are equally probable outcomes. She intends to make up her mind, and this intention causally contributes to her deciding to tell the truth, a decision that is based on her reasons for doing so.

Since the outcomes in Cases 1–3 aren’t up to anyone, the idea is, maintaining that the outcome in Case 5 is up to Alice will require the libertarian to explain how Case 5 is crucially different from Case 4, or, alternatively, how Cases 4 and 5 are crucially different from Case 3. That is, the libertarian must identify a feature of Case 5 (or of Cases 4 and 5) in virtue of which it seems plausible to say that the outcome is up to Alice in this case(s), given that the outcomes in the previous cases aren’t up to anyone.

I contend that it won’t be enough to observe that Alice acts in Case 5 (or in Cases 4 and 5) while there is no action in the preceding cases (2013, 302–03; cf. Shabo 2011, 19–20). This is because, on the assumption emphasized above, it’s possible for an undetermined action not to be up to the agent who performs it. Just as supposing that an action is undetermined is no guarantee that it’s up to the agent, so supposing that an undetermined event is someone’s action is no guarantee that its occurrence is up to that person. In short, there is a “logical gap” between the premise that an agent-involving event is undetermined and the conclusion that it’s up to the agent whether it occurs, a gap that isn’t closed by adding that the event is something the agent does (Shabo 2013, 301). Undoubtedly, adding this further premise is a step in the right direction, but given the above assumption, it isn’t enough. If the outcome in Case 5 (or in 4 and 5) is up to the agent in contrast to the outcomes in the preceding cases, this isn’t simply in virtue of the Case 5-outcome being an action.

A modified version of van Inwagen’s Rollback Argument shows why it isn’t enough for a libertarian to maintain that Alice acts in Case 5 but not in Case 4. In the modified version, we imagine that

…God sets up two cosmic video displays side by side. The first shows replays of the original Case 4; the second replays the original Case 5. Each display is programmed to repeat a five-second clip of the run-up to Alice’s decision, replete with stages that are causally undetermined by their predecessors. Now suppose that we are able to zoom in—way in—on the action. On each display, we see images of neural activity in the deliberative centers of Alice’s brain. On the first monitor, we see a graphical representation (in a cloud of red and blue dots) in these neural centers as Alice mulls her options; then the upgraded device emits an energy pulse, which is immediately followed by a particular activity signature (always the same) in the affected pathways, and then by one decision “node” or the other “lighting up.” Events on the second display unfold in much the same way, except that the device is absent. First we see the same activity in Alice’s deliberative centers as on the first display; then the same (or a seemingly indistinguishable) activity signature, only without the mediation of the device, followed by one decision node or the other lighting up.

After 16,027 dual replays, we will have seen Alice deciding to lie about 16,027 times and to tell the truth about 16,027. Looking from one display to the other, we will observe no difference in the activity signature that immediately precedes Alice’s decision. Nor, focusing on either monitor [individually], will we discover any difference in Alice’s neural state immediately before she makes the one decision or the other. In sum, here are the only differences we will have observed: the pulse is always emitted on the first display and never on the second; and (on each display) sometimes one decision node lights up and sometimes the other. What doesn’t vary is the neural activity signature immediately prior to a node’s lighting up; this is constant through the combined 32,054 replays. (2013, 304; cited in Franklin 2012, 405)

In Case 4, we are supposing arguendo, the neural event doesn’t qualify as a genuine action, owing to its history; instead, it merely resembles an action in its intrinsic qualities. By contrast, the neural event in Case 5 is a genuine mental act, something Alice does. On both screens, immediately before the undetermined outcome occurs, Alice’s brain is (intrinsically speaking) in exactly the same state, which is sometimes followed by the one outcome and sometimes the other. If the outcome isn’t up to Alice in Case 4, if the only difference between cases 4 and 5 is that the outcome in Case 5 counts as a mental act, and if this outcome can (like the outcome in Case 4) fail to be up to Alice, this difference seems insufficient. The supposition that Alice acts in Case 5 but not in Case 4 doesn’t plausibly support the claim that it’s up to Alice in Case 5 which choice she makes.

Franklin’s Critique

Franklin’s critique of the Assimilation Argument involves three separate claims. The first is that Alice’s acting in case 5 but not in Case 4 does adequately differentiate the two cases, even if the logical gap—the gap between “Alice performed an undetermined action” and “It was up to Alice whether or not she performed that action”—remains unfilled. The second is that libertarians are able to close this gap, even if an adequate response doesn’t depend on closing it. The third is that it’s a mistake to see the Rollback Argument as even a presumptive problem for libertarians, contrary to what is generally supposed; if this is right, this argument will be unable to play the supporting role I assign it. I address Franklin’s first two contentions in this section and his response to the Rollback Argument in the next.

Must the Libertarian Close the Logical Gap?

Franklin’s main criticism targets my contention that it isn’t enough for the libertarian’s purposes to distinguish the case(s) in which Alice acts from the ones in which no action occurs. Franklin believes that this is enough to establish a sufficient difference and thereby defuse the threat from the Assimilation Argument. In Franklin’s view, the most auspicious place to draw the line is between Cases 4 and 5.

Franklin argues that it’s enough to maintain that Alice acts in Case 4 but not in Case 5. To underpin this alleged difference between the two cases, Franklin appeals to a familiar theory of action on which something counts as an action only if it’s appropriately caused by the agent’s relevant mental states, such as her “beliefs, desires, reasons, and intentions” (396–97).⁴ Given the device’s intervention in Case 4, Franklin observes, libertarians can plausibly deny that the states that indeterministically cause the neural outcome are Alice’s mental states (however intrinsically similar they may be to their counterparts in Case 5), and thus that this outcome and the behavior it produces count as Alice’s actions (2012, 400–01). By contrast, the outcome in Case 5 has the right causal history to be something Alice does.

To repeat, I expressly allow that Cases 4 and 5 might be claimed to differ along these lines. What I deny is that this difference is by itself sufficient to meet the challenge. I write: “From the fact that the causally undetermined outcome in Case 5 is Alice’s decision to tell the truth, a mental act she performs for reasons she has, it doesn’t follow that she settles which outcome ensues, or which set of reasons she acts on” (Shabo 2013, 303). Since the supposition that Alice acts in Case 5 but in Case 4 fails to explain how she can settle which act she performs, it doesn’t meet the challenge posed by the Assimilation Argument.

Franklin disputes this assessment. In his view, I have merely given readers “the old bait and switch”:

Shabo begins by asking one question, “What is the relevant difference between cases 4 and 5?” but criticizes our [anticipated] answer because it fails to answer a different question, “In virtue of what does Alice exercise the power to settle in case 5?” Shabo appears to think his switching questions is justified because he assumes that an adequate explanation of the difference between cases 4 and 5 “would plausibly close the logical gap identified [above].” In other words, Shabo assumes that an answer to the first question is satisfactory only if it is a satisfactory answer to the second question. But this is not the case—answers to these questions can and do come apart. (Franklin 2012, 403)

There is no bait and switch. By ‘the relevant difference’ (and ‘the crucial difference’), I just mean ‘the difference in virtue of which Alice has the power to settle the outcome in Case 5 but not in Case 4.’ Thus, pace Franklin, the answers to these questions cannot come apart.

In any case, the important question is whether Franklin is right that I have set the bar too high for a successful libertarian response. Is it enough for libertarians to say that Alice acts only in Case 5, so that she satisfies a prerequisite for free will that she doesn’t satisfy in the other cases?

Franklin believes so. He writes:

The difference between cases 4 and 5 is not just some minor detail, but is a free-will-relevant difference. An agent exercises free will only if he acts. Since Alice does not act in case 4 we know that she does not exercise free will in case 4; but Alice does act in case 5 and so the obstacle to Alice’s exercising free will in case 4 is absent in case 5. The difference between these cases consists in the fact that a necessary condition for exercising free will is absent in case 4 but present in Case 5. So we have arrived at an explanation of how overtly randomized outcomes differ from exercises of free will: an overtly randomized outcome is not an action, while an exercise of free will must be an action. (401)

Let us continue to grant that Alice doesn’t act in Case 4 because the neural outcome doesn’t have the right causes, so that it doesn’t meet a prerequisite for free will that is met in Case 5. Is this difference sufficient to answer the Assimilation Argument?

Franklin’s defense of an affirmative answer involves comparing the Assimilation Argument with the Manipulation Argument, an influential argument against responsibility-determinism compatibilism. In standard versions of the Manipulation Argument, the agent in the initial case satisfies a host of influential compatibilist conditions for moral responsibility, notwithstanding that neuroscientists have covertly manipulated her motivational states in an intuitively responsibility-undermining way.⁵ The challenge for compatibilists⁶ is to explain how this action differs from ones with ordinary deterministic histories. Since the actions in the first and last cases appear to satisfy the same compatibilist conditions, it’s hard to see what resources compatibilists have for differentiating the two; or so the Manipulation Argument aims to show.

Franklin provides this schema for the Manipulation Argument (404):

• 1. Agents who are severely manipulated are not free.

• 2. There is no relevant difference between “ordinary” deterministic agents and severely manipulated agents.

• 3. Therefore, deterministic agents are not free.

As he correctly observes, in responding to an instance of the Manipulation Argument, it isn’t incumbent on compatibilists to show that ordinary deterministic agents are free (in the sense required for moral responsibility), only that the reasons for thinking that manipulated agents aren’t free don’t transfer to them (ibid.). If it can be shown, for example, that the manipulated action doesn’t satisfy the compatibilists’ conditions after all (say, because the nature and extent of the manipulation compromises the individual’s status as a person or agent),⁷ this will suffice; for compatibilists can deny that this case has any bearing on cases where the compatibilists’ conditions are met.

Franklin takes the fact that the two arguments are “structurally similar” (ibid.), and that such a response (assuming it is otherwise credible) will thwart the Manipulation Argument, to show that the Assimilation Argument admits of a similar response. Like the Manipulation Argument, he writes, the Assimilation Argument

…depends on a premise concerning the relevant similarity between the cases presented, and so it, like the Manipulation Argument, fails if there is a free-will-relevant difference between the cases…Hence, in order to defeat the Assimilation Argument it is sufficient to show that Alice acts in Case 5 but not in Case 4… (404)

I don’t see this. Instead of thinking that what works against the one argument should work against the other, I’m inclined to see the two arguments as less similar than Franklin supposes. Here are two key differences.

First, the success of the Manipulation Argument straightforwardly depends on the compatibilist’s conditions for free action being satisfied in the cases where manipulation is present and the cases where it’s absent. Thus its success depends on the agent’s acting—and thereby meeting a precondition for free action—when the agent is manipulated. If compatibilists can show that their conditions aren’t met when the agent is manipulated, they can deny that the agent’s lack of freedom generalizes to cases without manipulation where these conditions are met; and showing that the manipulated individual fails to act (and so doesn’t meet a precondition for moral responsibility) a way to show this. For this reason, showing that there is no action when the agent is manipulated effectively undermines the Manipulation Argument. (More carefully, it shows that the particular version of the Manipulation Argument under consideration doesn’t succeed against the particular compatibilist account to which it is addressed.)

The dialectic with the Assimilation Argument is importantly different. In particular, there is no comparable reason to suppose that this argument’s success depends on Alice’s meeting this precondition for free will in Case 4. Why think, then, that denying that Alice acts in Case 4 is an effective response to this argument? As we have seen, the argument’s defender aims to show that libertarians cannot adequately explain how the outcome in Case 5 could be up to Alice. The fact that Alice acts in Case 5 clearly isn’t enough to explain this, even though it does make the outcome in Case 5 a candidate for free will. If we want to understand how this outcome could be an exercise of free will and not just a candidate, denying that the outcome in Case 4 is even a candidate for free will won’t help. Thus, while Franklin is correct that there is a free-will-relevant difference between Cases 4 and 5 (granting that Alice doesn’t act in Case 4), it isn’t the right kind of difference to answer the Assimilation Argument, as emerges on a closer look at the relationship between the Assimilation Argument and the Manipulation Argument.

I turn now to the second and deeper difference between the two arguments, which we have already touched upon in the preceding paragraph. This second difference concerns the problem each argument is intended to bring out. Among committed libertarians, it has long been widely (though not universally) accepted that there is a real question about how a causally undetermined action could be up to its agent. As Robert Kane (1996) puts it, libertarians face an “intelligibility” problem, a problem of showing how we can make good sense of the thought that an action is both undetermined by its immediate causal antecedents and an exercise of the agent’s free will. Kane has cataloged a variety of exotic expedients to which libertarians have appealed in the hope of grounding this thought:

In this vein, libertarians have traditionally invoked factors such as noumenal selves; transempirical power centers; Cartesian egos; special ‘acts of will’; ‘acts of attention’ or ‘volitions’ that cannot in principle be determined; mental events that can exist outside of (but can also intervene within) the natural order; special forms of agent, or nonoccurent, causation that cannot be accounted for in terms of familiar kinds of event or occurrent causation; the ‘Will’ conceived as a kind of uncaused homunculus within the agent; and so on. How these special causes or agencies are supposed to operate, or how they are supposed to explain why one choice occurs rather than another, is never well enough explained to allay suspicions that libertarian accounts of free agency are mysterious. (1996, 115)

Kane believes that this problem ultimately has a satisfying solution, one that renders libertarianism intelligible while (not coincidentally) avoiding these expedients. But he doesn’t suggest that the intelligibility problem is misconceived; in his view, it needs to be solved, not dissolved or dismissed. Another committed libertarian, Peter van Inwagen, takes a less sanguine view. Convinced that the problem is intractable, van Inwagen reluctantly concludes that free will is a mystery (2002, 2011). While there may be differences in exactly how these (and many other libertarians) understand this problem and the prospects for solving it, they accept that moving from ‘It’s causally possible both that someone chooses to do A and that she chooses to do B instead’ to ‘It’s up to that person whether she chooses to do A or to do B instead’ isn’t plain sailing. I have suggested that the problem is aptly characterized in terms of the power to settle the outcome. Once we have taken into account all of the agent’s proximal, causally relevant mental states, it’s hard to see in virtue of what the agent could be said to settle which of the causally possible outcomes ensues. At this stage, there seems to be nothing more that she contributes to the outcome, even supposing that the outcome is something she does.⁸ The point of this observation is not that the problem is insoluble (though it may well be), but only that the need for a solution has traditionally been widely recognized. The problem, which is particular to libertarianism, is that it’s hard to state conditions whose satisfaction makes it intelligible how it could be up to someone which of two causally possible actions she performs. By contrast, the concern for compatibilists—the concern raised by the Manipulation Argument—is not about the intelligibility of being morally responsible for an ordinary, determined action. The concern is, rather, that the compatibilists’ conditions are arguably too weak, being satisfied even when, as in manipulation cases, the agent seems not to be morally responsible for his action.⁹

While Franklin is right to deny that successfully answering (a version of) the Manipulation Argument requires showing that an unmanipulated agent is morally responsible for her action, then, he is wrong to suppose that libertarians can adequately answer the Assimilation Argument without explaining how the outcome in Case 5 could be up to Alice. It isn’t enough to say that Alice acts, or that she acts for her reasons; for we can grant this while still wondering how she can be said to settle which action she performs. Thus, pace Franklin, it is a problem for libertarians that the logical gap identified earlier remains.

Summing up, Franklin’s analogy between the Assimilation Argument and the Manipulation Argument fails for two reasons. First, showing that the manipulated agent doesn’t act is a sufficient answer to the Manipulation Argument because this argument straightforwardly depends on the manipulated behavior’s meeting the compatibilist’s conditions for moral responsibility, and because a precondition for doing so is that the manipulated behavior is an action. By contrast, there is no comparable reason to suppose that the Assimilation Argument depends on Alice’s acting in Case 4. Second, the Manipulation Argument and the Assimilation Argument raise different worries for their respective targets. The Assimilation Argument uses Cases 1–4 to underscore the fundamental difficulty for libertarians of explaining how a causally undetermined action could be an exercise of free will, whereas the Manipulation Argument is meant to show that even manipulated agents who aren’t morally responsible for their actions meet the compatibilist’s conditions for moral responsibility, so that these conditions are too weak. If the manipulated agent doesn’t meet these conditions after all because he doesn’t act, the Manipulation Argument fails to show that these conditions are too weak. By contrast, showing that Alice doesn’t act in some or all of Cases 1–4, so that she doesn’t meet a precondition for free will in those cases, doesn’t resolve the pressing question for libertarians, the question of how we can make sense of its being up to Alice what she does in Case 5, as opposed to her action’s merely being another random outcome.

Franklin’s contention that the Assimilation Argument can be answered without closing the logical gap—the gap between “Alice performs a causally undetermined action for reasons that she has” and “It’s up to Alice which causally undetermined action she performs”—involves an analogy with the Manipulation Argument. Far from supporting his contention, close attention to the relationship between these arguments underscores how different they are. Insofar as the Assimilation Argument challenges libertarians to make it intelligible how a causally undetermined action could be up to the agent, an adequate response to this challenge will require libertarians to explain how the logical gap can be closed.

Let us now set the Manipulation Argument aside and focus directly on Franklin’s claim that it’s enough to dispatch the Assimilation Argument to maintain that Alice acts in Case 5 but not in Case 4. If there is a problem about how Alice’s choice in Case 5 could be up to her, a problem that is brought into relief by the intrinsic similarities between Alice’s neural events in Cases 4 and 5, the absence of a choice (or action) in Case 4 simply won’t be the kind of difference that resolves this problem. To resolve this problem, such a difference must close the logical gap. I turn now to Franklin’s claim that it can be closed.

Can the Libertarian Close the Logical Gap?

I have argued that the supposition that Alice acts only in Case 5 isn’t sufficient to undermine the Assimilation Argument. Even granting this supposition, the argument’s opponent must still identify the crucial difference between this action and the outcome in Case 4: the difference in virtue of which the former qualifies as up to Alice. The fact (as we’re supposing it to be) that Alice doesn’t act in Case 4 is a clear disqualifier. But to notice the absence of this disqualifier in Case 5 is not to explain how the outcome in Case 5 does qualify. Such an explanation would meet the challenge from the Assimilation Argument by closing the logical gap.

While Franklin denies that an adequate response must close this gap, he believes that it can be closed all the same. To show how, he notes a further presumptive difference between Cases 4 and 5:

Given the role of the upgraded device in case 4, Alice will not act either when the device causes her to “choose” to lie or to “choose” to tell the truth. The device appears to rob Alice of the ability to choose to lie and the ability to choose to tell the truth, at least for the time being. After all, given the workings of the device it is impossible for Alice to choose to tell the truth or choose to lie…In case 5, however, Alice acts (she chooses to tell the truth) and so presumably she also has the ability to tell the truth…And according to the libertarian theory I offered above, these differences are sufficient to close the logical gap. I claimed that an agent exercises his power to settle by making a decision, possessing the ability to make that decision, and possessing the ability to do otherwise. In case 5 Alice satisfies all these conditions… (405)

I want to focus on two of Franklin’s claims: that in Case 5 Alice is able to tell the truth and able to lie, and that the co-presence of these abilities is all there is to the power to settle the outcome.

Either of these claims might be disputed. Someone who accepted the second claim might deny that Alice is able to tell the truth or to lie, much as she is unable to intentionally role a 3 or a 6 with a fair die (cf. van Inwagen 2011, 476). While she will indeed tell the truth or lie, she lacks the ability to settle which act she performs if the outcome is undetermined by her proximal mental states.

On the other hand, someone who was prepared to say that Alice is able to lie and able to tell the truth, on the grounds that either action would constitute an exercise of the relevant ability, might deny that the power to settle the outcome is reducible to these abilities (cf. Shabo 2011, 9). Here the thought would be that the power to settle requires something more than each of these abilities, namely the further ability to determine which of them is exercised.

Of course, one might dispute both of Franklin’s premises. However, we needn’t commit to one of these responses to cast doubt on whether he has closed the logical gap. Consider his claim that Alice is able to tell the truth (or to lie instead) in Case 5. What does this claim amount to? More specifically, what is (b) supposed to add to (a)?

• (a) Alice decided to tell the truth, as it was causally possible for her to do.

• (b) Alice decided to tell the truth, as she was able to do.

If (b) adds nothing to (a), the logical gap remains, for as we have seen, (a) doesn’t entail that Alice’s decision was up to her.

Franklin doesn’t explicitly say what he takes (b) to add. What he does say involves a further contrast with Case 4. As noted, Franklin maintains that Alice doesn’t act in Case 4 because, owing to the device’s intervention, the resulting neural states don’t count as her desires, reasons, intentions, and so on, and hence the event that mediates between these states and her behavior doesn’t count as a mental act of deciding. In this scenario, Franklin goes on to say, the device’s intervention precludes the possibility of action on Alice’s part, a possibility that is present in Case 5 (p. 404). However, that it’s possible that Alice acts in Case 5 follows from the fact that she does act in Case 5. And that it’s possible that she tells the truth in Case 5 follows from the fact that she does actually tell the truth. But it clearly doesn’t follow that it’s up to her whether or not she tells the truth. (Cf. Franklin 2012, 414, n.46.)

To secure this further claim, something must be added about Alice’s abilities. At a minimum, this further something must make it clear how the ability in question goes beyond the mere possibility. Once we understand this, we can begin to assess whether the ability claim, together with the claim that Alice also has the ability to do otherwise, is logically sufficient for the outcome’s being up to her. What is clear is that not all abilities will do the trick. For example, Alice has the general capacities (or competencies) required for telling the truth and for lying, as well as for appreciating the moral and prudential reasons that support each option. But we can consistently acknowledge this, along with the fact that both actions are causally possible, while denying that the outcome is up to her. Hence the logical gap remains.

There is another way to make this point. Imagine a variant of Case 4 in which the device intervenes in a different way. In Case 4b, just before the device intervenes, Alice is torn between her reasons for lying and her reasons for telling the truth, and the device harnesses the indeterminacy in her neural pathways in such a way that one set of her reason states or the other is neutralized, with each outcome being equally probable. It’s plausible to suppose that, in the absence of the device, the insufficiency of each set of reason states is due to the “pull” of the countervailing reason states. If so, then rendering one set causally inert should make it the case that the other set, being unopposed, causally determines the corresponding choice.¹⁰

Suppose that the actual outcome in Case 4b is a neural state that leads Alice to tell the truth. The fact that the other reason states are neutralized isn’t a reason to deny that the causally effective states are indeed Alice’s reasons for telling the truth, or that her beliefs and so forth are the proximal causes of the neural outcome that produces her truth-telling behavior. Thus we have no reason in Case 4b to deny that this neural outcome is Alice’s mental act of deciding to tell the truth. Yet given that which set of reason states ends up being neutralized is random and not up to her, it isn’t up to her whether she chooses to tell the truth.¹¹

As in Case 5, it is possible in Case 4b that Alice choose to tell the truth and possible that she choose to lie instead. Unlike in Case 4 (as we are supposing for discussion’s sake), the device’s intervention does not preclude Alice’s choosing on the basis of her reasons to adopt the one course or the other. Recall Franklin’s claim that Alice is able in Case 5 to make the one choice and able to make the other instead. In Case 4b, just before the particle swerves inside the device, it’s possible that Alice choose to tell the truth and possible that she choose to lie instead, where each choice will exhibit the relevant agential competencies. Since it isn’t up to Alice which choice she makes in Case 4b, and since just before the device intervenes she has the same possibilities for action as in Case 5, the logical gap remains. Presumably, Franklin will say that Alice has additional abilities in Case 5. But until we are told what these abilities are and what difference they make, we cannot assess the claim that they close the gap.

Of course, there is this difference between Case 4b and Case 5. In Case 5, it remains possible up until the very moment that Alice chooses to tell the truth that she choose to lie instead. In Case 4b, by contrast, the choice to lie isn’t possible once the device neutralizes her reasons for doing so. (At that point, her reasons for telling the truth causally determine her choice to tell the truth.) But this difference doesn’t bear on the point of Case 4a. The point of this example is that there can be a time in the course of an agent’s deliberations such that (a) two alternative courses of action are causally possible relative to that time; (b) whichever possibility is realized, the agent will act (and thereby exercise various ordinary agential abilities); yet (c) which action the agent performs won’t be up to her then. Thus a logical gap exists between the premises (a) and (b) and the conclusion (c). In Case 5, the time relative to which (a) and (b) are true is the time immediately prior to Alice’s choice. But the logical gap between (a) and (b) (relativized to this different time) and (c) remains. This is not to dismiss out of hand the possibility that the temporal difference between the two cases might leave room for Franklin to make some further claim about Alice’s abilities in Case 5. The point of Case 4b is to show that, with respect to Case 5, Franklin has not yet succeeded in closing the gap separating (a) and (b) from (c).

Revisiting the Rollback Argument

I turn now to the final element of Franklin’s critique of the Assimilation Argument: his objection to the Rollback Argument, a variation of which (as we saw in “The Assimilation Argument and the Rollback Argument” section) I use to support my contention that Case 5 isn’t crucially different from Case 4. To repeat, the original Rollback Argument features a thought experiment in which we “roll back” the timeline until just before a causally undetermined action occurs, before letting events play out again. Suppose again that the undetermined action is Alice’s choice to tell the truth. If we repeat the experiment several hundred times, we will almost certainly see Alice tell the truth some of the time and lie some of the time, with nothing varying from one replay to the next except for which choice she makes. And we will very likely see the different outcomes converging on a definite ratio (say, 1:1), from which we will infer the objective probability of Alice’s telling the truth in each replay (say, 0.5). Moreover, we will have the strong impression that Alice’s choosing as she does in the next replay will be a mere reflection of this probability distribution—that is, a mere matter of luck or chance (van Inwagen 2002, 172). And with no grounds for dismissing this impression, we will have reason to conclude that Alice’s actual decision (of which the others are replays) is a mere matter of luck or chance as well.

According to Franklin, there is something odd about seeing the Rollback Argument as an objection to libertarianism in the first place. This is because he believes that the Rollback Argument merely identifies a constituent of the libertarian’s position. He writes:

The Rollback Argument does not draw out a consequence of libertarianism, but simply describes libertarianism and asserts that after understanding what this theory is, we ought to conclude that it is deeply problematic, or perhaps metaphysically incoherent. One cannot understand what libertarianism is without realizing that there will be observed variability from monitor to monitor.¹² To fail to appreciate this is to fail to understand indeterminism, and one cannot understand libertarianism without understanding indeterminism. Thus, to present the Rollback Argument is simply to describe libertarianism in a rather colorful way. But one cannot raise the cost of libertarianism by simply describing it. (409, second emphasis added)

Call the thesis that an undetermined action would exhibit this sort of variability across different replays (or at different possible worlds) in the thought experiment ‘The Varied Outcome Thesis’. More specifically, I shall understand the Varied Outcome Thesis to state that with a sufficiently large group of replays, we should expect to see a random-seeming sequence of outcomes, with the outcome distribution tending to converge on a definite ratio. In Case 4b (where the randomized device will neutralize one set of Alice’s reasons or the other, with even chances), the ratio toward which the genuinely random sequence will converge is 1:1. If we did not know in advance in Case 5 that Alice feels the pull of her competing reasons with equal force (as ex hypothesi she does), and that the two outcomes are consequently equally probable, we would infer this probability distribution upon seeing the different outcomes converge on a 1:1 ratio after hundreds of repetitions (as they very likely would).

Franklin contends that “one cannot understand what libertarianism is without realizing that there will be observed variability” of this general sort across the replays. I don’t see this. A libertarian who resisted this thesis might be fighting an uphill battle, but she wouldn’t thereby be exhibiting confusion about what libertarianism is or compromising her commitment to it.¹³ Indeed, one useful way to classify responses to the Rollback Argument is in terms of whether they reject the Varied Outcome Thesis or attempt to accommodate it.

The second approach has been more common in the literature, so I shall begin there. One instance of it involves maintaining that the force of the thought experiment depends on casting doubt on whether the agent’s choice has the right sort of explanation, an explanation for why she made that choice rather than the other one; and that free will doesn’t require such a “contrastive” explanation (O’Connor 2000, 91–93). A second instance begins the same way but maintains that undetermined actions do admit of contrastive explanations, or at least that we have no good reason to believe otherwise (Clarke 1996a, 2004).¹⁴ Still a third example does not focus specifically on contrastive explanation. This third response allows that enough replays of an undetermined action will (very likely) display a random succession of outcomes, while maintaining that this observation is innocuous; it simply falls short as a reason for libertarians to accept that each action in the series is itself a random outcome (Almeida and Bernstein 2011). A fourth instance claims that without special, agent-causal powers, each action in the series would be a random outcome, but that the presence of this power renders each outcome up to the agent, even if the succession is random and conforms to the Varied Outcome Thesis. This list is not comprehensive, but it provides a sense of how libertarians who have sought to accommodate what I’m calling the Varied Outcome Thesis have attempted to do so.

Turning to the first (and less common) approach, one recent instance of it involves rejecting a presupposition on which the Varied Outcome Thesis rests, namely that each possible choice has an objective probability. If the possible outcomes do not have objective probabilities, there is no reason to suppose that the undetermined outcomes will converge on a definite ratio or exhibit the random pattern we would otherwise expect; and so the thought experiment will lose much of its intuitive force.¹⁵ A second instance maintains that libertarians are vulnerable to this objection because they standardly suppose that free agency essentially involves a causally undetermined choice (or that some other discrete occurrence is the undetermined locus of responsibility), whereas the key to free agency is a suitably indeterministic process.¹⁶

I am mainly interested in the response that’s closest to Franklin’s, namely that the Varied Outcome Thesis simply falls short as a reason for libertarians to doubt their position. However, I will say something about agent-causal approaches as well.

Let us begin, then, with a libertarian response that embraces the Varied Outcome Thesis, as Franklin’s does, albeit without taking it to be part of the libertarian’s view. In a recent discussion, Almeida and Bernstein (2011) contend that the truth of this thesis shouldn’t worry libertarians: “We think the Rollback Argument should move no one to conclude that, in general, free will is incompatible with indeterminism” (485). As Almeida and Bernstein go on to say,

…[W]e reject the inference derived from the fact that Alice’s choices display a random distribution [across the successive replays] plus the fact that indeterminism is true to the conclusion that Alice was not able to tell the truth or was not free to tell the truth. Alice’s choice to tell the truth is part of a sequence of actions [across the successive replays] that display [a] random pattern in an indeterministic world. But it might also be true that Alice endorses her choice to tell the truth all the way up.¹⁷ If Alice endorses her decision to tell the truth all the way up, then we contend that Alice freely told the truth. (490)

I believe that such a random distribution is more problematic for libertarians than they (and Franklin) seem to realize.

Recall Case 4b, in which the neuroscientists’ device randomly neutralizes some of the neural states that embody Alice’s reasons to lie just before she acts. As it happens, it’s her reasons to lie that are neutralized and so she ends up telling the truth. Even after the effects of the device have worn off and Alice has learned of the intervention, she might well endorse her decision to tell the truth “all the way up” (it might be exactly what she hoped she would have done, without any reservations about her endorsement). And her decision and her act of telling the truth will exhibit her ordinary capacities to do these things. Finally, Alice will not have been coerced or compelled in any ordinary way to make this decision. Yet it won’t have been up to her which decision she made.¹⁸

Almeida and Bernstein might reply that Alice’s action in Case 4b is rendered unfree by the fact that her decision to tell the truth is determined by the reason states that proximally cause it. In Case 5, by contrast, the immediate antecedents of Alice’s decision underdetermine which decision she makes. What I wish to emphasize is that this doesn’t by itself allow a libertarian to claim that it’s up to Alice which decision she makes in Case 5, even if she endorses that decision “all the way up.” To see this, consider two presumptively possible worlds: one where each decision in the random sequence of replays really is up to Alice and one where each decision is truly random and not up to her. Let us add that in the second world, nothing marks Alice’s action as unfree in any garden-variety way: she is not forced to act against her will, she isn’t cognitively or morally incapacitated, she isn’t subject to an irresistible urge, she hasn’t been covertly manipulated, and so on.

Suppose that Almeida and Bernstein were to deny that the second world is genuinely possible on the grounds that Alice’s decision is up to her in the first world and that there is no relevant difference between the two worlds. This would be to deny our previous assumption that a causally undetermined action could in principle fail to be up to the agent (or at least that a suitably endorsed causally undetermined action could thus fail). But this seems incorrect, and Almeida and Bernstein have given us no reason to suppose otherwise.

Alternatively, suppose that Almeida and Bernstein were to say that there is a relevant difference between Alice’s decisions in these two worlds, such that the random distribution in the second world reflects the fact that each successive decision Alice makes is itself a random occurrence, whereas the random distribution in the first world reflects Alice’s sometimes exercising her free will in one way and sometimes in the other. That is, the metaphysical structure of the first world is such that the outcome is up to Alice, whereas the second world boasts no comparable metaphysical feature. If Almeida and Bernstein were to say this, we would need to know in what this crucial difference between the two worlds consists. What is it about the first world that confers on Alice the power that she lacks in the second?

A defender of the Assimilation Argument and the Rollback Argument will observe that Alice chooses in both worlds, and that each choice is free in the sense that nothing compels her to make it. But this is a separate question from whether she has a choice about which choice she makes; it’s a separate question from whether, in making the unconstrained choice she does, Alice exercises free will.¹⁹ What, then, would we have to add to the second world, where Alice’s choice is truly random, to give her a choice about which choice she makes, and thus to transform that choice from a random outcome into an exercise of free will?

If we adopt Franklin’s event-causal approach to libertarianism, it’s hard to see what materials are available to answer this question. For there doesn’t seem to be any event, set of events, or causal relationship between events, such that incorporating it in the etiology of Alice’s choice would ensure that the choice is up to her. If her choice could in principle be a random outcome, Almeida and Bernstein should accept that the Rollback Argument raises a genuine worry for libertarians. After all, a natural explanation for the random sequence we observe is simply that each individual choice in that sequence is a random outcome, and it isn’t clear how an alternative explanation might be supported. More specifically, it isn’t clear how any purely event-causal story about how Alice’s choice originates could give us reason to reject this explanation.

Summing up, unless libertarians can show that unconstrained, causally undetermined actions couldn’t be random outcomes after all, or else explain how exercises of free will differ from actions that are random outcomes, there is reason to believe that the Varied Outcome Thesis is less benign than Franklin, Almeida and Bernstein, and others have supposed.

Acknowledging the worry, some libertarians have appealed to agent-causal relationships to secure a difference between random actions and exercises of free will. Agent-causal accounts have been developed in numerous ways (see for example Chisholm 1966; Taylor 1966; O’Connor 1995 and 2000; Clarke 1993 and 1996b), but the central idea is that, qua agents or substances, we make a distinctive sort of causal contribution to what we do, a contribution that can’t be reduced to the contributions of our various states when we act. Such a theorist might hold, for example, that when Alice chooses to tell the truth, her competing reasons, qua states, confer an even probability on each choice, and that, if there were nothing more to say about the etiology of Alice’s choice, it would qualify as a random outcome. But, this theorist will continue, there is something more to say, for Alice settles the outcome by agent-causing her intention to tell the truth.

Objections to agent-causal accounts often focus on whether the notion of agent-causation is coherent. I shall set this issue aside. I want to focus instead on another important concern that is often raised about agent-causal accounts, namely whether positing agent-causation helps. Supposing that Alice agent-causes her choice to tell the truth, can she be said to settle the outcome by virtue of this?

As van Inwagen notes (2002, 173–75), it isn’t readily apparent how the appeal to agent-causation serves to counter the Rollback Argument.²⁰ Suppose for discussion’s sake that Alice agent-causes her intention to tell the truth in the actual course of events. Will this intention be subject to the Varied Outcome Thesis? If so, then given that Alice’s reasons for each choice remain equally weighted, the agent-caused outcomes should converge on the same ratio as they would in the absence of an agent-causal contribution. And if, as the agent-causal theorist allows, the outcome of each replay would be random in the absence of agent-causation, why think that what Alice agent-causes in each replay won’t be random as well? In effect, we will merely have exchanged a series of random event-caused choices with a series of random agent-caused choices.

One writer who invokes agent-causation in response to the Rollback Argument is Timothy O’Connor. In a situation like Alice’s, O’Connor would say, each of Alice’s competing reasons for action embodies an objective propensity for her to act in the relevant way (2011, 314–17, 326). Together with an individual’s “more enduring states of character, involving relatively fixed dispositions and long-standing general intentions and purposes around which her life has come to be organized,” such propensity-conferring reasons “probabilistically structure” that individual’s capacity to agent-cause the outcome (317). Ex hypothesi, Alice’s competing propensities are evenly matched. When she tells the truth, however, Alice herself (qua agent) is the sole cause of the intention she forms to tell the truth; it’s not the case that this intention is the joint product of her agent-causal capacity and the states that ground her propensity to tell the truth.

As an objection to his agent-causal account, O’Connor finds the Rollback Argument unpersuasive. In his view, the objection simply fails

…to take seriously the concept of agent causation. It [agent causation] is conceived as a primitive form of control over just such undetermined, single-case outcomes. The agent’s control is exercised not through the efficacy of prior states of the agent (as on causal theories of action), but in the action itself. Alice’s causing her intention to tell the truth is itself an exercise of control. And because, ex hypothesi, it is literally the agent herself generating the outcome, it is hard to see how the posited form of control could possibly be improved upon. So wherein lies the luck? (324)

Let us suppose, then, that Alice herself generates her intention to tell the truth and that her doing so exemplifies a primitive sort of control. How exactly does this supposition help to counter the Rollback Argument?

O’Connor doesn’t say explicitly what difference, if any, positing agent-causal control would make to the sequence we should expect to observe in the thought experiment. In particular, he doesn’t expressly affirm or reject what I’m calling the Varied Outcome Thesis in responding to the Rollback Argument (323–24). This no doubt reflects his tacit acceptance that agent-caused intentions such as Alice’s would conform to this thesis. (If O’Connor were to reject some aspect of this thesis, he would need to say so, and to explain why.) In O’Connor’s view, what differentiates this intention from a similar state that is event-caused and genuinely random is precisely that Alice herself generates this intention (something that, qua agent-cause, she cannot in turn have been caused to do),²¹ thereby exercising a primitive form of control over this action. For this reason, O’Connor maintains, “it is hard to see how the posited form of control could possibly be improved on.”

But is it up to Alice how she exercises the posited form of control? That is, is it up to her whether or not she generates this intention? If not, then such control falls short of what the libertarian requires. O’Connor might say that the outcome’s being up to Alice just is for her to agent-cause it when it was possible for her to agent-cause a different intention (or not to cause any intention at that moment). If this is right, the claim that Alice exercised this primitive form of control in agent-causing her intention, but that it wasn’t up to her whether she exercised it in this way, is incoherent. But is this claim incoherent? In earlier work, O’Connor expressly maintained that it is.²² With regard to an agent-caused event, e, he wrote:

Now the event e is itself clearly under the control of the agent, since he caused it (directly). But would it not, then, be perfectly absurd to raise a doubt concerning whether the agent controlled his causing e? Indeed, it seems to me that the question whether the agent has control over this event is ill framed—it simply is an instance of an agent’s exercising direct control over another event. (1995, 187, italics original)

However, it’s hard to see why we should accept this (cf. Widerker 2005, 89–90). After all, when we observe the random-seeming succession of outcomes, we will have the same impression that we did before we added Alice’s agent-causal power to the story. It will continue to seem as though Alice chooses willy-nilly from one replay to the next, and that the outcome of the next replay will be a matter of chance. And it seems that one possible explanation for what we observe is that which intention she causes in each replay is indeed random. (If so, then while it remains importantly unlike random nonactions, such as particles swerving one way or another, the difference lies in its being an action—something Alice does—and not in its being nonrandom.) Even if Alice’s generating her intention constitutes a sort of control over that intention, a sort that she would lack if she weren’t its agent-cause, it’s hard to see why such control ensures that her agent-caused intention is up to her.

Of course, we might opt to understand the relevant notion of control in such a way that possessing such control does logically guarantee that the outcome is up to Alice. But then we will have to ask why Alice’s agent-causing her intention, though it constitutes a kind of control, constitutes this kind. Again, it doesn’t seem incoherent to suppose that Alice might stand in an agent-causal relationship to her intention without possessing this kind of control over which intention she agent-causes. The agent-causal theorist will readily agree that such control is absent in the circumstances that Franklin (qua event-causal libertarian) believes entail it: Alice’s intention to tell the truth is appropriately caused by her reasons for doing so, when it’s possible, via an ordinary exercise of her agential abilities, that she actively forms the intention to lie for the relevant reasons instead. If, as I have argued, this isn’t enough to close the logical gap, it’s hard to see how positing a further, agent-causal relationship will do the trick. Indeed, even if this causal relationship makes the intention Alice’s in a distinctive way (a way that it wouldn’t be if it were caused partly or solely by her states), it seems that we have as much reason as ever to be skeptical about whether it’s up to her which intention she forms in any particular replay. In sum, we don’t transform a random outcome into an exercise of free will simply by positing a causal relationship, whether between the outcome and the agent (directly) or between the outcome and the agent’s states. This will be true even if we suppose that the presence of this particular type of causal relationship is essential to there being an action on Alice’s part.

I have argued that the Varied Outcome Thesis presents a serious problem for libertarians, contrary to what some writers have thought, and (focusing on O’Connor’s account) that positing agent-causal control doesn’t resolve this problem, even if we accept that there is such a thing as agent-causation. If libertarians can neither plausibly reject nor safely accommodate this thesis, they will be unable to resist the assimilation of the causally undetermined to the random. In that case, accepting that free will is a mystery may be the libertarian’s least bad option.²³