In this essay I reject the view that libertarian free will is compatible with infallible divine foreknowledge. David HuntFootnote 1 and Linda ZagzebskiFootnote 2 both argue that although each human action is accidentally necessary given God’s foreknowledge, these actions may yet be performed freely (where we are to understand freedom as libertarian freedom). I will critically evaluate Hunt’s and Zagzebski’s arguments in support of this thesis and conclude that these arguments fail, since the case of God’s foreknowledge is relevantly similar to other cases in which we should deny that an agent’s action was freely performed.

The view with which I am concerned here is the simple foreknowledge view, according to which God knows at t 1 that an agent will perform act A at t 1+n . This view contrasts with other views that one might hold regarding the nature of God’s omniscience and its relation to human action and free will. Among these are the middle-knowledge view, also known as the Molinist view, (according to which God knows the truth values of each agent’s counterfactuals of freedom, together with facts about the states of affairs in which these agents will find themselves), the Boethian view (according to which God’s knowledge of our future actions is outside of time, and thus not foreknowledge at all, strictly speaking), the Augustinian-Calvinist view (a compatibilist position according to which nothing happens that is not in accordance with God’s will, but our actions are nevertheless freely performed insofar as their causes are internal to us), and the open-theism view (according to which God does not know the truth values of propositions about the future because these propositions do not have truth values). I will have nothing to say here about these alternative views. In this paper I will focus exclusively on the simple foreknowledge view, to which I now turn.

Following Hunt (2001), let us consider these four inconsistent propositions:

  • (CK) If God foreknows that Adam will sin, then it is accidentally necessary that Adam will sin.

  • (N) If it is accidentally necessary that Adam will sin, then Adam will not sin of his own free will.

  • (K) God foreknows that Adam will sin.

  • (L) Adam will sin of his own free will.

Propositions (CK) and (N) involve a variant on the notion of accidental necessity. For present purposes, an event is accidentally necessary if (a) the event is contingent, (b) the event becomes inevitable if certain contingent conditions obtain, and (c) these conditions do in fact obtain. For example, it is a matter of contingency whether a pen I hold in my hand just prior to time t 1 will hit the floor at t 2 , but given that (a) the pen’s hitting the floor at t 2 is inevitable if I drop the pen at t 1 , and (b) I do in fact drop the pen at t 1 , the pen hits the floor at t 2 as a matter of accidental necessity.Footnote 3

We can now proceed to an argument against the compatibility of simple foreknowledge and human freedom, which may be stated as follows:

  1. (1)

    If God believes at t 1 that S will do A at t 3 , then S’s doing A at t 3 is accidentally necessary at t 2 .

  2. (2)

    If S’s doing A at t 3 is accidentally necessary at t 2 , then S cannot do otherwise than perform A at t 3 .

  3. (3)

    If S cannot do otherwise than perform A, then S does not perform A freely (the Principle of Alternative Possibilities).

  4. (4)

    Therefore, if God believes at t 1 that S will perform A at t 3 , then S does not freely perform A at t 3 .

The proponent of the simple foreknowledge view must deny one of the three premises of this argument. Typically, the foreknowledge defender accepts premises (1) and (2) and rejects premise (3), the Principle of Alternative Possibilities (hereafter, PAP). This move entails the denial of proposition (N) above: if it is accidentally necessary that Adam will sin, then Adam does not sin of his own free will. Hunt’s own argument for the rejection of (N) can be stated as follows:

  1. (1)

    ‘If it is accidentally necessary that Adam will sin, then Adam does not sin of his own free will’ expresses a true proposition only if PAP is true.

  2. (2)

    PAP is false.

  3. (3)

    Therefore, it is not true that Adam’s sin is unfree if his sinning is accidentally necessary.Footnote 4

If PAP is true, then we get the following result: if Adam’s sin is accidentally necessary then he could not have done otherwise, and therefore he does not sin of his own free will. Hunt argues that PAP is false by appealing to a Frankfurt-style case.Footnote 5 We are to imagine that Black wants Smith dead, and Black knows that Jones also shares this desire and is likely to act on it. Since Black wants to ensure that Smith dies, he creates a device that does the following: it will monitor Jones’s mental states, and if it determines that Jones is not going to kill Smith, the device will interfere with Jones’s mental processes so that Jones will kill Smith; on the other hand, if the device determines that Jones is going to go ahead and kill Smith on his own, the device will do nothing. Either way, Jones will kill Smith and it will not be the case that he could have done otherwise. In the end, Jones kills Smith without the device intervening.

It is reasonable to think that Jones kills Smith of his own free will, even though Jones could not have done otherwise, since the device never intervenes to force Jones to act. Since Jones is not causally determined to kill Smith, his killing Smith is a freely performed action. One could argue to this conclusion as follows:

  1. (1)

    The Frankfurt case shows that PAP is false.

  2. (2)

    A reasonable alternative to PAP is the following principle, call it the causal noninterference principle (CNP): S freely performs action A if and only if S’s performance of A is not the result of S’s having been causally determined to perform A.

  3. (3)

    CNP is consistent with Jones’s acting freely.

  4. (4)

    Therefore, we cannot conclude that Jones is unfree on the ground that Jones does not have alternative possibilities.

Hunt argues that God’s foreknowledge that Jones will (freely) kill Smith is no different from Black’s device (provided that the device does not in fact interfere causally with Jones, but instead merely monitors Jones’s actions, thereby ensuring that Jones will kill Smith but not, ultimately, as the result of the device’s causal interference). As Hunt writes, ‘[t]hough both Black’s device and God’s foreknowledge eliminate all alternatives to Jones’s action, neither does so in such a way as to jeopardize Jones’s free agency.’Footnote 6 Though they are logically entailed by God’s foreknowledge, our actions are not caused by God’s foreknowledge. Since ‘divine foreknowledge does not cause or compel what it knows,’Footnote 7 freedom is compatible with divine foreknowledge. We can say, then, that the proponent of the simple foreknowledge view denies leeway incompatibilism, the view that endorses PAP, but is committed to source incompatibilism (or causal history incompatibilism), the view that Jones acts freely if the cause of his action is nothing but Jones himself, regardless of the lack of alternative possibilities.Footnote 8 Source incompatibilists need only embrace CNP, as opposed to PAP.

I wish to set aside the possibility that the Frankfurtian argument against PAP fails, and that PAP is in fact true. Instead, I will simply grant the falsity of PAP and pursue a different line of objection. Hunt argues that since neither Black’s device nor God’s foreknowledge causally interact with Jones, neither fact (the device’s preparedness to force Jones to kill Smith, God’s foreknowledge that Jones will kill Smith) jeopardizes Jones’s free agency, since it is CNP rather than PAP that is true. But CNP, as it turns out, is false. It is not true that the only way in which a prior fact can interfere with human freedom is by way of a causal interference. No agent S, for example, is free to make true the proposition ‘S calculates the sum of 1 and the greatest prime number.’ This proposition is false at all possible worlds, but not because S is causally compelled at every possible world to fail to make the proposition true. Consider as well the following hypothetical case: Dr. Evil offers ‘Jimmy the Gun’ $50,000 to do the following: at precisely 12:00 pm on 31 July 2007, Jimmy will kill the wealthiest bachelor on the sparsely populated Isle of Plenty. Jimmy accepts the offer, and goes to sleep on the night of 30 July with assassination plans in hand. When he awakes at 9:30 am on 31 July, the world has changed: it so happens that the last three of the island’s bachelors were all married to different brides at 9:00 that morning. At 9:30, then, it is accidentally necessary that Jimmy does not kill the wealthiest bachelor at 12:00, and its accidental necessity is not the result of his being caused or coerced to perform any particular action. Indeed, Jimmy’s inability to perform the act of killing the wealthiest bachelor has nothing at all to do with any causal interference with Jimmy’s actions; it is a purely social fact that there are no longer any bachelors on the island. The result is that had the weddings not occurred, Jimmy could have freely refrained from killing the wealthiest bachelor at 12:00, but now it is the case that Jimmy’s not killing the wealthiest bachelor on the island at 12:00 pm cannot be freely performed, since there is nothing for Jimmy freely to refrain from doing. He cannot freely refrain from killing the wealthiest bachelor, since it is impossible for him to make true the proposition Jimmy kills the wealthiest bachelor at noon if he so chose. Therefore, CNP is false.

We should note that facts about Jimmy’s psychology are irrelevant. It need not be the case that Jimmy discovers what has happened, and then chooses a course of action based on the belief that there are no longer any bachelors on the island. In fact, we may stipulate that Jimmy never learns about the marriages, and proceeds according to his plan. It is still the case that Jimmy fails to make true the proposition Jimmy kills the wealthiest bachelor at noon, even if Jimmy kills the man who had been the wealthiest bachelor on the island before the marriages took place.

We can take the example a step further. Suppose that twenty-four hours in advance of the planned assassination, Dr. Evil decides that he does not want the wealthiest bachelor killed after all. Rather than withdraw the offer to the unpredictable and volatile Jimmy, Dr. Evil decides to deploy a device very similar to the Frankfurt device, except that instead of effecting a change in Jimmy’s mental states if it appears that he will go ahead with the plan, this device will monitor the Jimmy’s mental states as well as the mental states of the three remaining bachelors and their fiancées and, if Jimmy appears ready to go ahead with the murder, cause them to marry and thwart the plan. Suppose that Jimmy goes to bed on the night of 30 July intending to carry out the murder on the next day and consequently the device springs into action, causing the three couples to marry before Jimmy awakes. The result is the same as in the original case, except that in this case Dr. Evil unilaterally disrupts Jimmy’s freedom and does so without interacting causally with Jimmy. Jimmy’s not killing the wealthiest bachelor is determined by a device that does not interact causally with Jimmy at all (though it does interact causally with the three bachelors and their fiancées); as far as Jimmy is concerned, all that the device has done is to logically foreclose a certain action. As in the first case, Jimmy’s freedom appears to be impaired without any external causes determining Jimmy’s action. This conclusion undermines CNP.

One might be tempted to reply that the alternative-eliminators in the Jimmy stories are such that Jimmy would likely behave differently in the absence of those alternative-eliminators; just as Jones, in the Frankfurt case, would act differently in the absence of Black’s device, so too would Jimmy act differently in the absence of the alternative-eliminating social facts that I have stipulated. This reply is inadequate for two reasons. First, as I argue above, it may be true that in fact Jimmy does kill the man who had been the wealthiest bachelor on the island, and yet false that Jimmy may freely kill (or refrain from killing) the wealthiest bachelor, since it is accidentally necessary that Jimmy cannot make true the proposition Jimmy kills the wealthiest bachelor at noon. Second, the Jimmy cases involve actual-sequence compulsion, meaning that Jimmy’s ability to freely kill or refrain from killing the wealthiest bachelor is undermined in the actual sequence of events. In the Frankfurt case, Jones’s alternatives are eliminated because in the alternate sequence (in which Jones’s mental states indicate that he will not choose to kill Smith), Jones’s act of murder is compelled by the device. I assume for the sake of argument that the Frankfurt case may show that free action is possible under alternate-sequence compulsion, but no libertarian could accept that one acts freely under actual-sequence compulsion. Since the Jimmy stories involve actual-sequence, non-causal compulsion, appeal to Frankfurt-style cases, which involve alternate-sequence compulsion, is fruitless. The distinction between actual-sequence compulsion and alternate-sequence compulsion will be of central importance at a later stage in the argument; what is significant at this point is that causal compulsion is not the only kind of compulsion that can undermine an agent’s freedom. Human freedom can be undermined when actions are non-causally compelled. Henceforth, any occurrence of ‘compulsion’ should be understood to cover cases of non-causal as well as causal necessitation of a given action. If the reader insists that ‘compulsion’ should be reserved for cases of causal necessitation, then he or she should treat my usage as the employment of a term of art, keeping in mind my thesis that causal necessitation is not the only type of necessitation that can undermine human freedom.

Of course, in our hypothetical Jimmy cases, we wouldn’t consider Jimmy’s freedom to be radically impaired, since his freedom is limited only with respect to a single possible action—killing the wealthiest bachelor. That some actions on some occasions are logically foreclosed to us, and in such a way that our freedom can be undermined with respect to the performance of those actions, is not a troubling conclusion. When we consider God’s foreknowledge, however, we must conclude that the logical foreclosure of alternative possibilities by God’s foreknowledge would radically impair our freedom; we would have a ‘Jimmy case’ with respect to all actions by all agents.

The proponent of the simple foreknowledge view must offer an argument that does not presuppose that causal compulsion is the only type of compulsion that undermines human freedom. Linda Zagzebski has offered such an argumentFootnote 9 (call this argument F):

  1. (1)

    I freely do S at t 3 if it is the case that if God had not believed that I would do S at t 3 then I would have done S at t 3 anyway.

  2. (2)

    I would have done S at t 3 even if God had not believed at t 1 that I would.

  3. (3)

    Therefore, I freely do S.

This argument is valid, and premise (2) parallels the claim that Jones may act freely despite the presence of Black’s device, since when Jones acts ‘on his own’ it is true that he would have killed Smith even if the device had not been present. The weakness of the argument lies in the fact that the first premise is false, since the fact that I would have done S even if God had not believed I would is insufficient to guarantee that I do S freely. If premise (1) is to be more than an ad hoc claim, then it is best construed as an instance of the following more general principle:

  • (P) Some fact f does not prevent my freely doing S if it is the case that had f not obtained, I would have done S anyway.

It is this principle that seems to underlie the claim that Jones acts freely when the device does not intervene, and that Adam sins freely when God knows that Adam will sin. In both cases, so the story goes, the agent acts freely if he would have performed the act in question even in the absence of the facts that determine his inability to do otherwise. Unfortunately, (P) is false, and a slight modification to the Frankfurt case will illustrate this. Suppose that Black’s machine malfunctions in such a way that if it detects that Jones is going to kill Smith freely, then the machine intervenes to cause Jones to kill Smith, and suppose that this is in fact what happens. In that case, it is just as obvious that Jones does not freely kill Smith as it is that Jones freely kills Smith in the original case. But if (P) is true, then the fact that the machine causes Jones to kill Smith does not prevent Jones from freely killing Smith, since had the machine not intervened, Jones would have killed Smith anyway. Therefore, (P) is false, and F is rendered unsound by the falsity of its first premise.

Now it is quite clear that in the modified Frankfurt case, Jones’s act is not just modally overdetermined but also causally overdetermined. The proponent of the simple foreknowledge view may argue that God’s foreknowledge modally overdetermines, but does not causally overdetermine, my free actions. We have already seen that such an argument is a non-starter. An alternative strategy would be to argue, as Zagzebski does, that in the modified Frankfurt case Jones does not act ‘on his own.’ In support of this position, one might argue that in the original Frankfurt case, although there is an alternate sequence in which Jones is causally determined to kill Smith, the actual sequence does not involve compulsion, causal or otherwise, and therefore does not render Smith’s action unfree. Since the actual sequence involves no compulsion, the fact that Jones would kill Smith in the alternate sequence is irrelevant to the question of Jones’s freedom in the actual sequence. Likewise, the foreknowledge proponent argues, God’s infallible beliefs about the future do not make it true that an agent fails to act ‘on his own.’

This strategy may appear promising, since it acknowledges that freedom under alternate-sequence compulsion does not guarantee that one acts freely under actual-sequence compulsion.Footnote 10 But it is actual-sequence compulsion that the theological fatalist claims is a necessary consequence of divine foreknowledge. To see why this is so, consider that in all possible alternate sequences in which Adam sins, it is true that God foreknew that Adam would sin. Therefore, Adam sins if and only if God foreknows that Adam will sin. But the reason that we think freedom is undermined in cases of causal compulsion is just that

  • (C) Action A would not have occurred if state of affairs S had not obtained.

When (C) is true, A is determined by S. But of course, (C) is consistent with the case of Adam’s sinning, since if the state of affairs in which God believed that Adam would sin had not obtained, then Adam’s action, sinning, would not have occurred. If God had believed that Adam would not sin, then Adam would not have sinned. If God exists in all possible worlds, then in every possible world, God’s belief that Adam will sin is necessary and sufficient for Adam’s sinning.

Consider the following cases:

  1. Case 1:

    • (Actual Sequence) Jones kills Smith without the device’s intervention.

    • (Alternate Sequence) The device intervenes to cause Jones to kill Smith.

    If these are the actual and possible sequences, then Jones’s killing Smith is modally overdetermined but not causally overdetermined. There is alternate-sequence compulsion, but not actual-sequence compulsion. Now consider:

  2. Case 2:

    • (Actual Sequence) The device malfunctions and causes Jones to kill Smith when Jones would have killed Smith on his own.

    • (Alternate Sequence) The device intervenes to cause Jones to kill Smith when Jones would not have killed Smith on his own.

    If these are the actual and possible sequences, then Jones’s killing Smith is both modally and causally overdetermined. Furthermore, there is both actual-sequence compulsion and alternate-sequence compulsion. Finally, consider:

  3. Case 3:

    • (Actual Sequence) The device malfunctions and causes Jones to kill Smith when Jones would have killed Smith on his own.

    • (Alternate Sequence) The device malfunctions and does not intervene to cause Jones to kill Smith when Jones would not kill Smith on his own.

    In this case, Jones’s killing Smith is causally overdetermined but not modally overdetermined. Furthermore, there is actual sequence compulsion but not alternate-sequence compulsion.

The Frankfurtian move to avoid theological fatalism requires that God’s foreknowledge that Adam would sin is relevantly similar to Case 1, in which we find alternate sequence compulsion but not actual sequence compulsion. For it is in this case that Jones’s action is independent (in the actual sequence) of the non-causally necessitating features of the situation (the conditions that make the case that Jones cannot do otherwise), since those features entail only alternate-sequence compulsion.Footnote 11 But there is no reason at all to think that the foreknowledge case is not relevantly similar to Case 2 instead, where we find both actual-sequence compulsion and alternate-sequence compulsion. While the proponent of the simple foreknowledge view might counter that God’s foreknowledge does not causally intervene, and thus does not ‘compel’ an agent’s actions, I have argued above that causal interference is not necessary for the diminution of freedom. If foreknowledge involves actual-sequence compulsion (where ‘compulsion’ is understood to include cases of non-causal necessitation), then it would be unreasonable to suppose that divine foreknowledge is consistent with libertarian freedom.

The proponent of simple foreknowledge must therefore demonstrate that there is no actual-sequence compulsion due to God’s foreknowledge. This is in fact the route taken by Zagzebski, who writes that the problem with causal determinism is not the lack of alternative possibilities, but rather that ‘if the world is causally determined, my act does not really originate in me. For my act really to be my own, it must have a certain kind of independence of any conditions which make it the case that I cannot do otherwise.’Footnote 12 One might use this intuition to motivate the following argument:

  1. (1)

    If God had not existed or had not been omniscient, Adam would still have sinned.

  2. (2)

    If Adam would have sinned even if God had not existed or had not been omniscient, then Adam’s sinning is independent of God’s knowledge that Adam would sin.

  3. (3)

    If Adam’s sinning is independent of God’s foreknowledge, then God’s foreknowledge does not interfere with Adam’s freedom.

  4. (4)

    Therefore, God’s foreknowledge does not interfere with Adam’s freedom.Footnote 13

Unfortunately, one could offer a parallel argument with respect to the modified Frankfurt case, in which the device malfunctions and causes Jones to kill Smith even when Jones would have killed Smith without the device’s intervention:

  1. (1a)

    If the device had not existed, Jones would still have killed Smith.

  2. (2a)

    If Jones would still have killed Smith, then Jones’s killing Smith is independent of the device.

  3. (3a)

    If Jones’s killing Smith is independent of the device, then the device does not interfere with Jones’s freedom.

  4. (4a)

    Therefore, the device does not interfere with Jones’s freedom.

The conclusion of this argument is absurd; the device does interfere with Jones’s freedom in the modified case. Premise (2a), which drives us to the absurd conclusion, is false. But (2a) has the same form as (2), so if (2a) is false, then it is far from clear that (2) is true. If we cannot establish that (2) is true, then we cannot establish that there is no actual-sequence compulsion given divine foreknowledge. The only substantial difference between (2) and (2a) is that the latter involves causal overdetermination while the former does not. Since we have already seen that this difference is not enough to drive the foreknowledge proponent’s argument, it would be futile to appeal to this difference to motivate a response.

We should conclude that a person’s freedom can be undermined (meaning that an action that she could have performed freely is now accidentally necessary and cannot be performed freely) even if she is not causally determined to perform any particular action. If this is true, then Hunt’s and Zagzebski’s arguments for the rejection of proposition (N) fail, since the arguments require that at least one of the following two assumptions be true:

  1. (1)

    that freedom can only be impaired by causal determination and not by mere logical foreclosure of alternative possibilities;

  2. (2)

    that the non-causally necessitating features of the case must be such that the agent’s actual-sequence action is independent of those features.

The Jimmy cases show that assumption (1) is false, and we have already seen that the argument from the independence of non-causally necessitating features, which depends upon assumption (2), must fail. We must accept that if God foreknows which actions we will perform, and that logical foreclosure of alternative possibilities can indeed impair freedom, then the arguments for the compatibility of foreknowledge and human freedom appear to be unsuccessful.