A counterexample to Plantinga’s Free Will Defense


Abstract: Plantinga’s Free Will Defense is an argument that, possibly, God cannot actualize a world containing significantly creaturely free will and no wrongdoings.  I will argue that if standard Molinism is true, there is a pair of worlds w1 and w2 each of which contains a significantly free creature who never chooses wrongly, and that are such that, necessarily, at least one of these worlds is a world that God can actualize. 


1. Introduction

            Plantinga’s Free Will Defense (FWD) is an argument that possibly the truth values of counterfactuals of creaturely freedom (CCFs) are such that God cannot actualize a morally perfect world.  A “morally perfect world” is a world where at least one created agent has a libertarian-free choice between something wrong and something not wrong (“is significantly free”) and where no created agent does anything wrong.  If God cannot create a morally perfect world but can instead actualize a very good, but morally imperfect, world containing created significantly free agents, he will be justified in actualizing some such world.  Therefore, if Plantinga’s FWD succeeds, possibly God actualizes a world containing a moral evil, and so the existence of God is logically compatible with the existence of evil, which refutes the deductive argument from evil.

            According to standard Molinism, which Plantinga’s FWD in the Nature of Necessity[1] does assume, CCFs have non-trivial, and at least typically contingent, truth values.  I shall assume standard Molinism, and argue that there is a pair of possible worlds, w1 and w2, each of which is morally perfect and which are such that, necessarily, God can actualize w1 or God can actualize w2 (or both).  Consequently, the claim in Plantinga’s FWD that, possibly, God cannot actualize a morally perfect world is false, if standard Molinism is true.  Plantinga’s FWD is based on the possibility of transworld depravity.  Roughly speaking, transworld depravity says that every possible person is such that were she created in any context that God could create her in, she would either never have significant freedom or she would do wrong at some time.  The argument I shall give directly shows that transworld depravity is impossible.

            In the next section, I will give some crucial preliminaries, including arguing for a controversial counterfactual Domination Principle inspired by an example of Plantinga’s.  In the subsequent section, I will construct the pair w1 and w2 and argue on the basis of the Domination Principle that God can actualize at least one of them, assuming standard Molinism.  I will also briefly sketch an argument that even if the Domination Principle is false, a modified principle should be correct which at least has the result that probably God possibly could actualize at least one of w1 and w2.  I will then consider some objections.  Finally, I will discuss some options for free will defenses that are not affected by the objections.  I am inclined to think the problem lies not so much with free will defenses as such, but with Molinism.

2. Crucial preliminaries

2.1. CCFs

            By stipulation, a CCF is a subjunctive conditional of the form C®F where C is an “appropriate antecedent for F” and F reports that an agent did or did not freely choose something at a given time.[2]  Oddly enough, although it is traditional to use the word “counterfactual” for C®F, no assumption that C is false is made.  It is somewhat difficult to characterize what an “appropriate antecedent for F” is.  Plantinga’s characterization is that C is the conjunction of all the states of affairs that God “strongly actualizes”, where God strongly actualizes S provided that God causes S and every contingent state of affairs that is “included” (entailed?) by S.[3]  Another characterization is that C reports all the events in the temporal sequence of a world right up to the choice that F describes, including the fact that a free choice takes place, but not including any information on what the agent in fact did choose, or anything that follows from that.[4]  In particular, the appropriate antecedents will include all the considerations operative for the agent in the choice.  I shall also assume that only propositions that are possibly true are appropriate antecedents, and whether an antecedent is appropriate to F is not a contingent matter.

            Standard Molinism then says that CCFs have truth value independently of God’s activity, and God knows these truth values and can make use of them in deciding which antecedents to actualize.[5]  But just to say this much is not enough to be a standard Molinist.  For instance, Robert M. Adams[6], a paradigmatic anti-Molinist, thinks that CCFs have truth value independently of God’s will, because, possibly with some exceptions (one of which will actually be central to the argument for the Domination Principle), they are simply false.  I shall characterize standard Molinism by saying that (a) God knows the truth values of CCFs, (b) God can act on these truth values, and (c) the Conditional Law of Excluded Middle (CLEM) holds as restricted to CCFs.

            Unrestricted CLEM is the claim that, necessarily, p®q or p®~q, for all p and q (to be distinguished, of course, from the claim that p®(q or ~q), which is trivial at least if p is possible).  While one might argue for standard Molinism on the basis of CLEM, unrestricted CLEM is usually taken to be implausible.  Surely it is neither the case that were aliens to have inscribed a giant six digit integer on the far side of moon, then that integer would be even, nor that if they were to do that, then the integer would be odd.  However, the standard Molinist accepts CLEM in the special case of conditionals of the form C®F where F reports what a created person does or does not freely choose and C is sufficiently determinate—and all the “appropriate” antecedents will count as sufficiently determinate.

2.2. Dominance

            Plantinga has offered the following interesting argument against those who deny truth values to CCFs.  Suppose that Mayor Curley Smith has accepted a bribe of $35,000 to drop his opposition to a bill.  Then, surely:

(1)         Smith is offered $36,000 ® Smith (still) freely accepts the bribe.[7]

Hence, at least one conditional of free will (though perhaps not a CCF in our present terminology as the antecedent may not be sufficiently determinate) is true.  Interestingly, even Adams, though generally opposed to Molinism, accepts this example, albeit holding that (1) is made true by the facts of Smith’s actual choice.[8]

            Now, notice that we would not accept (1) in every case in which Smith has accepted a $35K bribe.  For instance, if state law has a higher penalty for accepting a bribe over $35K, it might well be the case that although Smith accepted $35K, he wouldn’t have risked accepting $36K.  Likewise, it could be that Smith has found a way of laundering $35K which wouldn’t work for $36K, or Smith has a special liking for numbers that are divisible by seven.  Of course, when we are pulled to say that (1) is true, we are assuming none of these situations occur.  Thus, what we really committed to is this:

(2)         If any consideration in favor of taking the smaller bribe would apply at least as well to the larger and any consideration there would be against taking the larger bribe also in fact applied at least as well to the smaller, and if Smith prefers more money to less money, then: Smith is offered a larger bribe ® Smith (still) freely accepts the bribe.

            Now, (2) is a consequence of a Categorical Domination Principle, which I submit we should accept as explaining (1) and (2):

(CDP)  Necessarily: If (a) C and C* are antecedents appropriate to <x freely chooses A>, (b) C* dominates C for x choosing A, (c) C obtains and (d) x freely chooses A, then C*®(x freely chooses A).

Here, <…> is shorthand for “the proposition that …”.  I call this principle “categorical”, because condition (d) talks of what x freely chooses, rather than what x would freely choose.  

            The notion of domination in CDP might be stipulated as follows.  C* dominates C for x choosing A if and only if: (a) C* counts as an alternative to C, so that the choice it describes x as having before her is spatiotemporally located just as in C, between the same options as in C, and both C and C* include the same laws of nature (more conditions may need to be added here), (b) every consideration included in C in favor of x choosing A is present in C* with at least as great a force in favor of x choosing A, (c) every consideration in C* against x choosing A is present in C with at least as great a force against x choosing A, and (d) either (i) some consideration is present in C* in favor of choosing A that either isn’t present in C or did not favor A in C or favored A less in C than it does in C* or (ii) some consideration is present in C against choosing A that either isn’t present in C* or does not count against A in C* or opposes A less in C* than it does in C.  If we like, we might add the condition that the agent has no inclination to act unreasonably for the sake of acting unreasonably.

            I shall only use the notion of domination for “appropriate” antecedents C and C*, and I shall take it that appropriate antecedents are sufficiently determinate that whether C* does or does not dominate C is not a contingent matter.

            The “considerations” mentioned in the definition of domination are subjective in nature.  What exactly they are depends on one’s theory of action, but they are meant to be all the subjective factors that influence the action.  Candidates include one or more of: motives, desires, choice-relevant beliefs, subjective reasons, inclinations, etc.

            Now, Molinist freedom is a species of libertarian freedom, and in order that libertarian freedom not fall prey to the randomness objection, the libertarian must hold that free choices are always made because of considerations, even though the choices are not determined by these considerations.[9]  And this gives rise to the reason why libertarians, whether Molinist or not, should accept CDP. 

            For, plausibly, a part of what one is saying when one says that an action was chosen because of the considerations in favor of the action is that if all the other considerations were the same, but the considerations in favor of the action were strengthened, or if instead the considerations against the action were weakened, one would still have chosen in the same way.  If this is not the case, then it seems to become unintelligible that the action is done because of the considerations.

            It’s worth noting that CDP is particularly plausible given the Lewis-Stalnaker account counterfactuals, since a world where one makes the same decision on account of the same considerations appears closer than a world where one makes a different decision and acts against the considerations that had moved one in the actual world.

            Now, consider the following modified principle which I will call simply the “Domination Principle”:

(DP)    Necessarily: If (a) C and C* are antecedents appropriate to <x freely chooses A>, (b) C* dominates C for x choosing A and (c) C®(x freely chooses A), then C*®(x freely chooses A).

The reasons for accepting CDP are, I think, reasons for accepting DP as well.  However, I can also give an explicit argument for DP from CDP.

            Suppose that conditions (a)-(c) of DP are satisfied at a world w, and assume CDP.  Let F be <x freely chooses A>.  For a reductio, suppose that it is not true at w that C*®F.  For the sake of brevity, assume that w is the actual world.  So, we have (C®F) & ~(C*®F).  Now, were C to hold, then (C®F) & ~(C*®F) would still hold.  For it is within God’s power to make C hold (because C is an appropriate antecedent), but according to standard Molinism it is not within God’s power to affect the truth values of CCFs.  Therefore:

(3)         C®((C®F) & ~(C*®F)). 

Now, I will make use of three axioms about counterfactuals with possibly true antecedents (recall that all “appropriate” antecedents are possibly true):

(4)         If p®q, then p®(p&q).

(5)         If p®q and necessarily (q if and only if r), then p®r.

(6)         Necessarily: If p and p®q, then q.

It is a very easy exercise to see that from (3)-(6) we can derive:

(7)         C®(C & F & (C®F) & ~(C*®F)).[10]

            Now, neither domination nor appropriateness of antecedents are contingent matters.  Therefore, conditions (a) and (b) in CDP hold necessarily if they hold at w (which we assumed was the actual world), and so CDP tells us that:

(8)         Necessarily: If C & F, then C*®F.


(9)         Necessarily: C & F if and only if C & F & (C*®F).

By (5) and (7), we get:

(10)     C®(C & F & (C*®F) & (C®F) & ~(C*®F)).

But the consequent of (10) is contradictory, and a counterfactual with possible antecedent and impossible consequent is always false.  Hence, we have an absurdity, and so the assumption that C*®F is true at w (which for convenience we took to be the actual world) is false.  And, thus, DP is true.

3. The construction

            Consider a family W of worlds at each of which God creates Eve, an apple and a dancing ground.  At t1, Eve must freely choose between either eating the apple or dancing a jig.  She cannot do both and she cannot fail to choose one of these two options.  These facts are going to be all a part of all the relevant appropriate antecedents.  Let A=<At t1 Eve chooses to eat the apple>, and J=<At t1 Eve chooses to dance the jig>.  Moreover, in the worlds in W, this is the only free choice any creature ever gets, and the laws of nature are deterministic except for that choice.  In all the worlds in W, Eve wants to eat the apple on account of its yumminess and to dance the jig on account of merriness.  In none of the worlds in W is Eve in any way motivated by a desire to act contrary to the will of God as such or inclined to act unreasonably, but Eve does have a motivation, though not an overwhelming one, to obey God.  There are no other relevant desires (in some subsequent discussions, this condition will be relaxed).

            Now, consider a world wJ in W.  In wJ, God forbade Eve to eat the apple (I stipulate that whenever I talk of prohibitions, the agent under the prohibition is aware of the prohibition), and Eve chose to dance the jig.  Let C be the antecedent appropriate to J in wJ.  Moreover, there is a world wA at which C also holds, but where, alas, Eve chooses to eat the apple. 

            Next, let wJ* be a world just like wJ, except that in wJ*, God forbade Eve to dance the jig instead of forbidding eating the apple, but Eve still chose to dance the jig just as in wJ.  Let C* be the antecedent appropriate to J in wJ*.  Finally, let wA* be a world just like wJ*, with the same divine prohibition on dancing a jig, except that Eve eats the apple there.

            I shall assume that Eve has a moral obligation—perhaps in virtue of benefits received or her relationship with God—to obey God, at least when God doesn’t command something innately immoral.  Thus, in wJ and wA, Eve is prohibited from eating the apple, whereas in wJ* and wA*, Eve is instead prohibited from dancing the jig.

            Observe that both wJ and wA* are morally perfect worlds: there is significant creaturely freedom, and nobody does wrong.  I will now show that if standard Molinism holds, then God can actualize wJ or God can actualize wA*.  All my premises will be necessary truths, and so the argument will establish (assuming standard Molinism) that, necessarily, God can actualize a morally perfect world, which is all I need to show to refute Plantinga’s FWD.

            To see this, observe that if C®J, then God can actualize wJ by making C hold.  And if C*®A, then God can actualize wA* by making C* hold.  Hence, it suffices to show that C®J or C*®A.  Moreover, if I can show this using only necessary truths as premises, then it will follow that transworld depravity is necessarily false, since Eve would either be significantly free and sinless were C to hold and were she to have no further opportunities for significantly free action, or else she would be significantly free and sinless were C* to hold and were she to have no further opportunities for significantly free action.

            I will now show that C®J or C*®A by showing that if ~(C®J), then C*®A.  To that end, suppose ~(C®J).  Then, by CLEM as restricted to CCFs, C®~J is true.  Moreover, because according to C, Eve must choose between the jig and the apple, it follows that C®A is true.  But C* dominates C in respect of Eve choosing to dance the jig.  For, all the considerations present in C in favor of Eve’s dancing the jig are also present in C* with equal strength: the jig is just as merry in C* as in C, and Eve has no motivation to disobey God as such, so the fact that in C* Eve is not forbidden to dance the jig is not a reason to dance the jig.  Similarly, all the considerations present in C* against dancing the jig are also present in C with equal strength: the apple is just as yummy in C as in C*, and Eve has no desire to disobey God as such.  On the other hand, there is a consideration in C against dancing the jig that is not present in C*: according to C, God has prohibited the jig, and Eve has a motivation to obey God.  Hence, indeed, we have domination.  Since C®A, it follows by DP that C*®A.  Hence, we have shown that if ~(C®J), then C*®A

            Thus, we have shown that God can actualize wJ or God can actualize wA*.  Since both are morally perfect, it follows that God actualize a morally perfect world.  Moreover, all the premises in the argument were necessary truths.  Hence, necessarily, God can actualize a morally perfect world.

            It is pretty intuitive that if Eve would eat the apple given C despite the prohibition, she would a fortiori eat the apple given C*.  Granted, sometimes, the fact that something is prohibited may motivate one—think of the use of “sinful” in advertising—but we have assumed that Eve does not suffer from that perversity.

            The argument used DP.  Suppose, however, that we do not accept DP.  Nonetheless, it is very plausible that if Smith accepted the $35K bribe, he would have accepted a $36K bribe.  This suggests that even if we reject DP, we should accept DP’s probabilistic variant:

(DPPV)     If (a) C and C* are antecedents appropriate to <x freely chooses A>, (b) C* dominates C for x choosing A, and (c) C®(x freely chooses A), then probably C*®(x freely chooses A).

And my argument above then shows that if God cannot actualize wJ, he can probably actualize wA*.  Hence, probably, God can actualize either wJ or wA*.  Hence, probably, God can actualize a morally perfect world.

            As the interested reader can easily verify, the above examples also work just as well on the refined versions of transworld depravity given by Otte.[11] 

4. Objections

            Objection 1: The construction supposes that it is wrong for Eve to disobey God.  But, as Mark Murphy has argued at length[12], there is no duty for x to obey God absent a special relationship between God and x, and without Eve having earlier been free, there is no way to ensure there is such a special relationship.

            Response: Rather than refuting Murphy’s argument, the simplest response is just to modify the case to evade the objection.  Instead of letting wA and wJ be worlds where God has forbidden eating the apple, let them be worlds where Eve knows that eating the apple causes (through some odd causal law) a severe harm to some non-consenting third party who does not deserve the harm (e.g., an Adam who is not significantly free).  Then let wA* and wJ* be just like that, except this time it is the dancing of the jig that is known to cause the harm.  It’s wrong to eat an apple or dance a jig simply on account of yumminess or merriment at the cost of severe harm to a third party, and the rest of the argument goes through.  Similar cases can be multiplied.

            Objection 2: Surely God has good reason to create a world with more than one significantly free creature.  But the example in this paper only shows that God could create a world with one significantly free creature that never goes wrong.

            Response: The example generalizes.  I will leave the details to the interested reader, but sketch the idea.  Consider worlds with n persons, E1,…,En.  Each is facing a simultaneous and independent (maybe they are doing this in far-separated parts of Eden) choice whether to eat an apple or dance a jig, and each is either forbidden the apple or forbidden the jig.  Now, there are 2n possible combinations of relevant divine prohibitions: maybe E1 is forbidden the apple, E2 the jig, and so on.  We also assume that the prohibitions are individually and independently communicated to each person, so that what each person does is, plausibly (see also the discussion of CRP, below), counterfactually independent of any combination of others’ actions and of divine commands to others.  DP then ensures that God can choose what to command Ei such that Ei would obey.  If Ei is such that she would eat the apple were she forbidden it, then by DP, Ei would also eat the apple were she forbidden the jig instead, so God need only forbid the jig to Ei.  And if Ei is not such that she would eat the apple were she forbidden it, and hence (by restricted CLEM) she is such that she would not eat the apple were she forbidden the apple, then God can safely forbid her to eat the apple.  And, so, God can ensure that everybody makes the right significantly free choice.

            Objection 3: Libertarians with whom I’ve discussed principles like CDP and DP have tended to deny them.  One intuitive reason for the denial appears to be the Repeat Intuition:

(RI)     If an agent freely chooses A in circumstances C, then were her memories of her choice and of its consequences deleted and she put for a second time in circumstances just like C, she might choose differently from how she did the first time.

            Response: I fully endorse RI.  But I do not know of a good argument from RI to the denial of CDP.  CDP does not talk of a repeat of C.  It talks of her having instead been in C*.  Maybe, though, the idea is that if RI is true, then a fortiori if one had been in circumstances at all different from C, one might have acted differently.  Call this the Difference Intuition:

(DI)     Necessarily: If a created agent freely chooses A in circumstances C, and C* is any different alternative set of circumstances, then the agent might have chosen differently in C*.

            The exact rationale for the move from RI to DI is unclear.  But in any case, we can argue directly against DI.  Start with the Causal Relevance Principle:

(CRP)  Suppose that E and D occur, but were D not to have occurred, E might not have occurred.  Then whether D occurs is causally relevant to the occurrence of E.  

Now, suppose x freely chose A in C, and suppose that C has some aspect that’s causally irrelevant to x’s choice.  For instance, suppose that I let C be the whole history of the world up to my choosing to write this paper.  Then, maybe, C contains a butterfly’s hatching in Beijing, but the hatching is causally irrelevant to my writing this paper.  Let C* be counterfactual circumstances just like C without the butterfly hatching.  According to DI, in C* I might not have chosen to write this paper.  But by CRP, this contradicts the causal irrelevance of the butterfly hatching. 

            This argument shows that if DI and CRP hold, then every event in the history of the world prior to a free choice is causally relevant to that choice.  This is not so plausible. 

            Moreover if DI holds, then we are mistaken when we say ordinary things like: “You didn’t check before your trip if you had enough gas, and had I not filled up the tank the day before, you would have been stuck in the middle of nowhere.”  For the circumstances in which I had filled up the tank the day before are different from the actual circumstances, and, according to DI, in these counterfactual circumstances you might not have chosen to take the trip, or you might have chosen to check the gauge. 

            We can directly argue for CDP by using a modified version of CRP:

(11)     Necessarily: If D and E occur, but E would not have occurred had D* taken place in place of D, then D’s occurring in place of D* is positively causally relevant to E.

This modified version of CRP uses a notion of positive causal relevance as opposed to mere causal relevance.  Moreover, it differs from CRP in two other ways.  First, it claims to be a necessary truth.  Second, instead of supposing that E might not have occurred, it has in its antecedent the typically stronger supposition that E would not have occurred.  Proposition (11) is quite plausible.

            Now, to argue for CDP from (11), assume (11), suppose that conditions (a)-(d) of CDP are satisfied, but, for a reductio, also suppose that the consequent C*®F is false, where F=<x freely chooses A>.  Because C* is an appropriate antecedent, we can apply CLEM to get the claim that C*®~F.  Encapsulate all the relevant circumstances in C as a really big event D, and similarly encapsulate all the relevant circumstances in C* into the event D*.  Let E be the event of x choosing A freely.  So, D and E occur, but E wouldn’t have occurred if D didn’t occur.  Then, by (11), we conclude that D’s occurring in place of D* is positively causally relevant to E.  But this is absurd.  If anything, D’s occurring in place of D* should be negatively causally relevant to E, given that C* dominates C in respect of x’s freely choosing A.  How could the occurrence of fewer or weaker considerations in favor of A, or the non-occurrence of more or stronger considerations against A, be positively relevant to someone’s choosing A?[13]

            Objection 4: DP (and (1), CDP, CRP and (11)) involve a different kind of subjunctive conditional from that involved in DI and in Plantinga’s Molinist CCFs.  For instance, perhaps, the conditionals in DP are Lewis-Stalnaker similarity-based conditionals or maybe Edgington-style conditional-probability conditionals[14], while the Molinist CCFs that God is guided by are sui generis, and this difference in the kinds of conditionals explains the clash in intuitions between those behind DI and those behind DP.

            Response: I think this objection is plausible, but fatal to Molinism if correct.  The reason it is fatal to Molinism is that it leaves the Molinist conditionals without a sufficiently robust connection to ordinary language.  For instance, (1) is a paradigm example of the sort of conditional of free will that comes up in ordinary language.  Indeed, I submit that in ordinary language, when we talk about what people would have done, we either are making probabilistic claims on the basis of similar circumstances they were in (conditional-probability conditionals) or else we are making claims, informed by principles like CDP and (11), about what people would have done, while keeping fixed as much as possible of their actual decision-making (this might be rather like Lewis-Stalnaker conditionals).  Plantinga’s (1) is one of these last ones: we keep fixed as much as possible of Smith’s motivations for his choice when we move to the counterfactual scenario.  If Molinist CCFs are completely different from these ordinary language subjunctive conditional claims about free choices, then, I submit, we really have no idea what the Molinist CCFs mean and why they would be relevant to divine decisions.  If this is right, then the Molinist cannot deny CDP (and hence DP) without undercutting Molinism.

            Objection 5: CDP is implausible in the case of patently unreasonable choices.  For instance, suppose Jim chooses a searing pain of magnitude 9 (on a 10 point scale) in place of a burning pain of magnitude 5, even though he prefers a smaller pain to a greater and has no belief that pain will make him stronger or the like.  He just chooses that which disprefers.  Let C be the actual appropriate antecedent.  Let C* be a modified antecedent in a case where the searing pain has only magnitude 4, and the burning pain still has magnitude 5.  Given Jim’s weirdness, it is far from clear that in C*, Jim would have chosen the searing pain, even though C* dominates C in respect of choosing the searing pain. 

            Response: It is not clear whether Jim’s case is possible.  In the case described, Jim does not act on account of any considerations.  In this regard, he differs from the masochist for whom the magnitude of pain may be a consideration in favor of the pain.  Thus, if we take acting on considerations to be a necessary condition for making a choice, we need to dismiss Jim’s case as impossible.  And if we modify the case to that of the sadist, then it becomes false that C* dominates C in respect of choosing the searing pain.  If, on the other hand, we attribute to Jim an inclination to act patently unreasonably, then either the same response as in the case of the sadist works or else, as I suggested earlier, in the definition of domination we need only add the condition that the agent has no inclination to act unreasonably for its own sake. 

            Objection 6: It is possible, pace everything that has been said about the randomness objection to libertarianism, to make reasonless choices—choices that do not come from any consideration.

            Response: I can grant this and still run a variant of the argument.  Simply add an extra condition to the antecedents in the necessary conditional in DP that x is unable to make a reasonless choice.  Maybe x is psychologically incapable of that, or maybe God makes it impossible for x.  And then stipulate in the construction of my counterexample worlds that Eve is unable to make a reasonless choice.  Granted, if normally people can make reasonless choices, this will restrict Eve’s freedom.  But since she has two non-reasonless options available to her, and one of them is wrong and the other isn’t, she is still significantly free.  Our freedom typically is restricted in various ways.  A choice between punching someone who insulted me and ignoring the insult can be significantly free even if I am psychologically incapable of smiling gently in response to the insult.  What is needed is that the choice not be so restricted that there be no significant freedom left.

            The only real danger to the argument would be from a claim that agents never make significantly free choices on the basis of considerations.  But that would be an implausible doctrine.

            Objection 6: God may want more than a world where everybody is significantly free and always does what is right: he may want a world where some creature makes a significantly free choice out of duty.  In my examples of “morally perfect worlds”, Eve dances the jig or eats the apple because of the jig’s merriness or the apple’s yumminess.  Granted, she is obligated to choose as she does, because the other choice is forbidden, but it does not seem that she chooses because of the obligation.  It may still be that, given the contingent values of CCFs, God’s only way of getting a world where a creature acts significantly freely out of duty is to actualize a world where some (other or same) creature does wrong.  (This suggestion was basically made by Michael Bergmann in correspondence.)

            Response: This objection amounts to a new Molinist FWD, not based on the possibility of transworld depravity but on the possibility that the CCFs preclude God from actualizing a world where both nobody does wrong and yet some creature acts significantly freely out of duty.  A fuller evaluation of this FWD may well require a deeper analysis of the logic of CCFs as well as of the nature of motivation.  After all, the counterexample to Plantinga’s FWD has shown that not all prima facie possible combinations of truth values of CCFs are in fact possible: DP places a significant constraint on the combinatorics.  There may be other such constraints, and if one wishes to give a fully satisfactory Plantinga-style defense, the onus of proof is now on one to show that the given combination of truth values of CCFs that one posits the possibility of is in fact possible. 

            However, a further response can be made.  We can imagine an agent who has the property of motivational maximalism in respect of a decision: necessarily, when she chooses an action A, she acts on all the undefeated considerations that favor A (or disfavor non-A—I shall neglect this disjunct below).  Her rationality is such that she is simply unable to ignore any considerations that she neither is choosing against nor has a defeater for.  One might even think that all agents have motivational maximalism in their choices, but there is no need for that controversial, and probably false, assumption. 

            What I need here is that motivational maximalism is compatible with significant freedom, and this appears quite plausible.  Now, then, simply add to my stories about Eve the assumption that she has motivational maximalism and that she has a consideration in favor of doing her duty.  Then, in these refined versions of wJ and wA*, Eve acts from all her considerations in favor of the jig and the apple, respectively, and duty is among these considerations. 

            Now, maybe, this does not satisfy.  For, maybe, God would want an agent not only to choose significantly freely, but to choose it solely out of duty.  But it is not clear that this is all that desirable.  It may well be the case that, necessarily, whenever one has a duty to do something, there is some other reason for the action which a virtuous agent will also be moved by.  It is one’s duty to visit a sick friend, and virtuous agents act on their duties.  But the virtuous agent will visit her friend not just out of duty, but also out of a desire to be with her friend.  If this kind of a choice with mixed motives is the best kind of choice, then there will be no special value in choosing something solely out of duty, and so God will have no special reason to prefer a world where somebody does that.  And Eve’s motives of yumminess and merriness may well in fact be motives that a virtuous person would have: a virtuous person appreciates perceptual goods and rejoices in good circumstances—we may even, if we so wish, add a theological significance to the yumminess and merriness (enjoying and rejoicing in God’s creation, respectively).

            Or maybe the case I gave where Eve has motivational maximalism is unsatisfactory because the motivational maximalism is imposed on her by God, rather than a result of her formation of her own character.  But further wielding of DP could, perhaps, ensure that Eve earlier achieved motivational maximalism through free choices.  For instance, maybe prior to her apple/jig choice, she had a choice between a pear and a waltz, in which choice she had a consideration in favor of gaining motivational maximalism and none to the contrary, and the situation was set up so that she knew that which one of her potential choices would lead to her character changing in a way that would make her have motivational maximalism.  By choosing whether the causal connection to motivational maximalism is tied to the pear or to the waltz, God could ensure that Eve chose motivational maximalism and did so freely, at least if standard Molinism is compatible with freedom.  And then God inserts the apple/jig choice, as in my original case. 

            The point of this response is not the rejection of all possible Molinist defenses along the lines of the “out of duty” suggestion.  The point is simply to highlight that finding an alternative along these lines is not easy and requires significant additional effort going beyond Plantinga’s FWD.

5. Closing remarks

            A FWD is an argument that an omniscient, omnipotent and perfectly good being could create a world containing no evil.  If DP is true, then the most famous version of the FWD, Plantinga’s Molinist FWD, does not work.  However, non-Molinist FWDs continue to have a chance of working.  It was essential to the argument that God be in a position to choose between actualizing wJ and actualizing wA* on the basis of his knowledge of CCFs, and that required Molinism.  If, for instance, CCFs have no truth values, or they are all false, or God only knows the truth values of a few CCFs and his knowledge is explanatorily posterior to his decision which world to actualize, then the argument will fail to establish that God had the power to knowingly actualize a morally perfect world.  Indeed, Adams has offered a defense based on such assumptions.[15]  Moreover, FWDs that are not based on God’s possible inability to actualize a morally perfect world are unaffected.[16]

            Plantinga’ thesis of the possibility of transworld depravity required an appropriate independence thesis for CCFs, such as that when C and C* are logically incompatible and appropriate, then C®F and C*®G are always going to be logically independent.  DP questions such independence assumptions.  An interesting question would be to map out more of the logical structure of the space of CCFs.  Is it, for instance, the case that all combinatorial combinations of truth-values of CCFs consistent with DP, (11) and some logical axioms about counterfactuals are possible?  Or are there other substantive axioms that need to be added?  The question is interesting in itself, and important to FWDs.  It will, for  instance, be relevant to a fuller evaluation of Molinist defenses like the one discussed in the last objection in the preceding section.

            The Molinist defender of Plantinga’s FWD should probably deny CDP and DP.  In justification of this denial, she may very well be forced to accept a version of libertarianism on which free choices need not be explained, even indeterministically, by any considerations.  This would allow the Molinist to hold that CCFs enjoy a logical independence from one another, so that, for instance, if C and C* are logically incompatible and appropriate, then C®F and C*®G are logically independent.  It appears that Plantinga’s FWD requires some such logical independence thesis in order to ensure the possibility of the scenario where all the CCFs come out in a way that precludes God from actualizing a morally perfect world. 

            However, even if there is such logical independence, it is implausible that there will be probabilistic independence.  And so at the very least we have yet another argument that the hypothesis that God is precluded by CCFs from actualizing a morally perfect world is improbable.[17]

[1] Nature of Necessity (Oxford: Oxford, 1974).  In “Self-Profile” (in: Alvin Plantinga, ed. by J. E. Tomberlin and P. van Inwagen [Dordrecht: Reidel, 1985], 49-52), Plantinga seems to be attempts to give a version of the FWD that will be acceptable to non-Molinists.  However, as Richard M. Gale (On the Nature and The Existence of God [New York: Cambridge University Press, 1993], 136ff has argued, this attempt fails.  For instance, if all the relevant CCFs are necessarily false, then Plantinga’s account in “Self-Profile” implies that God cannot weakly actualize any world containing free agents.  For, to do that, God would have to cause something that counterfactually implies that the agents acts as they do, and that would require a true CCF.

[2] To reduce verbiage, I shall assume that if C®F is a CCF, so is C®~F.  This means that a double, triple, etc. negation of a proposition reporting what someone did or did not freely choose also counts as a proposition reporting what someone did or did not freely choose.

[3] Plantinga, “Self-Profile”, 49.

[4] Cf. Richard Otte, “Transworld Depravity and Unobtainable Worlds”, Philosophy and Phenomenological Research 78 (2009) 165-177.

[5] I will use the assumption that God knows these conditionals only in one place in the argument.  Even without the assumption, the conclusion follows that, necessarily, there is a state of affairs that God could actualize that is such that were God to actualize it, there would be a morally perfect world.  However, the assumption is needed to yield the claim that God knows how to identify that state of affairs—for what that state of affairs is will depend on the actual truth values of CCFs.

[6] “Middle Knowledge and the Problem of Evil”, American Philosophical Quarterly 14 (1977)  109-117.

[7] Nature of Necessity, 177.

[8] “Middle Knowledge and the Problem of Evil”, 115.

[9] For libertarian accounts that claim to satisfy this constraint, see, for instance, Randolph Clarke, “Indeterminism and Control”, American Philosophical Quarterly 32 (1995) 125-138 or Robert H. Kane, The Significance of Free Will (New York: Oxford, 1996).

[10] By (3) and (4), we have C®(C & (C®F) & ~(C*®F)).  By (6), we have necessarily: C & (C®F) if and only if C & F & (C®F).  Applying (5) and the fact that C®(C & (C®F) & ~(C*®F)), we obtain (7).

[11] “Transworld Depravity and Unobtainable Worlds”.

[12] An Essay on Divine Authority (Ithaca, NY: Cornell, 2002).

[13] For an argument very similar to this one, see Alexander R. Pruss, “Prophecy Without Middle Knowledge”, Faith and Philosophy 24 (2007) 433-457.

[14] Dorothy Edgington, “On Conditionals”, Mind 104 (1995) 235-329.

[15] “Middle Knowledge and the Problem of Evil”, 90-91.

[16] For instance, Alexander R. Pruss, “A New Free-Will Defense”, Religious Studies 39 (2003) 211-223.

[17] I am grateful Michael Almeida, Michael Bergmann and Trent Dougherty for discussions of this paper and/or its arguments.