Introduction

The end of the first episode of the BBC series ‘Sherlock’ concludes with a scene in which the villain gives Sherlock Holmes two bottles. One of them contains poisonous pills, the other harmless ones. The villain points a gun at Sherlock’s head and instructs him to pick one of the two bottles. He then pushes one forward, implying that it is the one he wants Sherlock to take. Is it a bluff? Or a double bluff? Triple bluff? Now imagine two different alterations to the scene. First, imagine that the villain was not allowed to deceive Sherlock—on this scenario, Sherlock would know perfectly well which one of them to take (the one not suggested to him by the villain). Second, suppose that the villain was omniscient—for any thought-process Sherlock could ever go through, the villain would always know which bottle Sherlock would pick. On this second scenario, is there a way that the villain’s suggestion to take one of the bottles could be considered as anything short of deception? And what if Sherlock knew that the villain was omniscient? Which bottle would he pick? Would he know that the villain knows which one he was going to pick? And would the villain know that he knows that?

The reason why Sherlock has a chance of winning in the original story is that the villain is neither omniscient nor barred from deceiving—otherwise he would not be a proper villain! I will propose that the key feature of the story—the expression of intentions about the future to agents with the aim of their actualising that future—poses a new difficulty for Molinism. In particular, I will argue that the problems Molinism has when dealing with the prophecy of free actions, highlighted in recent scholarship (Oppy & Saward, 2014; Warfield, 2009), are even more severe than has been suggested. I think these problems best come to light when considering the phenomenon of stubbornness, that is, the conscious refusal to actualise a future intended for us by another agent. The reason why some of the complications highlighted by Oppy and Saward, to which I add below, have not come to light is that the paradigmatic cases of divine prophecy of free actions (especially the story of Peter’s denial) are highly specific and cannot be used as a basis for a generalisation regarding Molinism’s capacity to incorporate divine prophecies and revelations.

In what follows, I describe three worlds, the possibility of which requires a rejection of one or more of the following claims: the claim (i) that God does not or cannot deceive, (ii) that He can reveal anything true, that is, any item of His knowledge, (iii) that He has middle knowledge, and/or (iv) that God’s natural knowledge is conceptually distinct from His middle knowledge. I then consider three objections against these worlds: the objection (a) that the counterfactuals I discuss are not the types of counterfactuals the Molinist has in mind, (b) that the worlds I discuss are impossible, and (c) that God would not make certain revelations in them.

Molinism

The gist of Molinism, at least as it has been understood in recent literature, consists of the claim that there are three elements, or moments of God’s knowledge. A full characterisation of the other concomitants of de Molina’s original position (de Molina, 1988, p. 168) may be found, amongst other places, in Flint (1998), Perszyk (2013), Hasker (1989, pp. 16–52), Kvanvig (2013), Mares and Perszyk (2011), and Fischer (2011).

The first moment is ‘natural knowledge,’ that is, ‘knowledge of all the metaphysically necessary states of affairs’ (Freddoso, 1988, p. 11), which include the set of possible worlds that God could actualise.

The second is ‘middle knowledge,’ that is, God’s knowledge of the set of possible contingent states of affairs, which, crucially for our purposes, include knowledge of ‘counterfactuals of creaturely freedom’ (CCFs), which take the following form: ‘If person P were in circumstances C, P would freely do A.’ These counterfactuals ensure, on the one hand, that God always knows what creatures would do, but, on the other hand, also that they can do it freely:

… There [has] to be a middle knowledge through which God, foreseeing what would occur through any creature’s faculty of choice on any hypothesis and in any turn of events among things, was able, from eternity … to predestine what He pleased on the part of those creatures whom He should decide to create, while preserving their freedom. (de Molina, 1988, pp. 251–2)

The third is ‘free knowledge,’ which is the knowledge that results from God’s decision to perform His creative act. Unlike God’s ‘natural’ and ‘middle’ knowledge, God’s ‘free’ knowledge is post-volitional; it is the only ‘moment’ of the three that is posterior to the creative act.Footnote 1

The picture of divine providence that emerges from Molinism is roughly the following: God first (although the precedence here is logical, not chronological) contemplates the set of possible worlds. He then observes what free creatures would do when placed in those circumstances, as captured by CCFs. This grants Him perfect knowledge of what would happen should He decide to accord the possible combination existence. Finally, He accords it existence.

Scenario 1: revealing the counterfactual

Imagine a possible world w1. It is a possible world containing an agent who is stubborn, that is, whenever the agent finds out whatever God providentially arranged for her to do, she does exactly the opposite. Let us call the agent ‘the stubborn heretic’ (SH). Furthermore, add the condition that SH is not particularly good at making logical inferences. Let us stipulate that SH finds herself in circumstances C and is about to perform a particular act x. By His middle knowledge, God knows the following CCF:

CCF1: If SH were in circumstances C, she would do x.

As the SH is on the cusp of performing x, God decides to reveal CCF1 to SH. The heretic pauses for a moment and goes through the following mental process:

Well, I am in circumstances C, and CCF1 says I would do x. But since I am stubborn, I will show God I have free will, negate CCF1, and perform non-x.

SH goes ahead and performs non-x. Obviously, this does not mean that the performance of non-x caused CCF1 to be false. It is just that CCF1 no longer captures what SH is doing. The CCFs are supposed to capture the ‘complete circumstances’ of an action (Flint, 1998, pp. 47–8; Oppy & Saward, 2014, p. 241). So when SH performs non-x, the relevant CCF known by God is the following:

CCF2: If SH were in circumstances C’ (which are the same as circumstances C, but include the revelation of CCF1), she would do non-x.

This particular possible world presents some troubling consequences for Molinism, which will become more obvious in the further articulations of the thought experiment in the next two sections.

The first is that the thought experiment seems to put into question the order linking God’s natural and middle knowledge, causing the latter to leak into the former. Originally, the two items of knowledge were supposed to be distinct and ordered. We first have divine knowledge of w1 (stubborn heretic and a revelation of a truth) on the one hand and then various CCFs on the other hand. Of course, both of these are still pre-volitional, since we are merely discussing what happens in possible worlds—God has not created them, and nothing obliges Him to do so. This is not an argument against Molinism as such, but it adds further complexity to the widely-discussed ‘grounding objection’ against the existence of CCFs (see Freddoso, 1988, pp. 68–75; Hasker, 1989, pp. 29–31, 2011, pp. 25–9; Perszyk, 2013, pp. 757–8). In the case of w1, the Molinist no longer has just the job of working out the truthmakers for CCFs, but also for CCFs containing lower-order CCFs about a different possible world (one where the lower-order ones do not get revealed): What grounds God’s knowledge of what creatures would do when placed in circumstances that are such that those circumstances include the creatures’ knowledge of what they would do had those circumstances not included it?

The second consequence is that in w1, God seems to be deceiving the heretic. God’s middle knowledge is supposed to capture all possible actions an agent would perform in any circumstances:

[I]t falls under His immense and altogether unlimited knowledge, by which He comprehends in the deepest and most eminent way whatever falls under His omnipotence, to penetrate free choice in such a way as to discern and intuit with certainty which part it is going to turn itself to by its own innate freedom. (de Molina, 1988, p. 141)

Thus, God not only knows by CCF2 that were He to reveal CCF1, SH would do non-x, but He also knows that SH would not be able to realise that CCF1 does not capture SH’s particular situation and make the inference to CCF2. By revealing CCF1 to the SH, He causes her to have a false belief, which, to say the least, is theologically problematic. Note that in this particular situation God is not lying, He is not revealing something that is not true; CCF1 remains true even after its revelation, it just becomes irrelevant. But it is misidentified by SH as being relevant.

The role that deception plays in the revelation of CCF1 is best illustrated by contrasting w1 with the case of a mother and a stubborn child. Say a mother knows her child would eat the entire cake when placed in a room with the cake on the table. She also knows the child is stubborn. This means that she is fully aware that if the child were told they would eat the cake when left alone, they wouldn’t. She says:

I can’t leave you alone in this room, because I know that if you were left alone, you would eat the cake!

with the obvious intention of using the child’s stubbornness to preserve the cake when leaving them alone. The child thinks as follows:

Oh, how well mommy knows me! However, to demonstrate to her how stubborn I really am, despite the fact that I am in these circumstances, I am not going to eat the cake.

The reason why this case is not particularly problematic (certainly not theologically) is that the mother can willingly deceive. As a matter of fact, she might even be omniscient—in the eyes of the child, anyway. Since she can deceive, she can reveal not only something that she knows to be false (the simple future contingent ‘I know you will eat the cake’) but also the true CCFa:

CCFa: If the child were left alone in the room, they would eat the cake,

since at that very moment, the relevant CCF describing the situation becomes:

CCFb: If the child were alone in the room and were aware of the truth of CCFa, they would not eat the cake.

But her whole strategy relies on the child not working out CCFb: If the child managed to do that, they would just devour the cake to assert their stubbornness by negating CCFb. It also relies on being able to deceive. If she were not allowed to do that—knowing the stubbornness of the child beforehand—any revelation of CCFa would be prohibited, since she would know the child would be unable to make the inference and form a false belief about her intentions.

Scenario 2: further revelations

Let’s imagine a different possible world, w2, which is identical to w1 up to the point of God revealing to SH CCF1, but just before SH performs non-x, God reveals to her CCF2, since God knows (from Scenario 1) that SH would fail to infer CCF2 herself. Recall that God has revealed:

CCF1: If SH were in circumstances C, she would do x.

The stubborn heretic misidentifies CCF1 as applying to her situation and decides to do non-x. But God knows this, that is, He knows:

CCF2: If SH were in circumstances C’ (which are the same as circumstances C, but include the revelation of CCF1), she would do non-x.

SH is ready to perform non-x, when God reveals CCF2 to her. Once again, SH misidentifies circumstances C’ as applying to her circumstances and, out of stubbornness, performs non-non-x. Which is simply performing x. Therefore, in w2, God knows:

CCF3: If SH were in circumstances C’’(which are circumstances C’, but include the revelation of CCF2), she would do x.

But of course, we could easily imagine God revealing CCF3 too. We, therefore, have a series of possible worlds with revelations of CCFs of growing complexities—each member of which contains the previous members—and either the performance of x or non-x.

The first difficulty each of the worlds w1…n presents is that it even further intensifies the problem of middle knowledge leaking into natural knowledge. Can there be middle knowledge about middle knowledge about middle knowledge? The grounding objection against Molinism gets multiplied: CCFs and possible worlds (known by natural knowledge) become even more intertwined. On the original Molinist picture, God was supposed to contemplate three distinct things: (i) circumstances, (ii) all essences of agents, which ground what those agents would do (Freddoso, 1988, p. 12), in our case stubbornness and inability to infer the higher-order CCF, and (iii) CCFs linking the two. In any w1…n, some circumstances include CCFs in their descriptions—knowledge of what happens in some possible worlds depends on what happens in others (with fewer revelations). Any Molinist keen to elaborate both the truthmakers and semantics for the various CCFs would have to bear this in mind.

The second difficulty may be stated as a dilemma. The series w1…n is either finite or infinite. Prima facie it seems that the series cannot be infinite. Apart from the orthodox prohibition on God revealing actual infinites, there is the fact that God’s revelation and SH’s misidentification take time. If the series has no last member, then SH never does anything, but simply receives an infinite series of revelations, God disclosing one layer of CCFs after another. The only way SH could avoid this would be by receiving the entire series w1…∞ in one timeless act. But that would make SH more like God—timeless absolute cognition of infinites is traditionally reserved for God alone (see for example Aquinas, 1952, sec. Q2a9). Working out that a CCF applies to a particular situation takes time, and this time matters (see Lacan, 1998).

However, if the series is finite, this does not fare much better for the Molinist. In that case, at some point, God no longer makes any further revelations and SH simply takes an action (x or non-x). Her action is then captured by the CCF of the higher order. But this merely converts itself to the initial scenario discussed in the previous section, where God causes SH to have a false belief about the relevant order of CCF she thinks applies to her situation. God would be revealing the following:

CCFn: If SH were in circumstances Cm (which are just circumstances Cm-1 with the revelation of CCFn-1), she would do y.

But the heretic would misidentify CCFn as applying to her situation, even though the relevant counterfactual would in fact be:

CCFn+1: If SH were in circumstances Cm+1 (which are just circumstances Cm with the revelation of CCFn), she would do non-y.

The problem of deception (i.e., divine revelation causing an agent to have a false belief) appears again since it is precisely the process of ceasing the revelations that is required for SH to act.

The Molinist might now be tempted to simplify this case by asserting that all of the above can simply be reduced to the claim that God has just one item of knowledge, namely:

CCFn+1: If SH were in circumstances Cm+1 (which are just circumstances Cm with the revelation of CCFn), she would do the opposite than whatever is specified by the consequent of CCFn.

But, unfortunately, the stacking of the counterfactuals is also more complicated than I have made it to seem. For possible worlds where SH can only choose to do one of two possible actions (e.g., her actions are limited to pressing either button A or button B), that might work. In such worlds, SH resembles Buridan’s ass who is being made to starve by God revealing further and further CCFs. But we can easily throw SH into worlds with much more complex circumstances. Recall my earlier claim that God knows that if He reveals

CCF1: If SH were in circumstances C, she would do x,

SH would do non-x. But this way of formulating it was highly inaccurate. It is not only that SH would do non-x (or in other words, not do x), it is rather that she would do a particular y, which is not x. And it is y that God must know. This is frequently emphasised by Molina against Bañezians, especially when it comes to God’s knowing which evil acts would be performed by agents when placed in different circumstances (de Molina, 1988, pp. 139, 253). Freddoso comments:

It is not enough simply to show how God knows that a good effect will not obtain; one must also show how God knows exactly which evil effects will obtain instead. (Freddoso, 1988, p. 39)

And the same applies to morally indifferent acts. To use the earlier analogy of the child, God must know not merely that if the child were told they would eat the cake, they would not eat it—He must also know what the child would do instead. And in the case of SH, for every CCF of every order, God must know in particular what the SH would do. If they try to avoid divine providence and attempt to negate CCF82 by going to the park and CCF83 by not going to the park but heading to the cinema—God must know all that. And this does not look like something that could be revealed to the SH. It does not even seem like something that can be known by anyone but God. And I have my doubts about that too, since the Molinist would now need to provide a truthmaker (clearly a very different one) for each one of the CCFs. If it indeed turns out that the SH decides to perform something else at every level (and it does not seem to be impossible for there to be a possible essence of an erratic heretic—resembling Luke Rhinehart’s Dice Man—committed to avoiding divine providence in a different way every time), then the truthmaker for each level would need to reflect the difference from the other levels. I confess I cannot imagine how such a truthmaker could be provided.

Scenario 3: the clever heretic

Before summarising and addressing objections against the previous thought experiments, let us look at the last reformulation of the situation. Let’s put God revealing further CCFs aside and simply use the first scenario (w1), but instead of stipulating that SH is not great at logical inferences, let us grant her exceptional logical skills. SH has studied theology, read Lewis’ Counterfactuals, and is thoroughly acquainted with Molinism. She has also read the first section of this essay, saw how the silly heretic was tricked earlier and is intent on not misidentifying the CCF applicable to her situation. SH wants to work out the truly relevant CCF. How would she go about doing it?

SH is in circumstances C. As in the first case, God reveals:

CCF1: If SH were in circumstances C, she would do x.

But now, instead of mistakenly thinking that CCF1 applies to her situation, SH thinks thus:

Would performing x be in line with my commitment to stubbornness? Well, it turns out, I would certainly not be abandoning my stubbornness if I performed x, since the circumstances specified in CCF1 no longer obtain—God has now made a revelation, and that is not included in the antecedent of CCF1. I have also read in the Concordia that God is omniscient, i.e., He must know that I would work this out. So do I do x or not? Either way, God knows. So He either knows CCFi: ‘If SH were placed in circumstances C’ (circumstances C + revelation of CCF1), she would do x’ or CCFii: ‘If SH were placed in circumstances C’ (circumstances C + revelation of CCF1), she would do non-x.’Footnote 2 But how do I know which one of these is true?

Of course, the Molinist is committed to the claim that one and only one of these must be true. The SH then thinks about providence and argues as follows:

God is omniscient, so He must know which CCF is right. I have also been told that God does not deceive or lie and that each one of His actions is part of His providential plan. As Freddoso says: ‘Since God is the perfect artisan, not even the most trivial details escape His providential decrees.’ (Freddoso, 1988, p. 3) If God had not revealed anything to me, then I know I would do x (since if that was not the case, CCF1 would be false and I know God does not lie). But God did reveal something. There must be a reason! But this must mean CCFii is true—for otherwise the revelation of CCF1 would be superfluous.

Is SH’s reasoning correct? It seems to me that it is. If it was not, then the following counterfactuals would feature in the set of God’s middle knowledge:

CCF1: If SH were in circumstances C, she would do x.

CCFi: If SH were placed in circumstances C and revealed CCF1, she would do x.

But this seems to conflict with the key parameters of the situation set out earlier. Not only would it make the divine revelation of the true CCF1 completely superfluous, but it would also seem to conflict with SH’s free will. Recall that SH is committed to stubbornly avoiding whatever it is that is revealed about herself by God. The divine revelation of the counterfactuals underlying a particular situation is supposed to be a cause for change in SH’s behaviour in that situation, perhaps the only cause of change.

Of course, it could be objected that x starts being a different act in virtue of a different causal chain leading up to it—crying ‘wolf!’ when there is no wolf is an action different from an action when there is one (see García-Encinas, 2003). While this objection may be open to the Molinist, it certainly contradicts what Molina says himself. In Disputation 33 and 53 (part 2, Sect. 13), he makes an argument to the effect that

numerically one and the same act in the natural order, elicited here and now, is indifferently able to be morally good or morally evil with the alteration of just a single circumstance that does not change the identity of that act within the natural order. (de Molina, 1988, p. 221)

As Freddoso comments on the passage above:

[Molina] assumes that an action … is distinct from at least some of the action’s circumstances, so that the very same action that in fact occurred in, say, circumstances C might have occurred in circumstances C* (where C is distinct from C*). (de Molina, 1988, pp. 221, footnote 15)

This means that SH’s reasoning is correct and, when revealed CCF1, she knows that the higher-order counterfactual is CCFii.

But then, if SH was able to make the inference from CCF1 to CCFii, which implies God knows she is about to perform something that is not the x in the consequent of CCF1, SH could simply run the same process again and infer also the higher-level counterfactual which includes the inference to CCFii in its antecedent. And so on. In that case, we end up with something quite similar to the earlier dilemma implied by God providing further revelations, except that in this case, the revelation of subsequent higher-order CCFs is replaced with SH’s working out what they are. Either the series of these inferences is infinite or not. If it is, then SH never performs any action and spends her life making inferences to ever-higher CCFs—and we also get the unwelcome consequence of God creating an actual infinite. If it is not, then SH stops at some point and fails to make the inference to the higher-level CCF. God would know that His revelation of the original CCF1 would, at some point further down the line, result in SH failing to make that inference—once again, the revelation of CCF1 would result in a false belief.

Apart from a further complication the Molinist needs to address when dealing with the grounding objection (what grounds the truth of CCFii in the way it does not ground CCFi?), the situation above poses a problem that the previous two cases did not. What is the basis of God’s knowledge of the point at which SH fails to make the higher-order inference? Or in other words, how can God know:

CCFx: If SH were in circumstances C’ and revealed CCF1, she would keep making inferences to higher-order CCFs and stop with CCFn.

The idea that God would know at which point the SH would stop the inference-making process is very strange. The grounds for the fact of SH ceasing to make further inferences do not feature in the description of the circumstances or of the agent. The circumstances C’ are simple: SH, C, and a revelation. In order for a CCF to capture the link between those and an inference stopping at CCFn, we would need to include some sort of tendency on the part of the stubborn agent to cease inferring logically after a certain number of operations. But then, the SH can also work this out (just like I did now) and construct a super-computer to get more inferring power. In either case, the existence of the stacked series of counterfactuals is supposed to arise out of the circumstances of the revelation of CCF1 to the agent, not from the agent herself.

Of course, it might be objected that there is a hidden CCF* that forms part of God’s knowledge:

CCF*: If SH were in circumstances C and revealed CCF1 and made the inference to CCFii, then she would do x.

But this objection is otiose. God still needs to know whether SH could make the inference to the higher-order CCF. If she can, then we’re back to square one. If she can’t, we get the same situation that we discussed at the beginning of the essay. Furthermore, the SH in CCF*, CCFi, and CCFii is supposed to be the same agent. CCFs are supposed to capture the essence of the person and then describe what they do—stating that the essence of the agent includes a CCF would seem to be begging the question. Earlier we saw that w1 entails middle knowledge leaking into natural knowledge. Here we seem to be getting middle knowledge leaking into the very essences of agents that middle knowledge is about.

A good contrasting case with this situation is the Sherlock one cited at the beginning of this paper. The reason why that situation faces none of these problems is that Sherlock knows the criminal is not omniscient. There is no CCFx capturing where his inferences might stop. Sherlock knows that his knowledge of where the villain thinks he might stop his inferences is not infinite. His confidence when choosing one of the bottles is grounded in his confidence that he knows better. But this clearly cannot be the case with the omniscient God, who always knows better than any agent. And any good Molinist agent knows this.

Peter and Oedipus

All of the cases from above can also be contrasted with the case of the revelation to Peter that he would deny Christ (see de Molina, 1988, pp. 145–6, 151–2, 158; Edidin & Normore, 1982; Flint, 1998, pp. 197–211) and the prophecy given to Oedipus that he would kill his father and sleep with his mother (see Flint, 1998, p. 205). The reason why the Molinist can consistently explain the revelation of the true simple future contingent that Peter will deny Christ is that such a revelation plays no role in Peter’s action when he denies Christ. There is a sufficient temporal and causal gap between the action’s prophecy and the action prophesied that ensures knowledge of the prophecy does not feature in the immediate circumstances of Peter’s denial. He just forgot. And if he hadn’t forgotten, he does not seem as stubborn as our heretics, maybe complacent at worst. That is why he weeps when he remembers what the prophecy had been (Matthew 26:75). His action would have doubtless been different had he written the prophecy on the back of his hand with a permanent markerFootnote 3; but then if Christ knew he would do that, he would not have revealed the simple future contingent. The prophecy itself is also not a CCF, although it is, of course, guided by the relevant CCF. Specifically, Christ’s revelation that Peter will deny Christ is only made possible by Christ’s knowledge of

CCFp: If Peter were told a prophecy about his denial, his complacency would be such that he would forget it by the following morning.

The reason why the prophecy to Peter is a special case is that (i) although Christ reveals something true, something that is part of His knowledge, it is not a CCF, and (ii) the revelation relies on Peter’s forgetfulness. Had Peter not been forgetful, Christ would not have revealed this (Freddoso, 1988, pp. 60–61).

In the case of Oedipus, we do get Oedipus’ constant remembrance of the prophecy—in fact, it is precisely the prophecy itself that guides him into ruin. But the story is consistent because Oedipus (like Sherlock) feels confident he can outwit the Oracle: the Oracle knows that Oedipus thinks it could be wrong, which is what causes him to fulfil the prophecy. Our clever stubborn heretics are not so—they may be stubborn, but they still know that God is omniscient and does not lie. They know that had He revealed to them a simple future contingent (‘you will do x’), no matter how much they tried avoiding it, the course of events would always result in their performing x.

Contrary to the case of Peter and Oedipus, the revelations to the stubborn heretics may be called ‘immediate counterfactual revelations’—they are revelations of counterfactuals to agents in the course of performing actions described by those counterfactuals and in the circumstances described by them. We encounter these in everyday life and we can know them with certainty (see Leftow, 1991, pp. 256–50), though not with the absolute certainty that is available to God (Freddoso, 1988, p. 52). However, their effectiveness in achieving the ends we want to achieve by our revelation of them relies on the inability of the agent to make the inference to the higher-order counterfactual. I tell a friend, as we are queueing to the cake section of a coffee shop, ‘Ah, I know what you’re like! If you were in a coffee shop just like this one, you would buy the last piece of cake,’ and it is this utterance that causes them (out of stubbornness) to refrain from snatching away the last Esterházy on the shelf. I know my revelation will be efficacious because I know they would fail to infer we are only telling them because we want the last slice ourselves.

In the descriptions of w1 and w2, God seems to be pushed into a corner—when the CCF that captures the world including its revelation is revealed, it either results in God’s deception of the agent (the agent misidentifying which CCF applies to her own situation), the CCF’s irrelevance (since when it is revealed, it describes a different possible world) and/or the revelation’s inefficaciousness (if SH makes the correct inference, it does not change a thing about what she in fact does). Of course, the Molinist can just agree that God is allowed to deceive if the deception is part of His overall providential plan (see Edidin & Normore, 1982, pp. 186–7). This is not only theologically suspect but also slightly out of touch with what is happening in w1 and w2, neither of which specifies what the providential plan for SH actually is. Without a further specification of the reasons pushing one to deceive (reasons not contained in any of the stories above), one should assume, prima facie that deception is not the sort of thing God does.

Objection: the antecedent of CCF1 is incomplete

The Molinist might protest that the three stories I have told above are incoherent. Observe that the second and third articulations of the argument rely on there being something theologically or logically nefarious with the initial revelation of CCF1 in Scenario 1, something that gets carried over into the further versions of the argument. If the Molinst can show that CCF1 cannot be revealed or that its revelation does not entail the consequences I have sketched, then the entire argument can be stopped in its tracks.

The main way the Molinist might want to do this is by arguing that CCF1 is not a genuine CCF. Its antecedent is incomplete. It does not specify the full conditions of the situation, including whether SH is aware of the truth of CCF1 or not. For example, commenting on the prophecy to Peter discussed above, Flint says:

What are the circumstances in which an action is performed? As we have seen, Molinists typically respond that the circumstances should be thought of as including the entire causal history of the world prior to the time specified by the consequent, along with whatever influences are acting on the agent at that time. But if the entire causal history is included in C, then it would seem that C must include Jesus’ prophesying that Peter will raise his arm. (Flint, 1998, pp. 199–200)

The C in CCF1 contains nothing of the sort. Since nothing commits the Molinist to the truth of incomplete counterfactuals (let alone knowledge of them or revelation of them by God), no problem arises. We should expect God to reveal something like this:

CCF1(aware): If SH were in circumstances C and were aware of the revelation of the counterfactual just being revealed to them, she would do x.

For SH from the initial formulation of the scenario, the further specification of awareness of the revelation would in no way change her behaviour: she would stubbornly decide to do non-x. But, as the Molinist would point out, this option is impossible, since CCF1(aware) would not actually be true upon revelation, and God would thus never reveal it. Recall the earlier analogy with the queue in the coffee shop. Suppose I tell my friend:

Ah, I know what you’re like! If you were in a coffee shop just like this one and I told you what I’m telling you right now, you would buy the last piece of cake.

If my stubborn friend does not work out why I’m telling them this, then they go ahead and refrain from buying the cake. But in that case, what I told them was a lie: they did not actually buy the cake. God cannot lie. The Molinist would thus block the possibility of the revelation of CCF1(aware). If revealing CCF1(aware) is off the table, then the further articulations are off the table too.

One way of refuting this objection consists in simply rejecting the claim that the antecedent of CCF1 is incomplete. When CCF1 is revealed in the first scenario (the one where SH misidentifies the application of CCF1 to her), its antecedent is complete: it just applies to a completely different possible world. When God reveals:

CCF1: If SH were in circumstances C, she would do x,

the circumstances C specified in the antecedent do not include the awareness of the revelation of CCF1. This is because the possible world where CCF1 is true includes no revelation, and hence no awareness on the part of the agent and no need to include it in the specification of the circumstances. If the Molinist still asked for a more detailed specification of the antecedent of CCF1 in Scenario 1, it would look something like this:

CCF1: If SH was in circumstances C and was not aware of the counterfactual just being revealed to her (since in circumstances C where CCF1 is true no revelation takes place), she would do x.

In Scenario 1, which does include a revelation, CCF1 remains true and God can thus reveal it. So although the Molinist is right about CCF1(aware) being unrevealable (since by revealing it to SH, God would be revealing something false), the objection does not touch the claim that CCF1 is revealable, since its antecedent applies to a merely possible situation: not the one in which a revelation is actually happening. This is analogous to me telling my friend the true claim that if they were in this coffee shop and I wasn’t there telling them anything, they would buy the cake. So the argument still stands: if SH misidentifies CCF1 as applying to their situation (despite the rather obvious nudge pointing out that the revelation is somehow relevant), we get Scenario 1. If she doesn’t misidentify it and works out what the relevant CCF is, we get Scenario 3.

In more general terms, the power of this objection depends on one’s decision about how detailed the specification of the revelation has to be. This decision does not have that much to do with Molinism. Could God ever reveal the complete circumstances required by the Molinist? The complete circumstances of my writing this paper include all sorts of stuff, which I do not know and might never know: all the digits of pi, the birthplace of the person just passing outside my window, a particular unconscious bias, or the proof of the Goldbach theorem. Of course, God knows it all. All the antecedents of all the CCFs discussed in this paper include all the digits of pi. But could He reveal this to a finite mind? What form would that revelation take? Even a requirement of God revealing all the relevant circumstances does not help: as we have just seen, we would expect the awareness or unawareness of the revelation to be highly relevant, and yet its inclusion in the complete circumstances being revealed does not block the argument.

Objection: the worlds are impossible

The Molinist may rephrase the argument by stating instead that w1 and w2 are impossible and therefore do not form part of God’s ‘natural knowledge.’ Initially, this seems false. God’s natural knowledge is supposed to contain all possible courses of events (de Molina, 1988, p. 209). If I can conceive of a world (just like I did a couple of pages back), it should form part of God’s natural knowledge. They might not be worlds that God may want to actualise, but God must know them in order to exclude them from the set of candidates for actualisation.

However, the Molinist can strengthen her claim by saying that the worlds might be possible, but that they are excluded from the set of worlds that God could actualise. They are excluded from God’s ‘creation situation’ (Freddoso, 1988, pp. 48–50), since they compromise divine ability to reveal truths. W1 and w2 would thus fall into the same category as worlds where God performs a completely gratuitous evil act. God knows what goes on in those worlds, but would never actualise them.Footnote 4

The Molinist is right about this. But middle knowledge still remains the culprit. Consider the way we have constructed w1 and w2. We started with a simple description of a possible world—one with a stubborn heretic who, when told what she would do, does the opposite. So far so good.

We then added the ingredient of God being able to reveal any true proposition that forms part of His knowledge. Again, still no problem: God can reveal lots of things, about the heretics’ past and also about their futures—if God reveals they will do something, then they will in fact do it, most probably because they will have forgotten about the prophecy. They merely act like Peter. Note that I have nowhere claimed that when SH finds out from God that she will do something, she will not do it. That is impossible: if she was not going to do it, God would not reveal that she would, since that would imply that God can reveal a lie. The peculiarity of CCFs being revealed in w1 and w2 is that they deceive even though they are true.

We then added counterfactuals of creaturely freedom into the set of truths that God knows (and that, in the earlier stage, we had permitted Him to reveal whenever He wants). And it is at that point where w1 and w2 cause problems. In the case of w1, we get SH being deceived. In w2…n we either get an infinite regress or simply SH being deceived all over again. When God reveals the CCFs, they continue to be true even when the heretic avoids the action in their consequents: in the case of the heretic from Scenario 1 since she performs an action described by a CCF of a higher order; in the case of the heretic from Scenarios 2 and 3, she does something described by a CCF of a much higher order.

So if we run into difficulties, it is because we have included CCFs in the set of truths God can reveal. Of course, we could try to exclude w1 and w2 from a creation situation, because when combined with the totality of revealable CCFs, they lead to divine deception. But that would either give middle knowledge explanatory priority over natural knowledge, which is certainly not in keeping with the original picture of Molinism, or further strengthen the problem of middle knowledge leaking into natural knowledge mentioned earlier.Footnote 5

Objection: God would never make such revelations

The Molinist may then rephrase the argument from Objection 1 into a more general form by simply claiming that although the worlds I have described are possible, and that God would (possibly) be allowed to prophesy lots of things in them, He would never reveal these ‘immediate counterfactuals.’ Oppy and Saward consider the objection with regards to ‘prophecies to the effect that certain agents would freely perform certain actions in certain circumstances.’ (see also Flint, 1998; Oppy & Saward, 2014, p. 240).

Molina himself was aware of the perils connected with the possibility that humans might know with certainty about future contingents that apply to themselves:

Thus, from the fact that future contingents are known by God with certainty it does not follow that they are going to occur necessarily in reality—in the way that this would follow if our cognition of those things were certain or if the propositions we form about those things were determinately true. (de Molina, 1988, p. 193)

In the case of our knowledge of simple future contingents, this is certainly true. If I know with necessity that I will do something tomorrow (putting aside questions of whether something like this can be known by me), it seems inconceivable that I’d be free with respect to doing it.

In a remarkable passage in the Concordia Molina also develops an argument that does not apply merely to simple future contingents, but also to counterfactuals about God Himself, i.e., conditionals of divine freedom (discussed extensively in Flint, 1998, pp. 54–63). Molina argues that God knows through His middle knowledge what each agent would do in particular circumstances—except Himself. If He did, He would not be free with respect to His creative act:

[W]e … do not concede that through His natural or middle knowledge (which we deny of Him in this regard) God sees, before the determination of His will, which part He Himself is going to choose. … Now in order that you might understand this point better, notice that it is one thing for a suppositum, because of its pre-eminence over another suppositum, to know through middle knowledge what is going to be chosen by that other suppositum by virtue of its freedom; it is a far different thing for one and the same suppositum to foreknow through middle knowledge what it itself is freely going to choose. (de Molina, 1988, p. 171)

Freddoso comments:

[Molina contends that] God does not know prevolitionally what He Himself would choose to do in any creation situation, including the actual one. For His cognitive power does not surpass His own nature in the way that it surpasses creaturely natures. … Molinists are burdened with the task of finding some way to explain the fact that God has prevolitional cognition of the free actions of creatures but lacks prevolitional cognition of His own free actions. (Freddoso, 1988, pp. 52–3)

W1 and w2 add to this burden. Because if God is prohibited from having certain middle knowledge about Himself (due to Molina’s dictum that agents would not be free had they had certain middle knowledge about themselves), but permitted to have middle knowledge about creatures, we could only satisfy Molina’s dictum by prohibiting God from revealing it to the agents in question.

But the Molinist would need to find some sort of independent reason for this prohibition—the fact that it causes complications for the doctrine of middle knowledge is not a very good one. Trading off God’s ability to reveal something true (that continues being true once it is revealed) for the inclusion of CCFs in God’s prevolitional cognitive package is not particularly economical. This is even more so when we consider that divine revelation of CCFs to non-stubborn agents (misidentification-prone or not) would actually be highly desirable, a bit like when we reassure our children with: ‘I always knew that if someone asked you to help a person in need, you’d help them. Your friend is in need, so I am certain you will help them.’Footnote 6 If that is true, then it seems that it is purely the existence of a stubborn person in w1 and w2 that limits which CCFs He might and might not reveal, which, again is rather strange. God’s ability to prophesy something should not be limited by the nature of the agent to whom it is prophesied, especially if this prohibition is motivated purely by an attempt to retain middle knowledge.

Conclusion

By way of conclusion, the final objection to consider is one that says that the worlds above might be possible, God might even be permitted to make the revelations of immediate counterfactuals in them, but that they are just worlds He would never create.

This objection cannot be refuted by pointing out that there are some worlds that God cannot actualise. For example, He is not free to actualise a world in which Adam is placed in the garden and does not sin (see Freddoso, 1988, pp. 48–50). This claim would merely refute the connected claim that God cannot create w1 and w2 and ensure that SH does not behave in those worlds in the ways I have sketched out above.

The objection misses the point. Of course, God would not create these worlds. It does not seem like He did, anyway. The problem here is not that these worlds might not be created, but rather that God’s knowledge of what would happen in these possible worlds, if they were created, which was supposed to be guaranteed by the machinery of middle knowledge, comes at a cost. The Molinist either has to sacrifice God’s ability to reveal a great deal of true things (and also explain what the difference is between some that He can and some that He can’t reveal) or give Him the power to do so for the cost of being deceptive, for leading stubborn people to have false beliefs about what was in fact revealed to them. In such a trade-off, getting rid of middle knowledge altogether seems like the best deal one can get.