Open Theism

As I noted above, the traditional notion of divine providence involves two components: complete, certain foreknowledge and specific sovereignty. The Open Theist rethinks both components of this traditional notion of divine providence. According to the Openist, God cannot have complete, certain foreknowledge about free human actions. He argues that God's foreknowledge of a human action precludes the action's being free.Footnote ⁴ Consider the following proposition that describes a future free action: My sister will vote for the Democratic candidate in the next US presidential election. According to the Openist, if this proposition were true and God knew that it was true, my sister's vote would not be a free action on her part. Therefore, the Openist concludes that if this proposition describes a free action, it cannot be true prior to my sister's performing the action and, therefore, God cannot know it prior to my sister's performing the action. Of course, God could have foreknown all human actions by determining the actions as opposed to creating creatures endowed with freedom. But Hasker, on behalf of Openists, writes, ‘[I]t seems to us that God found such a world less desirable – less appealing to his creative goodness – than a world which contains genuinely free creatures.’Footnote ⁵ As long as God in his goodness creates free human creatures, he will lack complete foreknowledge, says the Openist.

The Openist's revised conception of the traditional notion of foreknowledge results in a similar revision to the traditional notion of divine sovereignty. Because God lacks foreknowledge of free human actions, he cannot confidently design his plan for the world around his knowledge of what human beings will freely do. The Openist thinks that God has the resources to make a good guess as to what free creatures will do in particular circumstances, but free creatures may not always conform to God's conjectures. As a result, God's sovereignty does not extend to certain specific features of the world, importantly, free human actions. As with foreknowledge, God could have specific sovereignty over the world by choosing not to create a world with free creatures. But again, the Openist suggests that God would surely not prefer such a world.

Instead, God in his goodness has chosen to create a world with free creatures. But if God creates creatures endowed with libertarian freedom, then he must accept that he cannot shape the future to the degree that it is shaped by the free actions of these creatures. While the Openist maintains that God has control over the general course of the world, he does not exercise sovereignty over the specifics when the specifics include free human actions.

The Openist's revision of the notion of divine providence results in a God who is a risk-taker. Hasker defines this feature of God as follows: ‘God takes risks if he makes decisions that depend for their outcomes on the responses of free creatures in which the decisions themselves are not informed by knowledge of the outcome.’Footnote ⁷ For example, let's suppose that it was God's plan for Moses to free the Israelites from slavery in Egypt. Let's suppose, also, that Moses was endowed with libertarian freedom; when God asked him to return to Egypt to free the Israelites, he could have either gone or not gone. If Openism is true, then God had no foreknowledge of what Moses would in fact choose to do. So God took a risk in giving Moses freedom; God's plan might have been thwarted if Moses had chosen not to do as God asked.

One might expect the Openist to be uncomfortable with the idea of God as a risk-taker; some Openists do count this as a cost of their view. But Hasker embraces the idea of God as a risk-taker. Just as the boldness and courage of a risk-taker is admired in human beings, so, too, we should admire this feature in God. Nonetheless, Hasker is quick to point out that God's risk-taking does not mean that there is no sense in which God has providential control over the course of human history. While God's sovereignty does not extend to the details of each particular free human action, he does exercise sovereignty over the world in terms of ‘general strategies’.Footnote ⁸ According to Hasker, these are ‘strategies which are, as a whole, ordered for the good of the creation, but whose detailed consequences are not foreseen or intended by God prior to the decision to adopt them’.Footnote ⁹ So the Openist maintains that while God may not have control over every minute detail of the world he created which contains free creatures, he is still able to guide the world towards his grand goal.

How is it that God accomplishes this general providence since he lacks foreknowledge of free human actions and, as a result, control over such actions? While Hasker acknowledges that God's precise mechanisms for carrying out his general strategy are largely unknown, he cautions that we ‘should not underestimate the tremendous resourcefulness of God in adapting his responses to human actions – even willful and disobedient human actions – so as to achieve his wise and loving purposes’.Footnote ¹⁰ Importantly, God has knowledge of many would-probably conditionals with respect to free human actions. He knows with respect to a given free agent, a given circumstance, and a given action what the probability is of the agent performing the action under the circumstances. His knowledge of such probabilities, so the Openist claims, gives him ample help in guiding the world towards his ends. While he might not know with certainty what a free agent will do under a given set of circumstances, he has a good guess and he will get it right enough times to accomplish his ends. Moreover, when he doesn't get it right because the agent acts against the probabilities (as is a free agent's prerogative), Hasker reminds us of God's endless resourcefulness in revamping his original plan to achieve his intended end.

Would-probably conditionals

We have seen that an important element of the Openist account of divine providence is God's knowledge of would-probably conditionals. If God had no idea what an agent would probably do in a given circumstance, then he would be further limited (beyond the limitation that stems from not having complete foreknowledge) in his ability to govern the course of the world. Each time a free agent acts, it would be a total surprise to God which choice he made. This would require God constantly to readjust his plan in order to get the world tracking towards his end. Without knowledge of the would-probably conditionals, it seems far less likely that God would be able to guide the world providentially than if God knew such conditionals even though free agents do not always act as they probably would.

Before we consider what grounds these would-probably conditionals, it is important to understand what would-probably conditionals claim. Consider the following example:

Given the wording of the would-probably conditional, it is initially tempting to symbolize a generic would-probably conditional as follows:

However, it is unclear how one can evaluate the probability of y apart from x. So a more elucidating way to symbolize the would-probably conditional might be as follows:

where P stands for ‘probability’. It seems to me that, for the sake of clarity, would-probably conditionals ought to be understood as probability claims of the form mentioned in (4). So (2) should be understood as follows:

where D stands for ‘my sister votes for the Democratic candidate’ and V stands for ‘my sister votes in the next US presidential election’. Moreover, it seems that everything else upon which the probability of D depends should be seen as built into V. For example, V will include no relevant change to my sister's character, the past history of the world including my sister's voting record and her past political involvement, the laws of nature, etc. So we should understand the V to be ‘fully loaded’; that is, it includes anything relevant to assessing the probability of D.

Reducing would-probably conditionals to probability claims brings to the forefront the difficult notion of probability that is employed by would-probably conditionals. What sort of probability does the Openist have in mind in attributing to God knowledge of the probability of a free agent performing an action in a given circumstance? According to D. H. Mellor, there are three main kinds of probability and a variety of interpretations of each.Footnote ¹¹ The three main kinds of probability that he identifies are (1) physical probability, (2) epistemic probability, and (3) subjective probability. Physical probability, or chance, is the objective likelihood of an event occurring. It is an objective aspect of the world regardless of whether we know the chance of the event occurring or even could know the chance. For example, the chance (i.e. physical probability) of a coin landing on heads is .5. Epistemic probability is the likelihood of an event occurring given the available evidence. Events with a high physical probability might have a low epistemic probability if we do not have access to the evidence which indicates that the event is likely to occur. Moreover, events with a low physical probability might have a high epistemic probability if we only have access to the evidence that seems to confirm the likelihood of those events (and we are missing the critical piece of data that indicates that the event is unlikely). A common example of epistemic probability is the likelihood of a particular suspect committing a crime given the available evidence. Subjective probability, or credence, has to do with how strongly we believe the proposition in question. A proposition can have a high subjective probability even if one believes the proposition against the evidence and the proposition, in fact, has a low physical probability. For example, a football fan might believe that it is likely that her favourite team will win their next football game despite the evidence that the team is struggling or any facts about the objective probability of their winning the game. Subjective probabilities are not features of the proposition that they are about; they are features of the believer.

So what sort of probability does the Openist have in mind in attributing to God knowledge of the probability of a free agent performing an action in a given circumstance? It seems to me that the Openist must attribute to God knowledge of the genuine chances of a given event taking place under specified circumstances; that is, God's knowledge of would-probably conditionals is knowledge of the physical probability of events under specified conditions. It seems that God's knowledge of would-probably conditionals could be helpful in exercising his sovereignty only if it is ultimately knowledge of the physical probability of the event occurring under the circumstances.

Given that God's knowledge of would-probably conditionals is fundamentally knowledge of physical probabilities (i.e. chances), it is important to understand these probabilities. It seems that there are two common interpretations of physical probability: one in terms of frequency, the other in terms of propensities. According to the frequency interpretation, probabilities do not measure the possibility of a single event given a particular circumstance, but the frequency of that event given the particular circumstance. To illustrate, Mellor refers back to the coin-tossing case. ‘[The frequency interpretation] takes the probability of tosses landing heads to measure not the possibility of that outcome on one toss but how often it occurs on many tosses.’Footnote ¹² According to this view, it is the frequency of the event occurring under the circumstances that ‘fixes’ its probability.Footnote ¹³

One result of the frequency interpretation of probability is that there is no probability value assigned to a single event. For example, what is the physical probability of a particular coin-toss coming up heads? You might have thought the probability is .5. However, there is only one instance of this particular coin-toss. So the probability with which it lands on heads will either be 1 (if it lands on heads) or 0 (if it lands on tails). Proponents of this view take this to be a reason to think that there are no physical probabilities associated with single events.

If there are no physical probabilities associated with single events, then the frequency interpretation of physical probabilities will not be very helpful to the Openist. The would-probably conditionals that God knows are in many (if not most or even all) cases probability claims with respect to single events. Consider our example: if my sister were to vote in the next US presidential election, she would probably vote for the Democratic candidate. My sister (assuming that the polling workers do their jobs) will only vote one time in the next US presidential election. So there will be no probability of her voting for the Democratic candidate in those circumstances if probability requires multiple instances of the same event in the same circumstances. If we understand the circumstances to be fully loaded (as discussed above), the circumstances under which any event takes place will plausibly only occur one time. So it seems to me that the Openist needs an interpretation of physical probability that allows single events to have probability assignments.

The Openist might reply that there is some sense in which the events described in would-probably claims are repeatable. While my sister will presumably vote, at maximum, one time in the next US presidential election, there will presumably be a number of votes cast by people that are a lot like my sister. Moreover, there have been in the past a number of votes cast by people like my sister (including my sister). If these sister-like votes are cast for Democratic candidates with a high relative frequency, then couldn't this pattern form the basis for a frequency interpretation of the would-probably conditional about my sister and her voting practices? These sister-like circumstances will not be the fully loaded circumstances of the would-probably conditional, but they might include enough of the relevant circumstances so that they could be included in a frequency understanding of the probability of my sister's vote being cast for the Democratic candidate in relevantly similar circumstances.

I do think that this is a way in which the Openist could embrace a frequency interpretation of would-probably claims. However, there is a good reason to reject frequency interpretations of probability claims that is not specific to would-probably claims. One problem with frequency interpretations that have a finite number of instances is the possibility of an odd run of outcomes. For example, suppose that you flip a coin 100,000 times and it lands on heads 30,000 times and tails 70,000 times. This does not make the probability of the coin landing on heads .3, even though that is what the frequency with which it has landed on heads suggests. Instead, we should conclude that you have had an odd run of outcomes. Likewise, it could be that Democratic votes in sister-like circumstances have been rare. But it seems unjustified to conclude that the likelihood of a Democratic vote in a sister-like circumstance is low; it may be the case that we have had a strange run of outcomes. Because of the possibility of these odd outcomes, one should reject the frequency interpretation of probability when the number of instances is finite.Footnote ¹⁴

The second common interpretation of physical probability is the propensity interpretation. According to this interpretation, physical probability is defined by an event's disposition to produce a particular outcome. For example, the probability of a coin toss landing on heads just is the disposition of that coin toss to land on heads. What sort of disposition will fix the probability of a coin toss landing on heads? Mellor answers that ‘it is a disposition to produce a limiting frequency’.Footnote ¹⁵ In other words, it is a disposition of an event to produce a certain outcome x per cent of the time if that event were repeated endlessly. Returning to our example, we should understand the claim that P (my sister votes for the Democratic candidate/my sister votes in the next US presidential election) > .5 as the claim that my sister's vote has the disposition to produce a limiting frequency of greater than .5. In other words, my sister's vote has a disposition that indicates that if the circumstances in which she votes in the next US presidential election were repeated endlessly, her vote would be cast for the Democratic candidate at a limiting frequency approaching infinity that is greater than .5.

The propensity interpretation of physical probability avoids the problem that was raised for the frequency interpretation, namely, that single events cannot have probability assignments. If an event has a disposition to produce a limiting frequency, such a disposition can fix the probability of the event even if the event is, in fact, never repeated or unrepeatable.

According to Mellor, this disposition is not a brute fact about the relevant event. It is based on features of the event. Returning to the coin-flipping case, Mellor writes, ‘[O]n the propensity view … a coin toss a's chance … of landing heads is a disposition whose basis F is a property or combination of properties which a actually has.’Footnote ¹⁶ So the coin toss has the disposition it has in light of certain properties of the coin toss. And this disposition fixes the physical probability (i.e. chance) of the relevant event taking place. Likewise, my sister's vote has the disposition that fixes the probability of the vote being cast for the Democratic candidate in light of certain properties about her vote. In other words, my sister's vote v's chance of being cast for the Democratic candidate under the condition that my sister votes in the next US presidential election is a disposition that is based on properties that v actually has.

By contrast, Paul Humphreys suggests that these dispositions (or, as he calls them, propensities) are not necessarily derived from features of the event. In fact, he states that ‘propensities are basic, often primitive, properties of the world’. So according to Humphreys, propensities or dispositions can be brute facts. It is not important for our purposes to take a side on this debate. In the subsequent discussion of the grounds for would-probably conditionals, we will consider grounds of both sorts: grounds that take propensities to be brute facts about the world and grounds that take propensities to be derived from other features of the world.

Given the unrepeatable nature of the events captured by would-probably conditionals (and other problems with the frequency interpretation), the Openist is committed to a propensity interpretation of physical probability. So the probability of my sister's vote being cast for the Democratic candidate under the conditions in which she votes is not based on the frequency of her vote going for the Democratic candidate under those conditions. Instead, the probability of my sister's vote being cast for the Democratic candidate under the conditions in which she votes is based on my sister having a certain propensity to vote for the Democratic candidate.

Understanding the would-probability conditionals as probability claims about the physical probability of conditional events where this probability is understood in terms of the propensity of the relevant object to act as specified by the probability claim gives the Openist his best shot at making sense of these conditionals. Moreover, Hasker seems to suggest a similar sort of understanding of would-probably conditionals in an exchange with Michael Robinson.Footnote ¹⁷ Hasker notes that ‘it is certainly true that, in general, our knowledge of probabilities is based on our observations of previous behavior of the individuals or systems for which probability is being assessed’.Footnote ¹⁸ However, he suggests an additional way to understand would-probably conditionals. He writes:

This suggests that claims about the probability of an object doing something under a given circumstance are claims about the objective propensity of that object to do that thing under that given circumstance (as opposed to claims about the frequency with which the object has done that thing under those circumstances in the past).

Given that the propensity understanding of physical probabilities is both the Openist's best shot at understanding would-probably conditionals and confirmed by one of Openism's leading proponents, I will focus on this understanding of would-probably conditionals in what follows. In order to keep this understanding before our minds, I will henceforth refer to would-probably conditionals as propensity claims.

Grounding propensity claims

Given the importance of propensity claims to the Openist's account of divine providence, it seems appropriate to ask the Open Theist the question so often asked of the Molinist with respect to counterfactuals of creaturely freedom: what grounds a propensity claim? Or to ask the question as Hasker does in his grounding objection: what makes the propensity claim true? Or, more generally, what is the truthmaker for a propensity claim?

The obvious answer to this grounding question is the relevant object having the relevant propensity. In the voting case, what grounds the propensity claim is the objectively existing propensity of my sister to cast her vote for the Democratic candidate under the circumstances of my sister voting in the next US presidential election. However, both Mellor and Hasker suggest that these propensities are not brute facts about the relevant object.Footnote ²⁰ They are ‘fixed’ by features of the object in question. For example, the propensity of a coin to land on heads if tossed might be grounded in the physical features of the coin. So, in the case of a coin, it is the physical features of the coin which fix the propensity of the coin to land on heads if tossed and thereby ground the corresponding propensity claim.

Open Theism is concerned with propensity claims not about coins but about free agents. When the propensity claim is a claim about what a free agent is likely to do, it seems that the psychological features of the agent will be particularly relevant to fixing the propensity of the agent to perform a given action. In a propensity claim about the likelihood of a free agent performing an action in specified circumstances, the psychological features of the agent appear in a fully loaded version of the specified circumstances. In the vote example, the circumstances are my sister's voting in the next US presidential election. Included in a fully loaded version of these circumstances are the relevant facts about my sister's character. It is these sorts of facts that seem like the best candidates for grounding the propensity claim about the probability of my sister voting for the Democratic candidate if she votes in the next US presidential election. Hence, I will begin by considering whether or not my sister's character is a good candidate to ground the propensity of my sister to vote for the Democratic candidate if she votes.

For our purposes, let's loosely understand a person's character as the sum total of her beliefs, habits, reasons, desires, etc. On this view, the propensity claim P (my sister votes for the Democratic candidate/my sister votes in the next US presidential election) > .5 is grounded by certain facts about my sister's character. What might these facts about my sister's character be? Let's suppose that my sister has always voted for Democrats. Moreover, let's suppose that she has worked on the presidential campaigns of several Democratic candidates in the past, continues to believe in the principles of the Democratic party, and desires for a Democrat to be president. It seems that my sister's character strongly indicates that the propensity of her vote to be cast for the Democratic candidate is greater than .5. So the Openist might appeal to the relevant agent's character to serve as the grounds for propensity claims involving agents.

But how exactly is it that an agent's character provides the ground for propensity claims involving that agent? Returning to our example, how is it that my sister's voting habits, political beliefs, etc. ground the claim that P (my sister votes for the Democratic candidate/my sister votes in the next US presidential election) > .5? An agent's character cannot ground the propensity claim about the agent by appealing to what the agent with that character actually does in the situation described by the propensity claim. This would not allow for an agent's character to ground a particular propensity claim and then the agent to freely act in a way other than what the propensity claim indicates. For example, let's suppose that my sister's character seems to indicate that P (my sister votes for the Democratic candidate/my sister votes in the next US presidential election) > .5. However, when she finds herself voting in the next US presidential election, she acts against her character and freely votes for the Republican candidate for president. If one appeals to what the agent with that character actually does in the circumstances in order to explain how character grounds the propensity claim, then it appears that my sister's character is good grounds for the claim that P (my sister votes for a Republican/my sister votes in the next US presidential election) > .5. But this is absurd! If character is to ground propensity claims, it must do so regardless of whether or not the agent actually acts in accord with her character. So it obviously isn't the case that character grounds a propensity claim in virtue of what someone with that character actually does in the situation described by the propensity claim.

Another explanation for how character can serve as the ground for propensity claims stems from Robinson's discussion of God's knowledge of probabilities.Footnote ²¹ He writes: ‘Probabilities are rooted in past regularities, which implies … that God's knowledge of the probability of future events is grounded in knowledge of past uniformities in the created order.’Footnote ²² As I noted above, Hasker denies Robinson's assertion that propensity claims are so grounded. However, given that our knowledge of propensity claims is often based on a person's character plus the past sequence of events involving that person or others with similar character, I think it is important to consider this candidate.

Recall that we are not concerned with the epistemological question of how one (in this case, God) knows propensity claims; we are concerned with what features of a person might serve as grounds of the truth of propensity claims about that person. But, as noted before, one way to identify a good candidate is to consider those things upon which one bases one's knowledge of propensity claims. Applied to our example, the idea is something like this: in the past, my sister has had the political beliefs that she has, the voting pattern that she has, the desires that she has and she has voted for the candidate who is a Democrat. So her present character plus facts about what she has done with that same character in the past is a good candidate for the grounds for propensity claims about her. Or even more broadly, the idea might be this: in the past, other agents with my sister's political beliefs, voting patterns, etc. have voted for the candidate that is a Democrat. So my sister's present character plus facts about what she and others have done with the same (or similar) character in the past is a good candidate for the grounds for propensity claims about her. It seems to me that this does capture how we often reason about what an agent would probably do in a circumstance; we look to what others with her character have done or what the agent herself has done in the past and we assume that the future will resemble the past. Maybe a good explanation for why reasoning in this way so often leads to true beliefs is that character plus the past is the sort of thing which grounds propensity claims.

Above I noted that this candidate stems from Robinson's discussion of God's knowledge of probabilities. Robinson points out that there is a famous objection to the claim that God's knowledge of probabilities can be grounded in an agent's character plus what agents with that character have done in the past. The objection comes from David Hume. Hume writes:

While Hume admits that we all do in fact reason from past cases to future instances, he does not think that such reasoning can be justified. This is the problem of induction. As Robinson puts it, our inductive reasoning is ‘grounded in a fundamental and ultimately unprovable assumption – a kind of base faith’.Footnote ²⁴ We cannot justify the assumption that the future will be like the past and such an assumption is necessary for inductive reasoning. Reasoning that is ultimately based on an unprovable assumption will not suffice to ground divine knowledge of propensity claims. So if Hume is correct, then God cannot base his knowledge of propensity claims on ‘past regularities’. The fact that agents with character like my sister's character have voted for Democrats in the past will not be good grounds for God's knowledge that P (my sister votes for the Democratic candidate/my sister votes in the next US presidential election) > .5.

Let's suppose that Robinson's argument succeeds in showing that God cannot base his knowledge of propensity claims on an agent's character and what agents with that character have done in the past. Does this argument in any way impugn the claim that past plus character grounds the truth of propensity claims? It does bring to light the fact that past plus character will not be sufficient to ground the propensity claim. If, in fact, the future is quite different from the past with respect to the relevant laws, then the past plus character will not ground the propensity claim. It might be that what my sister with her character has done in the past and what other agents with similar character have done in the past indicates that P (my sister votes for the Democratic candidate/my sister votes in the next US presidential election) > .5. However, if the future is different from the past in the relevant ways, then past and character will not suffice to ground this propensity claim. At the very least, one will need to add to the past and character the truth that the future will be like the past in order to ground propensity claims. Positing the existence of such a truth is, of course, highly controversial for the reasons offered by Hume above.

So far we have considered two possible explanations for how character can serve as the ground for propensity claims: (1) by appealing to what the agent with the character actually does in the circumstances described by the propensity claim or (2) by appealing to the past activity of the agent (or others with similar character) in similar circumstances. Neither of these explanations seems very promising: the first does not allow for character to ground a propensity claim when the agent acts against her character and the second requires controversially positing the existence of the fact that the future will be like the past in addition to the agent's character and the past to ground the propensity claim. In light of these problems, I will suggest a third explanation for how character might ground a propensity claim. I propose that character grounds a propensity claim by appealing to what the agent would do in nearby worlds in which she finds herself in the situation described by the propensity claim including the fact that the agent has the character that she has in the actual world. An Openist can consistently endorse the following: there are a number of equally nearby worlds with the same history as the actual world in which, for example, my sister finds herself voting in the next US presidential election with the character that she has. If in a significant majority of the nearby worlds in which she votes in the next US presidential election with her actual character she does in fact vote for the Democratic candidate, then it seems that we have grounds for the propensity claim with respect to her voting for the Democratic candidate if she were to vote in the next US presidential election. But if in many of the nearby worlds in which she votes in the next US presidential election with the character that she has, she still votes for the Republican, then we do not have grounds for the propensity claim. We explain how character grounds propensity claims by appealing to what the actual person with the character she actually has would do in nearby worlds in which she finds herself in the situation specified in the propensity claim.

This explanation solves the problems that arose for the first two candidates. First, it allows for an agent's character to ground the propensity claims about her even when she acts against her character. Suppose my sister acts against her character and votes for a Republican. Her character could still ground the propensity claim P (my sister votes for the Democratic candidate/my sister votes in the next US presidential election) > .5 if, in a majority of the nearby worlds in which she finds herself voting in the next US presidential election with the character that she has, she does in fact vote for the Democratic candidate. Moreover, it does not depend on positing the controversial fact that the future will be like the past. We are not appealing to what the agent (or others with similar character) have done in the past; we are instead appealing to what the specified agent does in the specified circumstances in nearby worlds.

It seems highly intuitive to me that an agent's character is what grounds the propensity claims about that agent. Moreover, the actions of the agent in nearby worlds in which she has the same character that she has in the actual world nicely explain how it is that character grounds these propensity claims. However, it is important to note that this explanation of how character grounds propensity claims does not ground propensity claims in something that is entirely actual. Part of what grounds the propensity claims are non-actual states of affairs. As such, this candidate for grounding propensity claims mirrors the Molinist candidate for grounding CCFs that I mentioned above. The Molinist suggests that CCFs that are genuinely counter-factual (i.e. have false antecedents) are grounded by the subject's actions in the nearest world in which the antecedent of the CCF is true. Hasker does not accept such grounds for CCFs as they are non-categorical; they fail to ground what would be the case in what is in fact the case.Footnote ²⁵ Hasker claims that grounds must be actual, must be what is in fact the case. Hasker's grounding requirement rules out not only the aforementioned candidate for grounding CCFs but it also rules out grounding propensity claims in what goes on in nearby, non-actual worlds. While this candidate for the grounds for propensity claims seems plausible in that it would entail the relevant propensity claim, we have not identified a ground for propensity claims that satisfies Hasker's own grounding requirements.

I have considered three explanations for how character grounds propensity claims. The first two do not explain how character entails the relevant propensity claim (or at least do not explain how character does so without the controversial claim that the future will be like the past); the third does not explain how character entails the relevant propensity claim in a way that satisfies Hasker's grounding requirements. However, I have not considered the possibility that character grounds propensity claims even though we may not be able to explain how character grounds propensity claims. As I stated above, it seems highly intuitive to say that my sister's character (her desires, beliefs, habits, etc.) makes it likely that she will vote for the Democratic candidate if she votes in the next US presidential election even if I cannot say how it is that her character grounds this likelihood. There are psychological laws (however unknowable) that govern this connection between my sister's character and her propensity to vote for the Democratic candidate if she votes just as there are physical laws that govern the connection between the physical features of a coin and its likelihood of landing on heads if flipped. Given the complexity of human psychology, it should come as no surprise, the Openist might say, that we cannot fully grasp these psychological laws (or even grasp them to the extent that we can grasp physical laws) and thus cannot explain how it is that character grounds propensity claims. However, God could surely know these laws and thus base his knowledge of propensity claims on his knowledge of a person's character plus the relevant psychological laws.

What can be said about these elusive psychological laws? Importantly, they need to be distinguished from physical laws if they are to govern the relationship between an agent's character and her free actions. It seems to me that, strictly speaking, the physical laws plus the complete list of physical facts about the coin (e.g. the angle that it is held in the hand, the speed at which it is flipped, etc.) entail that the propensity of the coin to land on heads is either 0 or 1. In other words, it seems to me that the physical laws plus the physical features of the coin determine whether or not it will land on heads; the coin cannot act against the indicated propensity. It only makes sense to talk about the propensity of the coin to land on heads as something other than 0 or 1 if we are talking about either epistemic probability or physical probability that only factors in a limited number of physical features of the coin. In the former case, even if the physical features of a coin plus the physical laws determine that the propensity of the coin to land on heads is either 0 or 1, we often do not know these physical features of the coin and thus the likelihood of the coin landing on heads given our knowledge of its physical features is .5. This is why we think that it is fair to determine who has the ball at the beginning of a football game by a coin toss; though the result of the coin toss is determined, we do not know the relevant physical features of the coin so we do not know which result is determined. In the latter case, we determine the physical propensity of the coin to land on heads excepting certain features, often the ones of which we have no knowledge. Presumably, the physical laws do entail a non-0, non-1 propensity of the coin landing on heads if you except certain physical features of the coin. But, again, the complete set of physical features of the coin plus the physical laws does entail that the propensity of the coin to land on heads is either 0 or 1.

By contrast, in the case of propensity claims that describe the likelihood of a free agent performing an action in a given circumstance, it will not be acceptable for this likelihood to be either 0 or 1. If the psychological laws plus the psychological features of an agent entail that the likelihood of an agent performing an action in a given circumstance is 0 or 1, then the agent is not free to act against her psychological features. She cannot do other than what her character plus the laws that govern character dictate that she will do and thus she is not free in the libertarian sense. This result is not acceptable to the Openist who proposes that there are psychological laws that govern free human actions. Such laws plus the features of the agent must always result in non-0, non-1 probabilities of the agent performing a specified action.

So what is wrong with saying that character grounds the propensity claims about an agent in virtue of psychological laws (about which we know very little) that govern the relationship between an agent's character and the probability of her subsequent actions so long as the probability entailed is always a non-0, non-1 probability? I guess there is nothing obviously wrong with this claim; the propensity claim is entailed by the agent's character and Hasker's grounding requirements are upheld. However, I cannot resist the worry that positing these mysterious psychological laws is unmotivated (besides the obvious motivation to ground propensity claims). It seems to me that our reasons for thinking that there are these psychological laws ultimately has to do with the fact that agents with a particular character have shown a significant tendency to perform certain actions in certain circumstances. However, if what agents have done in the past is not a good explanation for how character grounds propensity claims, then I don't see how it could be a good motivation for thinking that there are psychological laws that govern the relationship between character and propensity claims.

While it seems highly intuitive that the propensity claims about a person ultimately must be grounded in some way by the person's character, it is important to consider what other sorts of candidates are available to the Openist to serve as the grounds for propensity claims. Above I noted that Mellor and Hasker both seem to suggest that objectively existing propensities are based on certain features of the object in question. It seems to me that it is available to the Openists to say that these propensities are the grounds for the propensity claims. Moreover, instead of trying further to ground these propensities in the psychological features of the relevant agent involved, the Openist might just say that these are brute facts about the relevant agent that are not themselves in need of grounds (contra Mellor and Hasker). This is in keeping with Paul Humphreys's description of these propensities as ‘basic, often primitive, properties of the world’.Footnote ²⁶

If my sister has the property of being such that she has a propensity to vote for the Democratic candidate, if she were to vote in the next US presidential election, then surely the corresponding propensity claim is grounded. What makes it true that P (my sister votes for the Democratic candidate/my sister votes in the next US presidential election) > .5? My sister's propensity to vote for the Democratic candidate if she votes in the next US presidential election makes it true. And no further explanation is needed: the propensity fact is a brute fact about my sister.

Interestingly, positing brute propensity facts to ground propensity claims mirrors the Molinist's attempt to ground counterfactuals of creaturely freedom in brute counterfactual facts. Consider our above example: if my sister were to vote in the next US presidential election, she would vote for the Democratic candidate. Some Molinists have suggested that this counterfactual is grounded by the brute counterfactual fact that if my sister were to vote in the next US presidential election, she would vote for the Democratic candidate. It seems that if we allow for brute propensity facts, then there is no reason to rule out the possibility of brute counterfactual facts.