2. The Debate Over PRE

Double originally formulated PRE in his paper “Libertarianism and Rationality,” ^{Footnote 7} but I will rely predominantly on his more recent articulation and defence of the principle in “The Principle of Rational Explanation Defended.” ^{Footnote 8} PRE states: “Citing a person’s reasoning process R rationally explains choice C only if the probability of C given R is greater than the probability of not C given R.” ^{Footnote 9} Double thinks of PRE as a particularized case of a more general constraint on explanations. In his view, the explanans makes the occurrence of the explanandum more probable than its non-occurrence in all “insightful” explanations. ^{Footnote 10} So, if an indeterminate explanans makes A 0.6 likely and B 0.4 likely, the explanans can at best explain A; it cannot explain B. And if A and B are both 0.5 likely to occur, given the explanans, we can explain neither outcome. Hence, Double claims there is an intimate connection between our ability to explain a phenomenon by appealing to an event (or series of events) and the probability with which that event (or those events) will lead to the explanandum event. In the context of libertarian theories that appeal to indeterminism, PRE accounts for the idea that we require contrastive explanations for why the agent didn’t make the more probable choice. Double describes PRE and its connection to libertarianism as follows:

Kane disagrees with PRE on the grounds that agents have the capacity for dual (or plural) rationality, and that DR is at work when agents make certain kinds of free choices. ^{Footnote 12} In particular, DR is involved when agents make what Kane calls ‘self-forming actions’: undetermined choices that involve freely shaping the agent’s character. He defines DR as follows:

Although such situations are relatively rare, Kane claims that there are times when agents have divided wills when faced with a choice between two or more incompatible actions. On such occasions, the agent has reasons for and against both of the available options—reasons the agent endorses and thinks of as her own. This leads to a conflict of will the agent must overcome, but the outcome of this conflict is not determined by the agent’s prior character or motives. Instead, the agent exerts an effort of will to resolve the conflict (say, between moral and self-interested reasons). According to Kane, such efforts are indeterminate, leading to an undetermined choice. The fact such efforts are indeterminate, however, in no way undermines the explanatory worth of the agent’s reasons. In Kane’s view, whatever the outcome, the agent will have acted for the set of reasons that rationally supported that choice. The relative probabilities of each outcome are, for Kane, irrelevant to the explanatory connection between the agent’s reasons and her choice or action. What matters is that the agent acts for one set of the identified reasons and—given the agent’s character and beliefs—her choice is a rational one to make.

Double finds Kane’s approach implausible. In particular, he takes issue with the idea that we can explain the agent’s choice if it is the outcome of an indeterministic process or event.

According to Double, although the agent’s conflicting reasons or divided character might very well explain the motivational conflict and the agent’s effort of will to resolve it, “we will never reach the point where we see why either outcome results.” ^{Footnote 15} In his view, this severs the explanatory connection Kane claims exists between the agent’s reasons and her choice. Double is not alone here. A number of recent critics of Kane’s position have voiced similar concerns about the explanatory role of an agent’s reasons if they do not determine her choices. ^{Footnote 16}

In the third part of his discussion Double steps back and contemplates the source of the disagreement between him and Kane on this issue. He proposes that they are likely working with different meanings of ‘explain’ and, hence, different conceptions of understanding. Double suggests that Kane is working with an excessively weak conception of what it is to explain a choice or action, viz.: that to explain an action is to provide a narrative in terms of the agent’s reasons or psychological states that rationalizes the choice in the sense of rendering it understandable—that is, as something a rational person who possessed those reasons can reasonably do. Double thinks this is too weak because, as Richard Nisbett and Lee Ross ^{Footnote 17} have argued, we are prone to confabulate reasons in order to generate plausible sounding narratives that maintain our appearance as rational beings. Given this, an approach to action explanation that places so much emphasis on the rationalization of the agent’s choice is not sufficiently objective. Double claims we should also be wary of the idea that we can adopt Kane’s account alongside his own or make what it is to explain something a pragmatic matter. After all, our concern with the explanation of choices is driven by the deeper question of responsibility: can a dual-rational agent be sufficiently in control of her choices to be responsible for their outcomes? According to Double, a relativistic or pragmatic account of explanation is surely too divorced from the underlying metaphysics for the rationalizing relation to provide a meaningful answer to this question. Double, then, takes a meta-sceptical attitude toward this part of the debate, urging that a bottom-up approach will get us bogged down in questions about the nature of explanation and leave the real issues of responsibility untouched.

3. The Metaphysics and Epistemology of Explanation

In order to settle the above disagreement and to address Double’s scepticism, I want to introduce a framework for the discussion of explanations provided by Kim. ^{Footnote 18} This framework will highlight certain relevant features of both Double’s and Kane’s positions with the aim of showing how they can, in a sense, be reconciled. Kim embraces what he calls ‘ER’. For Kim, explanations of individual events take the form of propositions, but they are grounded in and track metaphysical dependency relations between events in the world.

Kim leaves room for the possibility that there are many different kinds of explanatory relations—as many as there are forms of metaphysical dependence—but he treats the causal relation as the most central case. This, of course, assumes what Kim calls ‘causal realism’ since causal relations would have to be objective, mind-independent relations in order to support ER.

Kim contrasts ER with EI. EI characterizes explanations in a way that makes explanation an ‘internal’ matter of how propositions about events relate logically to one another or to our overall body of knowledge. Kim describes EI as follows:

Kim claims that views that analyse the causal relation itself in terms of explanation, ^{Footnote 21} or that treat explanations as arguments, ^{Footnote 22} or that make explanation entirely a matter of what Wesley Salmon ^{Footnote 23} describes as the ‘psychological conception,’ are all irrealist because they do not ground the explanans relation between propositions in objective relations between the events the propositions are about.

It is reasonably clear that Double is committed to ER, given his insistence that genuine explanations should be “objective.” ^{Footnote 24} While he does not employ the same framework or vocabulary as Kim, he surely thinks that a genuine or insightful explanation should be grounded in an objective relation between events. After all, he is concerned not to lose sight of the issue of responsibility in the debate about libertarianism, and this calls for an appropriate dependency relation between the agent’s deliberation, her choice, and subsequent action. Double also seems to think that if, following Kane, we make explanation wholly a matter of rational coherence between the agent’s reasons or motives and choices, we lapse into EI. Double’s worries about our propensity for confabulation in the weaving of plausible sounding narratives about our reasons and actions indicates a commitment to the idea that a good explanation of a choice must be more than a matter of logical, epistemic, or rational relations between the descriptions of the explanans and explanandum events.

It is difficult to fault Double for the above commitments. I am quite sympathetic with ER and I tend to agree that, if Kanean explanations of free (dual-rational) choices were just a matter of rational coherence between the explanans and explanandum, this would be unacceptably irrealist and, as such, would be insufficient to ground moral responsibility. However, things are not so simple. The metaphysical relations highlighted by ER are but one dimension of explanation, and it is a mistake to think that Kane’s view is irrealist, or so I shall argue.

While our explanations are, as Kim claims, grounded in objective truths about the world, they also have an essentially epistemic dimension. Kim writes:

Unfortunately, Kim has very little to say about this epistemic dimension of explanation because his purpose is to formulate and to defend his principle of explanatory exclusion, which is driven primarily by considerations drawn from the metaphysical dimension of explanation—ER and causal realism, in particular. Nevertheless, he does acknowledge that explanations might involve more than the representation of metaphysical relations:

I don’t have a fully developed account of the epistemic dimension of explanation to offer either, but I will hazard a few remarks that are (I hope) relatively uncontroversial and that will prove useful for the evaluation of PRE.

Whatever else one might want to say about explanation, it is clear that it is an intensional relation. ER demands that the explanans and explanandum pick out events that stand in an objective dependency relation, but considerations about the substitution of logically inequivalent descriptions in explanation statements show that not just any descriptions of these events will be explanatory. Consider the following twist on Donald Davidson’s ^{Footnote 27} classic example offered by Ausonio Marras:

If such examples are compelling, they suggest two related lessons. The first is that explanations have an irreducibly epistemic dimension that involves the generation of understanding. If they did not, we would find causal tautologies and causal claims that employ oblique descriptions of events explanatory, given that they satisfy ER. The second lesson is that explanations are not individuated solely by the objective relations they represent, but also according to the way these events and relations are described or the properties they token.

These simple observations suggest an interesting possibility: depending on how we describe two events that stand in an objective dependency relation, there may be more than one way of producing understanding and, hence, more than one type of explanation available for the same event. ^{Footnote 30}

One way of generating understanding is already evident in PRE. Double makes this even more explicit by offering the following:

Although Double is agnostic about the correct logical form of explanations, his account of how explanations generate understanding appears very Hempelian. His emphasis on the role of laws is reminiscent of Carl Hempel’s claim ^{Footnote 32} that to explain an event is to see that its occurrence was something to be expected, given the explanans and laws of nature (whether deterministic or otherwise). In the light of this, it is clear why Double finds PRE so obvious. If we achieve understanding by coming to see how an event was to be expected, given certain laws and given the preceding events, it follows that we cannot understand the occurrence of an improbable event (i.e., one with a probability ≤ 0.5), since such events are by definition unexpected.

I have no quarrel with this view of understanding. I believe that it is a plausible and useful account of the epistemic dimension of explanation. However, in the light of the intensionality of explanation, I must disagree with Double’s assumption that this is the only acceptable account. Since the events that stand in an objective relation token a number of different properties, it is plausible to claim that alternative explanations of the same event are possible—ones that appeal to different explanatory relations between different pairs of properties of the cause and the effect. ^{Footnote 33} Double focuses on nomological properties of the related events, whereas Kane appeals to a rational relation between mental (or content) properties of the cause and the effect.

According to the alternative approach, then, when we explain an intentional action by appealing to the agent’s reasons or other motivational states, we do not achieve understanding by coming to see that the action was to be expected, given the agent’s reasons and certain laws of nature. Instead, in a manner similar to the approach to action explanation proposed by William Dray ^{Footnote 34} and more recently by Kim, ^{Footnote 35} we achieve understanding by seeing how the action is the rational thing to do, given the agent’s beliefs and desires. This is surely Kane’s view since he makes so much of the rational fit between the agent’s reasons and her choices—whichever way she chooses.

The best way to illustrate this account is to consider our way of making sense of situations in which an agent’s behaviour initially appears irrational. ^{Footnote 36} Imagine that you see your friend Bob in the supermarket and that he is in the process of wrapping his head in aluminum foil. This is not normal behaviour when shopping. Indeed, such behaviour should strike you as extremely odd. Now suppose that you ask Bob what he is doing and he tells you the following story: he thinks the CIA has implanted a microchip in his head that can transmit his thoughts to CIA headquarters. Since he would prefer to keep his thoughts to himself, and since he believes that wrapping his head in foil will block the transmission of his thoughts by interfering with the radio waves used by the transmitter, he has rushed to the supermarket to procure some foil and has quickly set to work wrapping his head.

I think we can all agree that this is not a particularly good reason to wrap one’s head in foil since it is justified by beliefs that are almost certainly false. But this does not mean that Bob’s reason fails to explain his action. If you believe that the CIA is monitoring your thoughts, if you don’t want the CIA to do this, and you believe that by wrapping your head in aluminum foil you can prevent it, as a rational person you should wrap your head in foil. ^{Footnote 37} It seems clear that it is in virtue of this rationalizing role of Bob’s reason that we find it explanatory. For it is not until we can grasp a rational connection between his action and his other mental states that we have a satisfactory account of what he is doing. To drive this point home, we need only to consider what would happen if we could not situate Bob’s action in the context of his mental states in a way that rationalizes it. If every attempt to uncover a rational connection between his action and his beliefs and desires failed, Bob’s behaviour would forever be mysterious. For example, if all we learned about Bob’s mental states was that he thinks there is a conspiracy to prevent the mass production and distribution of the electric car, there would be no way to make any sense of his action. Accordingly, we would cease to regard him as a competent agent.

These considerations are hardly conclusive but, in the light of examples such as these, and of our explanatory practices generally, there is a good case for thinking that the epistemic dimension of action explanations is a matter of seeing how the agent’s choice or action is rational, given the agent’s other mental states or deliberation. Hence, the alternative to Double’s account of the epistemic dimension of explanation cannot be dismissed out of hand, and so there is little reason to agree with his assumption that the only way in which an explanation can produce understanding is by showing that the explanandum event was to be expected with a degree of probability above 0.5.

Double would no doubt object that the above account in terms of the rationalization of the explanandum is excessively irrealist since it seems to make explanation entirely a matter of internal or logical relations between propositions about the agent’s deliberation and choice. After all, the explanatory relation (i.e., the rationalizing relation) appears not to be an objective relation between events, but a logical relation between one way of describing them. But appearances are misleading and the concern is misguided. There is no reason to think that the above account of understanding is incompatible with ER. In fact, such a commitment is already built into Kane’s account of DR (quoted earlier) in claim (c): “the agent does it [i.e., performs the action in question] for those reasons.” That is to say, there is an objective relation (a causal relation) between the agent’s reason and her choice or action. So long as the reason that explains the event in question caused it (either deterministically or via an indeterministic effort of will), there is no danger of EI. While our capacity for the confabulation of reasons is no doubt extensive, this fact does not undermine the claim that we achieve understanding by coming to see how an agent’s choice is rational in the light of her reasons, for we can distinguish as Davidson did ^{Footnote 38} between those reasons that did, and those that did not, cause her action, and this satisfies the metaphysical constraints on explanation that ER requires. The worry that Kane’s account of explanation and understanding amounts to a form of EI, then, is misguided. Nothing about the fact that an action explanation produces understanding via the rationalization of the choice or action implies that this internal relation is all there is to such explanations, or that they are not grounded in underlying metaphysical relations. Thus, Double is too quick to dismiss Kane’s account of explanation. Also, Double is premature to conclude that Kane’s denial of the possibility for contrastive explanations in the context of undetermined self-forming actions severs the explanatory connection between the agent’s reasons and her action.

4. Neural Indeterminism and the Canonical Form of Explanation Statements

While Kane’s emphasis on the rationalizing role of an agent’s reasons may be consistent with ER, Double and many of Kane’s other critics continue to object that the indeterminism involved in dual rational choices represents a fundamental obstacle to any satisfying explanation of such choices. In an illuminating passage, Double expresses the concern in a way that still speaks for many of Kane’s critics:

In this section, I argue that this kind of objection (and, hence, the concomitant demand for contrastive explanations) runs afoul of the intensionality of explanation.

We saw in Section 3 that the canonical form of a singular explanation statement (‘c’s being F explains e’s being G’) connects a property of the cause to a property of the effect. But, given that the cause and the effect both token multiple properties (under the assumption of nonreductive physicalism), how do we decide which property of the cause is explanatorily relevant to which property of the effect? Marras pairs the properties mentioned in the explanans and explanandum according to the causal relevance of the former to the latter, as revealed by a counterfactual test. ^{Footnote 40} That is, c’s being F explains e’s being G only if it is the case that, had c not tokened F, e would not have tokened G. This approach has some problems in the light of apparent counterexamples ^{Footnote 41} and so my preference is to pair the property of the cause with the property of the effect according to their epistemic features. ^{Footnote 42}

Suppose, then, that, in accordance with ER we have two events, c and e that stand in a causal relation. Let us also assume, in accordance with the tenets of nonreductive physicalism, that c and e each token two non-equivalent mental and physical properties. ^{Footnote 43} Hence, c tokens a mental property M₁ and a physical property P₁, while e tokens a mental property M₂ and a physical property P₂. What I have been proposing, in effect, is that the way we pair our descriptions of these events when we construct an explanation is determined by their epistemic features—to the ways in which the relations between the descriptions can produce understanding. The mental descriptions M₁ and M₂ stand in a rationalizing relation, such that the former rationalizes the latter. P₁ and P₂, we may assume, stand in a nomological relation, whereby the former would lead us to expect the latter with either certainty or a high degree of probability, depending on the kind of law involved. Relative to the appropriate descriptions, then, we can have a rationalizing explanation alongside an explanation in terms of nomic implication, both of which involve the same pair of events and both of which are grounded in the same underlying metaphysical (i.e., causal) relation.

In most contexts, it would be fruitless to combine the above descriptions and hope that we would have epistemically satisfying explanations. While the claim that ‘c’s being M₁ explains e’s being P₂’ satisfies the metaphysical dimension of explanation required by ER, since c causes e, it is difficult to see how this statement could be explanatory. We do not normally think of events such as desires as rationalizing a complex set of signals sent through the nervous system, along with their effects on the contraction and expansion of certain groups of muscles. Although such complex events are actions, when so described, they are explained by an appeal to the neurobiological properties of the cause, not by mental content. ^{Footnote 44} It seems to me that Double’s criticism, quoted above, makes precisely this mistake, though the descriptions are reversed: he describes the cause in physical terms (as a quantum neural event) and the effect as a mental event (a choice). If one works with these two descriptions, then, of course, explanation will fail, but this is not because no explanation is possible or because no objective relation exists between the identified events. The problem is with the way Double describes them. I take this to be what Kane is gesturing at when he says the following:

So, if we assume that the relevant events have physical descriptions according to which no laws, save perhaps probabilistic ones, apply, it will appear as though what happens is random and inexplicable. But, as Kane suggests, this is not the only available description of these events. Under their mental descriptions, the related events stand in a rationalizing relation, and this satisfies the epistemic dimension of explanation in the way described earlier. Those, like Double, who object to indeterminist theories like Kane’s by saying that at some stage in the deliberative process an indeterministic process ‘takes over’ and produces the outcome or choice when an agent’s will is divided are therefore doing something illegitimate: they are running roughshod over the fact that explanation is an intensional relation. They are ignoring the epistemic dimension of explanation and its dependence on the way we describe events or the properties they token.

One might object to this proposal that Kane is guilty of this himself when he interposes the effort of will between the agent’s reasons and her choice and characterizes the former in terms of ‘neural indeterminism.’ There is a sense in which this is true, but I think it is harmless. After all, Kane does not treat such efforts as a feature of our everyday action explanations since he claims that agents are typically unaware that they are making dual efforts. Instead, we should think of Kane’s appeal to neural indeterminism as an effort to uncover the metaphysical relations that are at work in such cases.

Recognizing these points goes a long way towards undermining PRE, or at least to limiting the scope of its relevance. Once we give the epistemic dimension of explanation its due, we can appreciate how there can be more than one way in which explanations generate understanding. To be sure, one is the way Double champions in PRE: some explanations generate understanding by helping us to see that the event in question was nomologically implied by the event(s) that preceded it. I have no doubt that many explanations of events operate in something like this way, and that our ability to explain such events is linked to the probability of their occurrence. But the application of this approach to the explanation of actions in terms of reasons is a mistake. Here, understanding is produced in a different way and considerations about probability appear largely irrelevant. Hence, PRE is not a general constraint on explanation as Double claims, but a constraint on a certain kind of explanation.

Employing Kim’s framework of ER, EI, and the metaphysics and epistemology of explanation has, I believe, proved to be extremely useful. It helps us not only to see the differences between Double and Kane more clearly, but also provides a way out of the impasse that Double claims exists between them. Both philosophers identify plausible accounts of understanding (and, hence, of explanation), but these are not incompatible in the way Double assumes. Indeed, they can operate side-by-side, provided that the events in question support the relevant descriptions or token the appropriate properties. Double worries that to relativize explanation to a description in this way will undermine our ability to say anything useful about responsibility, but we have also seen that this concern is misguided. So long as the Kanean account involved in DR satisfies the metaphysical commitments of ER, the connection to responsibility is guaranteed.

5. Another Problem on the Horizon

The foregoing argument should give non-Valerian libertarians like Kane hope, since it bolsters their existing attempts to create a bulwark against the unrelenting tide of objections to the effect that indeterminism renders choices random and inexplicable. There is, however, a further problem that libertarians who adopt this approach will have to face. My claims about the intensionality of explanation will no doubt sound rather Davidsonian. To the extent that they involve the adoption of some form of nonreductive physicalism (not necessarily Davidson’s), I suspect they are likely to provoke an objection with which nonreductive physicalists have struggled for some time now. The objection runs like this: If the mental properties that figure in rationalizing explanations are not the properties in virtue of which mental events cause actions, then they cannot be genuinely explanatory. Events have the power to cause what they do in virtue of their law-engaging physical properties; hence, an event’s mental properties are epiphenomenal.

Arguments abound that the nonreductive physicalist cannot secure the causal relevance of mental properties unless they embrace reduction and identify mental properties with physical properties. ^{Footnote 46} Kim succinctly describes this in terms of the ‘problem of causal exclusion’ as follows:

In the light of the nonreductive physicalist’s commitment to physicalism and to the causal closure of the physical domain, there is tremendous pressure to say that the neurophysiological property of the cause provides the real story, in which case the fact that the cause also has the mental property of being a desire for beer is irrelevant. Thus, the explanatory relevance of an event’s rationalizing properties might still be in jeopardy, in which case the worry that Kane’s approach leads to EI resurfaces. I cannot hope to resolve this issue on Kane’s behalf here, though I do think there is an available plausible response. Suffice it to say that, although the argument in the previous sections has created a clearing, Kane is not out of the woods yet.