Simulation and Reason Explanation: The Radical View

A Prefatory Note 

Alvin Goldman's early work in action theory and theory of knowledge was a major influence on my own thinking and writing about emotions. For that reason and others, it was a very happy moment in my professional life when I learned, in 1988, that in his presidential address to the Society for Philosophy and Psychology Goldman endorsed and defended the “simulation” theory I had put forward in a 1986 article. I discovered afterward that we share a strong conviction that empirical evidence is relevant to a full assessment of the theory. We both find the burgeoning evidence from cognitive neuroscience to be of particular interest, I believe, in part because it makes possible a major departure for the philosophy of mind: turning its attention from

(a)                  the neural basis of mental states,

to

(b)                 the neural basis of the epistemology of mental states.

Instead of looking to (a) with the hope of incorporating the mental into a naturalistic ontology, the newer focus on (b) aims at a naturalistic epistemology of the mental. It concerns the neural mechanisms that subserve our capacity to make mental attributions and to predict and explain (as well as to anticipate and understand nonverbally) the actions of others, with at least satisficing reliability. An excellent example of (b) is Goldman’s work with Gallese on the link between simulation and “mirror neurons,” high-level motor neurons that are activated under two distinct conditions: when the individual executes specific purposive actions, such as grasping an object, and when the individual visually perceives similar actions in others. [1]

Eventually, differences emerged between Goldman’s understanding of the simulation theory and my own. We differ on the logical structure of simulation, the self-ascription of mental states, and the implications for mental concepts. Concerning the brain, I am interested in (b), but I am also interested in

(c)                  the neural basis of possession of the concepts of mind and various mental states.

A word about (c) might help put the present paper in context. Quine held that it is only when we switch to "the essentially dramatic idiom" that we can speak of intentional states. However, the existence of mirror neurons, along with well-documented phenomena such as the automatic mimicry of facial expressions and the tracking of another’s direction of gaze, make it clear that this “switching muses” goes far beyond the elective methodology of putting ourselves in the other's shoes; it occurs also, in a rudimentary form, as an unavoidable feature of our perception of conspecifics. My view is that human beings equipped with such imitative mechanisms of social understanding are predisposed to populate the world with entities that have strange, not obviously naturalistic properties: entities with “points of view” or “perspectives” that have the potential of being simulated, with varying degrees of success.  Without the predisposing mechanisms, we would find attributions of propositional attitudes unintelligible.

If it could be established, the dependence of our concepts on such special-purpose equipment would not explain away minds and beliefs and desires. An ontology to which we are disposed because of mechanisms that evolved for everyday survival in a social world is not like an ontology of witches and phlogiston. It isn't a myth, illusion, or false theory that distinguishes our ontology of simulable points of view from what people with autism are said to perceive when they see human beings: mere objects, bags of flesh animated by an internal machinery of heart, lungs, and other organs, including, perhaps, an information-processing control center that makes the parts move about in unfathomable ways.

My version of the simulation theory in general has been tagged the “radical” view, Goldman’s being the more conservative, and here and in an earlier paper I have adopted that tag as my own. Discussions of our differences have been rewarding to me, so much so that I almost hope the present paper generates even more controversy. Here I argue that it is simulation that makes it possible to think of people as acting from or because of reasons. More narrowly, what is necessary is a simulative procedure for determining what a person would have done under counterfactual conditions. This gives us a bifurcated understanding of counterfactual possibility. “What would x have done had it been the case that p?” is interpreted one way for (x=an inanimate object) and another way for (x=a person), where the difference inheres in the way we suppose or pretend that p—Ramseyan pretense for x=an inanimate object, and simulation for x=a person.

***

Reason Explanations

I took the longer route to my office today, because it’s the more scenic one, and I had plenty of time. On the way, I stepped on the brake pedal on various occasions: once, because another car had swerved in front of me, another time because a pedestrian was crossing the street, on several occasions because the traffic light was red, and so on. Finally, I drove my car onto a ramp opening into a large edifice. Why there? Because that’s the entrance to my parking garage.

Each of these explanations cites as explanans a reason of mine for doing what I did. It explains action in terms of the reason or reasons for which the agent acted, where “a reason” is understood as a reason for, a reason in favor. This is a reason in the strict sense: a favorable consideration, something about the world—a fact—that, at least to the agent's eyes at the relevant time, favored, or argued in favor of, doing what he did. (Many philosophers have supposed reasons-in-favor to be enthymematic; usually, they are thought to be the non-normative, or “minor,” premises of practical syllogisms.) An explanation in terms of a reason of A’s for which A X’d I will call a reason explanation of A’s X’ing. [2] The typical form of a reason explanation is, “A X’d because p,” where it is left inexplicit that the explanans, that p, was a reason of A’s for X’ing, as opposed to a “brute” cause. The explanations presented in the previous paragraph are all reason explanations.

Since the early pioneering work of Donald Davidson, philosophers have noted that our everyday explanations of action perform two distinct functions. On the one hand, reasons-in-favor, or the practical arguments of which they are supposedly premises, may argue for, and in a weak and relative sense justify, A’s X’ing (that is, agent A’s performing an action of the type “X’ing”). On the other hand, they may also explain why A X’d. To show an act to be justified from the agent’s point of view, and even to cite one or more reasons-in-favor that justify the action from the agent's point of view, is not sufficient for explaining why the agent performed the action. Where two or more such arguments favored X’ing, at least in the eyes of the agent, the explanation specifies which one or more of these actually explains why A X’d. For example, on one occasion when I braked for a red light, I had another reason to step on the brake pedal: I had been going ten miles over the speed limit. However, there is an asymmetry between these reasons: I wouldn’t have stopped, other conditions being what they were, if it hadn’t been for the red light; but I would have stopped even if I hadn’t been speeding. The red light would have sufficed without the speeding, but the speeding would not have sufficed without the red light. I braked because there was a red light. [3]

In what I am calling a reason explanation of action, the explanans, what is proffered as explaining the action, is identical with a justificans or reason-in-favor, a fact that justifies the action (at least from the agent’s point of view). In my example, there were two reasons for stepping on the brake, two considerations that independently counted in favor:

I was going ten miles over the limit, and

there was a red light ahead.

Why, then, did I step on the brake? Because

there was a red light ahead.

Here, the fact that there was a red light ahead is both a justificans and an explanans.

The Breakup of Explanans and Justificans

In several decades of philosophical debate about explanation in terms of reasons, there has rarely been any discussion of reason explanations. Most who say they are discussing explanations in terms of the agent's “reasons” identify a reason, not with a justificans or reason-in-favor, but with a property of the agent, with, more specifically, the agent's being in a particular “state”: typically, that of having a particular belief or belief-desire pair. The usual course has been, first, to assume that explanations in terms of reasons are “really” explanations in terms of beliefs, or, beliefs and desires; then, to argue that these belief-desire explanations, in turn, are really causal explanations. Philosophers tend to speak of pro-attitudes rather than, as one might expect, of pro-considerations, or reasons-in-favor. Wedded to the idea of a “belief-desire psychology,” philosophers, when they do discuss reasons-in-favor, do so, not in the context of explanation, but strictly in the context of justification. On the one hand, it was my psychological state (believing that there was a red light ahead, etc.) that explains why I stepped on the brake pedal; on the other hand, my reason for doing so was that there was a red light ahead—the content of my belief. In effect, this is to cast aside what I call reason explanations, where it is not a state of the agent, but the justificans itself, that does the explaining. Most philosophers who discuss the topic do not even mention explanations in terms of facts or considerations that favor or argue for the action. They simply dub certain forms of belief-desire explanation “explanations in terms of reasons.” They leave to subsequent inquiry the troublesome question whether an action that is caused by beliefs can somehow be depicted as dependent on the content of the beliefs that cause it.

Why have reason explanations been ignored? Most likely, because they could not plausibly be regarded as a species of causal explanation. Davidson thought the difference between merely having a reason for X’ing and actually X’ing for, or because of, that reason was that in the latter case the reason and the action were causally related in a certain way. Given the assumption that reasons explain action causally, it is understandable that one would not identify the explanans with a justificans, a reason-in-favor. For in a variety of cases, reasons-in-favor are not plausible causal candidates; nor are they plausibly identical with or supervenient on something that might be a cause. Consider some of the explanations I gave in the opening paragraph. Although the light’s being red might be thought to qualify as a cause or causal condition, and perhaps also one’s going ten miles over the speed limit, it would be difficult to construe the quality of being the more scenic route as causing something to happen. [4] For it has a normative component that resists reduction to any properties, such as the preponderance of trees and the distance to the visible horizon, that might plausibly count as causes or causal conditions. Further, a reason may be a future-dated fact or condition (I am visiting today because I won’t be able to tomorrow), a necessary truth (The committee didn’t fund his project, because it is mathematically impossible to square the circle), or a moral truth (He quit the job because the company’s parental leave policy was unfair). But future events and conditions are not causes of past events, and necessary truths and moral truths are not causes of anything. It is preposterous to think of the mathematical impossibility of a task or the unfairness of a policy as causing  something to happen. The existence of such reasons clearly precludes at least a simple causal reading of the “because” in a reason explanation.

If the connective is to be understood as causal, then we are forced to separate explanans from justificans. It seems plausible to suppose that the belief that route A is more scenic than route B might be a causal factor; likewise, the belief that I won’t be able to visit tomorrow, the belief that it is mathematically impossible to square the circle, and so on. Thus it commonly held that a reason explanation is always, at least implicitly, a psychological explanation, an explanation in terms of a property of the agent, specifically, the agent's having certain beliefs or beliefs and desires. This is a point that has in recent years become dogmatically “obvious” to many philosophers. Their hope is that these explanations, at least, can be made out to be causal.

The point is even more pressing if one adopts, as Davidson and many others do, a nomological view of causation. To maintain his causal analysis without compromising his nomological conception of causation, [5] Davidson must claim that the real explanans in a reason explanation is not in general those reasons or considerations in favor of the action for which the agent acted, but rather something that can plausibly be held identical with an event covered by a law of nature. With a little luck—a providence that has, let’s admit, been long in coming—philosophers or empirical scientists might show beliefs and desires, or onsets of beliefs and desires, to be states or events that can plausibly be held identical with (or to be constituted by or at least to supervene on) states and events that are covered by laws of nature.

Where would this leave reason explanations? One might posit a causal chain, with belief a causal intermediary between fact and action.  On one side of the chain, the belief that A is the more scenic route is a cause, or a causal condition, of my choosing and taking route A. (The causal relation must be of the right sort, excluding the wrong, or “wayward,” sorts of causal chains—although there is no consensus among philosophers as to how to exclude all and only those chains.)

And on the other side of the chain? If we are to make causal sense of the ‘because’ in a reason explanation such as,

I took route A because it is the more scenic route,

then we must suppose that the belief, in turn, was caused by A’s being the more scenic route. Then, with belief causally linked on the anterior side to this external situation and on the posterior side to my action—and assuming the transitivity of the causal relation—one might boast that one has succeeded in showing the “because” of reason explanations to be causal.

Whatever one may think of the alleged causal relation between belief and action, there are obvious problems at the anterior side of the chain. The alleged link between beliefs and facts or reasons poses again the very problems mentioned earlier. Normative “facts” such as one route’s being more scenic than another and a company’s policy’s being unfair are not candidate causes, whether of actions or of beliefs. Nor is there a causal link from mathematical truth to mathematical belief. Something's being mathematically impossible isn't a cause of the belief that it is mathematically impossible. Arguably, there may be a relation of counterfactual dependence. In the circle-squaring example, it may be that, given the committee’s ability to “track the truth” in this particular subdomain of mathematics, given its familiarity, for example, with proofs of the transcendental character of pi, the committee wouldn’t have believed it mathematically impossible to square the circle, if it were not. However, we certainly cannot (i.e., should not) claim that this counterfactual reflects an underlying causal connection between a necessary truth and the corresponding belief.

So let us suppose that some day we will have a satisfactory causal-nomological account of beliefs and desires. Suppose even that beliefs and desires can be shown to be token-identical with undeniably physical states. These unlikely triumphs would bring us not a jot closer to a causal account of reason explanations, explanations in which a reason-in-favor explains as well as justifies.. They would not enable us to make strictly causal sense of the intuition that an agent may actually do something “because p,” where “p” specifies a reason-in-favor.

The Counterfactual Difference

It has not been seen as an easy task, however, to explain in a non-causal way the difference between a reason-in-favor that explains an action and one that does not. Jonathan Dancy expresses the frustration felt by philosophers who have tried and failed to find such an explanation:

The most direct response to Davidson … is just to say that the difference between those reasons for which the agent did in fact act and those for which he might have acted but did not is not a difference in causal role at all. It is just the difference between the considerations in the light of which he acted and other considerations which he took to favour acting as he did but which were not in fact the ones in the light of which he decided to do it. This is admittedly not very informative, since we have to allow that we have offered no analysis or philosophical account of the ‘in the light of’ relation. I suspect, however, that no such analysis or account is available to be given, without therefore supposing that this has any tendency to show that the relation concerned does not exist. It is what it is, and not another thing; and if it cannot be analysed, so much the worse for the more global pretensions of analysis. (I agree, however, that it would be good to produce an account of the ‘in the light of’ relation—if one could only think of some way of producing one.) [6]

I agree with Dancy that the distinction between a reason-in-favor that explains an action and one that does not is not based on a causal difference. However, the idiomatic “reason for which” and “reason in the light of which” might wrongly be taken to suggest that reason explanations are purely subjective: that if agent A means or intends to be X’ing “for” a particular reason, namely, that p, then the reason that explains A’s X’ing is that p. The phrase “in the light of” seems even to suggest that only introspection can decide which reason explains A’s X’ing.

There is, in fact, a clear counterfactual difference between a reason-in-favor that explains an action and one that does not. In my earlier example, there was the following difference between my two concurrent reasons for braking, the fact that the light was red and the fact that I was going ten miles over the limit:

It is the case that

1a.        I would not have braked if there hadn’t been a red light—other things being “equal,” including my going ten miles over the limit.

However, it is not the case that

1b.       I would not have braked if I hadn’t been going ten miles over the limit —other things being “equal,” including the fact that there was a red light.

The red light would have sufficed without the speeding, but the speeding would not have sufficed without the red light. The presence of the red light was sufficient (in its context) to move me to step on the brake; my going ten miles over the speed limit was not. The same holds for cases in which the explanans is a necessary or a normative truth. Suppose the grant committee thinks that two considerations count against accepting (and, given a forced choice, for rejecting) Sam’s project: one, that it is mathematically impossible; the other, that Sam has a mediocre “track record.” Suppose, however, that there was the following counterfactual difference between the two considerations: [7]

1a.        The committee would not have rejected Sam’s project (i.e., would have approved it) if it hadn’t been a mathematically impossible project—other things being “equal,” including the mediocrity of Sam’s record.

1b.       The committee would not have rejected Sam’s project if Sam’s record had not been mediocre—other things being “equal,” including the mathematical impossibility of carrying out the project.

The impossibility of the project would have sufficed without the track record, but the track record would not have sufficed without the impossibility. Although Sam’s mediocre track record was not sufficient to move the committee to reject his project, the mathematical impossibility of carrying it out was sufficient. (Again the sufficiency in question is contextual. That is, there are likely to be additional, unmentioned, conditions under which the committee would not or at least might not have funded the project, mathematical impossibility aside: a poorly written proposal, negative expert evaluations, or a negative sign from God.) This counterfactual difference appears to be all that is needed to distinguish a reason-in-favor that explains the action from a reason-in-favor that does not. [8]

But how are these counterfactuals to be understood, if not causally?

An Example

Here is the setting: I pre-heat the oven, put in the frozen pizza, and retire to the living room. An hour or so later, smoke fills the kitchen and starts lapping into the living room. I run into the kitchen and turn off the oven. Just as I do, the smoke alarm goes off (sounds the alarm). My smoke alarm and I both responded to the presence of smoke in the kitchen. Our respective behaviors appear to have the same explanation:

1. The smoke alarm went off because there was smoke in the kitchen.

2. I ran to the kitchen because there was smoke in the kitchen.

Superficially, at least, these explanations appear to make similar counterfactual commitments, namely, that if there hadn’t been smoke in the kitchen, then the smoke alarm would not have gone off, and I would not have run into the kitchen. There appear to be further parallels. The presence of smoke in the kitchen, considered in isolation, is not sufficient for setting off a smoke alarm. Conditions must allow the smoke to reach the sensor, and the alarm must be powered and in working condition. Thus, in the example, a number of further conditions obtained in the absence of which the alarm would not have sounded. These background conditions did not suffice in themselves to make the alarm go off, but the conjunction of these with the presence of smoke did suffice. In my own case, too, a number of further conditions obtained in the absence of which I would not have run into the kitchen. Some are intuitively "positive" conditions: for example, that if I am in the kitchen I will be in position to turn off the oven, that the sooner I turn off the oven the less the damage will be, that I run faster than I walk.  Others are intuitively "negative": that no one else is in the kitchen, that I cannot turn the oven off from the living room, that running to the kitchen doesn't require stepping through a rattlesnake pit. Given just this set of relevant background conditions--which I am not able to specify completely--I would not have run into the kitchen; but, given also that there was smoke in the kitchen, I would run.  That is, I would do so no matter what other conditions did or did not obtain: for example, whether or not the kitchen walls are green, whether or not China's capital is Beijing.

However, despite these parallels, reason explanations are different. For one thing, each of the conditions that contribute to a sufficient condition, both the explanans (there was smoke in the kitchen) and the background conditions (e.g., the sooner I turn off the oven the less the damage will be), is implicitly cognitive. Where the explanans, “there was smoke,” is a reason-in-favor, then for the explanation to be correct, I had to have known or been aware that there was smoke; therefore, I had to have believed there was. So, too, in the corresponding counterfactual conditional, we understand the antecedent, “If there hadn’t been smoke in the kitchen,” to be implicitly cognitive. We interpret it as excluding the possibility that there wasn’t any smoke in the kitchen but I mistakenly believed that there was. For if I had had that belief, then, of course, I would have run into the kitchen anyway. To put the point in terms of possible worlds: In interpreting the counterfactual that corresponds to a reason explanation, we consider only worlds in which the counterfactual condition c, specified by the antecedent of the conditional, lies within the agent's epistemic horizon: the agent knows or is aware that c—and, therefore, believes that c. We don’t allow the counterfactual “facts” to vary independently of the agent's beliefs.

I rejected the account that posits a causal chain, with belief an intermediary between fact and action.  But how are these counterfactuals to be understood, if not by such a chain? As the phrase “sufficient to move me” suggests, we seem to be talking about a sufficient reason or rational condition. That there was a red light just ahead was, together with contextual facts, reason enough to move me, whereas my speeding was not. Although Sam’s mediocre track record was not sufficient—that is, not reason enough—to move the committee to reject his project, the mathematical impossibility of carrying it out was rationally sufficient. In other words, although Sam’s track record was a consideration, the outcome did not hinge on it; only the mathematical impossibility of the project made a difference. Given these suppositions, then only one of the reasons was decisive and actually moved or led the committee to reject the project, namely, that it was mathematically impossible. [9]

It appears that we interpret some counterfactuals causally and others rationally, and I will try to show what the difference consists in. It will help to illustrate with some pairs of superficially similar counterfactual conditionals that are interpreted differently, one in each pair most naturally interpreted causally (c) and the other, rationally (r):

The distinction between (c) the causally interpreted conditionals and (r) the rationally interpreted conditionals is not to be understood in terms of a theoretical difference. It is a matter of a procedural difference, a difference in what we do in interpreting the corresponding conditionals. There are two distinct procedures for determining what is the case given a counterfactual premise, and the difference between these, I believe, explains the difference in the interpretations we give the (c) and the (r) conditionals in the above pairs. Any of these pairs, including the equivocal 5c/5r, could be used to illustrate the point, but I will work with 1c and 1r.

The Simulative Procedure

An important tenet of the simulation theory, at least as I conceive it, is that one may use simulated practical reasoning as a predictive device. [11] In its first step, the simulative procedure resembles a well-known procedure for evaluating conditionals, the Ramsey test. In the Ramsey test, the antecedent (in its indicative form),

There (wasn't) any smoke in the kitchen,

is added hypothetically to one’s existing stock of beliefs—which is to say, essentially, it is adopted as a pretend belief. Other beliefs are then allowed to change accordingly, as needed to avoid inconsistency. One might give up, for example, the knowledge that there is a loud, high-pitched alarm sound coming from the kitchen. (Some philosophers hold these adjustments to be justified by a possible world account, the minimal adjustments to beliefs corresponding to the closest possible world in which the antecedent is true.) [12] Then, the resulting modified set of beliefs may be sufficient to determine whether the consequent,

The alarm didn’t go off,

is true. The decision is constrained by generalizable properties of the actual world, properties that may justifiably be projected to the alternative world(s) stipulated by the counterfactual antecedent. Most people know very little about the way smoke detectors work, but it is part of our background knowledge that they are generally designed to go off if and only if smoke crosses their sensor. If we take it for granted that the smoke alarm is in working order, then, applying our background knowledge, we would conclude that the alarm did not go off.

Now, if there hadn’t been smoke in the kitchen, would I have run into the kitchen? Applying the simulative procedure, I pretend-believe that there isn't smoke in the kitchen, as before, and I make adjustments to avoid inconsistency. However, I add it, not only to my existing beliefs, but also to my existing desires and emotions, or, more broadly, to whatever psychological resources my decisions depend on. Then I decide whether to run to the kitchen or not. In other words, following this path, I use the counterfactual antecedent as a premise of simulated reasoning: as an input to reasoning the output of which is a simulation of what is to be predicted: a pretend decision if the counterfactual concerns what I would decide to do, a pretend belief if the counterfactual concerns what I would believe, and so forth.

Wondering what I would have done (“just now”) if it had been the case that p, I would typically pretend that p; then—in effect simulating or playing the role of myself in the world thus transformed—I decide what to do. This is also what I do sometimes in the case of others, albeit with additional adjustments necessary to transform myself imaginatively into them. Wondering what agent A would have done at time t if at t it had been the case that p, I first simulate A at t. Then, still simulating A, I pretend that p, in effect simulating A in the world thus transformed; then, in the role of A in the counterfactual condition p, I decide what to do. For example, having adopted the point of view of the committee members as they weigh the grant application—having simulated them—I then introduce a counterfactual premise: that Sam’s project is not objectionable on the grounds that it is mathematically impossible. This counterfactual supposition now becomes available as a premise of theoretical and practical reasoning; I can no longer avail myself of the premise that the project is impossible. To put in terms of “worlds,” first I imaginatively modify the world—the world as I myself take it to be, “our” world—to simulate “their” world. Once having “arrived” at their world, I then proceed to modify it. Then, within this twice-modified world, I decide what to do.

The two procedures I have distinguished are both familiar to us and have been so since early childhood—if not theoretically, then certainly in practice. [13] A child’s game of make-believe typically calls on both procedures: the Ramsey procedure for determining the nature of the counterfactual or make-believe situation within which one is called on to act [14] ; and the simulative, for deciding what to do in that situation. To take a favorite example of developmental psychologists: I pretend that the banana is a telephone. This initial premise connects with a body of general knowledge (or belief), including knowledge about manipulable physical objects in general, special knowledge about typical telephones, and a rudimentary understanding of sound. If the banana is a typical phone, then one end (perhaps arbitrarily chosen) is the mouthpiece into which one speaks and the other is the receiver out of which sound comes; there are imagined push-buttons or a rotating dial on the banana-phone or on an imagined base to which the phone may or may not be tethered by an imagined wire; the voices of familiar people can be heard through the phone, but only after it has rung and one picks it up, or after one has pushed buttons or made the dial rotate in certain ways; and these voices cease when one puts the banana-phone down; and so on. These “productive” consequences of the counterfactual premise are derived by applying projectible knowledge—that is, knowledge (or belief) that is projectible from the actual world to the pretend world of the game. [15]

Games of make-believe involve something more than adding a counterfactual premise to one’s existing beliefs. When young children pretend that the banana is a telephone, they don’t just theorize about the properties of their banana-telephone: They go on to use it as a telephone. They speak into it and they pretend to hear someone speaking through it. There is, then, a further productive feature of games of make-believe: a solution to the practical problem, “Now that I have a telephone in my hands, what shall I do with it?” What the child does with the phone depends on more than the stipulated pretend-facts along with applied general knowledge or beliefs. The counterfactual premise, “The banana is a telephone,” becomes a premise of practical as well as theoretical reasoning. They might reason: “This banana is a telephone. If it’s a telephone, then if I pick it up, I can speak to Mommy. So I’ll pick it up and push the buttons.” Obviously, this practical reasoning will call on mental resources beyond the general knowledge or beliefs needed for theoretical reasoning: desires, values, norms, and the like. [16]

One Important Consequence

One major consequence of the simulative departure from the Ramsey test is that on the simulative reading the counterfactual antecedent specifies a more restricted set of possible worlds. For one not only stipulates counterfactually that p, thereby specifying a set of possible worlds; one also plays the role of an agent A in the world(s) thus specified. Like the child who uses the banana-telephone as a telephone, one inserts oneself into the counterfactual world(s) one imagines. One does this either as oneself or, if one is role-playing or simulating another person (or oneself at another time), as the transformed self one plays. In the counterfactual world in which one acts, therefore, there exists at least one agent. Moreover, it is essential to the simulative procedure—at least, the simple procedure I have described—that the counterfactual “fact” be available to A as a possible basis for inference and emotional response. Putting the point differently, it is essential that A is aware of this “fact.” The procedure thus restricts possible worlds to those in which A exists and A knows, or is aware, that p and therefore believes that p (and does not believe that not-p). In the pizza example, the antecedent,

there isn't smoke in the kitchen

excludes possible worlds in which there isn't smoke in the kitchen but I falsely believe that there is. Even though the antecedent of the conditional makes no mention of beliefs, it is procedurally inconsistent with the statement that I do not believe that there is no smoke (and, a fortiori, the statement that I do not exist). In this respect, any counterfactual conditional interpreted by the simulative procedure is interpreted as implicitly cognitive. Thus, a procedural difference would readily explain why we understand the antecedent of,

Is it (counterfactually) the case that p?

and,

Does A (the agent simulated) believe that p?

In other words, it does not allow the possibility of disparate answers to the objective question,

What are the facts?

and the subjective question,

What are the facts to the agent, from the agent's point of view: that is, as the agent believes them to be? [17]

It is particularly interesting to note that the simulative interpretation, despite restricting possible worlds to those in which the target (the person simulated) knows that p, doesn’t demand that the simulator possess the concepts of belief or knowledge—again, as in the case of the child who uses the banana-phone as a phone. What I will argue elsewhere is that this has consequences for both the nature of these concepts and the means of their acquisition.

How Reasons Explain

How does a reason-in-favor or justificans explain action? My answer, shorn of refinements, is that in the simple case:

Reason R1 explains A’s X’ing in that, without reason R1, A would not have X’d.

The simple case is one without rational overdetermination or pre-emption. It leaves aside cases in which the following holds:

If A had either R1 or a second, independent reason R2, A would have X’d, but if A had neither, then A would not have X’d; and (in overdetermination) A did have both R1 and R2, or (in pre-emption) if A did not have reason R1, A would have had reason R2.

When I consider A as being counterfactually “without reason R,” I proceed by subtracting or taking away R from my simulation of A. This I may do in either of two ways. Where R=that p, one way to remove the reason—the simple way—is just to pretend or suppose that not-p. This step is the homologue of what we do when we try to isolate causal factors. Consider one of the examples given earlier. To test whether the windiness causally explains why the match failed to light, we would suppose it “taken away” or at least reduced in level, and then ask whether the match lit. In doing this, we evaluate the relevant counterfactual conditional,

(c) If it hadn’t been so windy, the match would have lit.

Likewise, to test whether the wind rationally explains—that is, explains as a reason—why Marge did not strike the match, we would again want to “take away” or at least reduce the windiness, and then ask whether Marge struck the match. Thus, we evaluate the relevant counterfactual conditional,

(r) If it hadn’t been so windy, Marge would have struck the match.

The difference (or, at least, the chief difference) is that, whereas in the (c) example, one uses the Ramsey procedure, in the (r) example, one uses the simulative procedure—as in the case of the smoke alarm’s response to the presence of smoke and my response.

This, I said, is the simple way to “take away” a reason. There is a different way, which has no causal homologue and appears to be peculiar to reason explanations. I remarked earlier that the reason explanation for my response to the smoke excludes the possibility that I mistakenly believed that there was smoke, noting that if I had had that belief, then I would have run into the kitchen anyway. However, we—a “we” that evidently includes nearly every human being over the age of four—understand that beliefs may vary independently of facts and may be manipulated to deviate from facts. Where their beliefs deviate from fact, we explain what people do in terms of their deviant belief or beliefs. Even when beliefs do not deviate, and (further) even when people know what is the case, we might persist in explaining their actions in terms of their beliefs.

However, I have not given an account of explanations in terms of beliefs (or belief-desire pairs). Until I do, it is open to someone to object that the account I have given of reason explanations is at best incomplete. Happily, I do have an account of belief explanations. It starts from the fact discussed earlier, that an ascription of knowledge, and consequently an ascription of belief, is implicit in and derivable from the simulative procedure for evaluating counterfactuals. What I will argue is that a belief explanation is a weakened form of a reason explanation. As Heal says, “the capacity to think about thoughts must be seen as an extension of the capacity to think about their objects.” To which I add that the capacity to explain in terms of thoughts—and beliefs—is an extension of the capacity to explain in terms of their objects. This will have consequences concerning the development of epistemic concepts such as awareness, knowledge, and belief. The simulative procedure is our bootstrap into cognitive concepts. If by using the procedure one is already making implicit attributions of knowledge (and therefore belief), and if children use such a procedure before they have these concepts, then the developmental question for psychologists should be, How does the child make these implicit attributions explicit? Acquisition or refinement of a theory does not seem to be the right sort of answer. Rather, what is required is a complex variant of the simulative procedure, involving a kind of double pretense that is also to be found in sophisticated games of make-believe. [18]

This is all promised for the “next chapter,” however. Until I make good on it, it is open to someone to argue, not only that my account is incomplete, but that it puts the cart before the horse: that, even if the simulative procedure allows us to give reason explanations without employing the concept of belief, reason explanations reduce to causal explanations in terms of beliefs or beliefs and desires—psychological explanations—plus some other component that somehow ties the believing that p to the fact that p.

However, if I am able to show how the ascriptions of knowledge and belief that are implicit in the simulative procedure can be made explicit, then I believe I will have made a very good case for my simulative account of reason explanation. According to that account, reason explanation is not causal explanation, but it is the simulative homologue of causal explanation. That is, it bears whatever logical connection to counterfactual conditionals causal explanation does, but to conditionals evaluated by the simulative procedure rather than the Ramsey procedure. This—the similarity and the difference—would explain why

·        like causal explanations, reason explanations explain

·        unlike causal explanations, reason explanations allow as explanans necessary, normative, or future-dated facts

·        unlike causal explanations, reason explanations presuppose that the explanans is a fact that is known to the agent.

Notes