Although the simulation theory (ST) was developed as a solution to problems in the philosophy of mind, it bears obvious affinities to theories that particularly concern the aims and methodology of the social sciences: most notably, those centered on the concepts of Verstehen, or “empathetic understanding” and historical reenactment.[1]
These older cousins of ST do not enjoy the best of reputations, however. Among analytically oriented philosophers, at least, they are best known for their anti-naturalist portrayal of ordinary, everyday reason explanations: that is, explanations of actions in terms of the reason (or reasons) for which they were performed. A major thesis of some Verstehen theorists was that, whereas the natural sciences attempt to explain (Erklären), so-called explanations in the social sciences, as well as ordinary "common sense" reason explanations, are really not explanations at all, but merely attempts to interpret, that is, to understand (Verstehen) actions, to indicate how they might have been seen by the agents themselves. In the 1950’s and 1960’s, several analytic philosophers, influenced by Wittgenstein as well as by Continental views, claimed that what I am calling reason explanations have merely a justificatory, or normative, aim.[2] According to William Dray’s account, for instance, they involve an attempt to view the world and the agent’s situation as the agent saw them at the time of action, and to pick out those aspects the agent saw as justifying the action. They portray the action as a reasonable thing to have done, given what the agent was aiming at and what the agent believed to be the relevant facts.
Does ST share these views? It would be presumptuous of me to speak for ST generally, for the term, “the simulation theory,” has a broader coverage than just the theory I intended to be putting forward in my 1986 paper, “Folk Psychology as Simulation,”[3] and in subsequent elaborations. My own version of ST definitely shares some of the anti-naturalism of its predecessors, in ways I hope to make clear. Most important, it denies that reason explanations must be interpreted against a background of laws, or at least that they imply, presuppose, or entail that there is a relevant causal law. Yet this anti-naturalism does not deny that they are explanations, and that they perform an explanatory function beyond merely portraying the action as reasonable from the agent's point of view. They surely do answer “Why?” questions, and they may have the form, “A X’d because p,” where “p” states a reason of A’s for X’ing. How they do this is the topic of this paper.
A major shortcoming of the predecessors of ST was pointed out by Donald Davidson in his classic 1963 paper, “Actions, Reasons, and Causes.”[4] One way to describe the problem is that there is a logical gap between having a reason for X’ing and actually X’ing for that reason. That there is such a gap is most obvious where the agent has more than one reason for doing something, yet is actually moved to action by only one. (The problem of filling that gap I will call the mixed motives problem.) For example, Sam, out for a stroll, is caught in a sudden rain, and also finds himself in a dangerous district. Sam runs. Is Sam running because it is raining (and he will get drenched if he doesn’t) or is he running because he is in a dangerous district (and he will be in danger if he doesn’t) -- or for both reasons? If one says,
1. Sam is running because it’s raining,
or,
2. Sam is running because he is in a dangerous district,
one is saying something that answers this question. One is not just specifying a fact that made the action appear to the agent reasonable, rational, justified, or attractive; one is more particularly specifying a fact that is a reason for which Sam is running.
Davidson’s solution to the mixed motives problem is that what differentiates those reasons that are reasons for which (or because of which) the action was performed from those that are not is a causal relation between the reasons and the action. The “because” in “He did it because...,” where we go on to name a reason, signifies a causal relation, and the explanation is construed as a causal explanation. What makes it not just an “accident” that A X’d, given that A had reason R for X’ing, is that reason and action are linked, under some description of each, by some law of nature -- though the explainer may not know what law this may be. Reason explanation thus not only requires something beyond justification from the agent’s point of view; the further ingredient is a naturalistic one, having to do with the way nature operates rather with the way the agent sees things. This is not to say that Davidson’s account of mind, of “the way the agent sees things,” is naturalistic; far from it. What is naturalistic -- and adds to his account the tension expressed patently in his doctrine of anomalous monism -- is his view, to be discussed shortly, that if mental events are to explain actions, they must be token-identical with events that are covered by laws.
Davidson does not explicitly discuss reasons such as,
It is raining.
I am in a dangerous district.
He does not claim that reasons such as these, reasons or considerations that one might put forward in arguing for running, are causes of the action for which they are reasons. It would not be his position that (1) entails,
Sam’s running was caused by its raining (or: by the fact that it is raining, by the precipitation, etc.).
and that (2) entails,
Sam’s running was caused by his being in a dangerous district.
This is just as well. Reasons for acting, some of which are the reasons for which people act, are in fact a motley lot. It would require a broad understanding of what a causal relation is to allow being in (or entering) a dangerous district as a cause of anything. If this is not convincing, consider:
Joan flew to Hawaii, because there will be a solar eclipse there tomorrow.
Max abandoned his thesis, because it was contradictory.
I bought 32 feet of fencing, because I’m enclosing a 10-foot diameter, and pi is just under 3.2.
Davidson’s conception of the relevant causal relation is not broad enough to accommodate these reason explanations as they stand. Central to his conception is the Principle of the Nomological Character of Causality: that two events are causally related only if they are covered, under some description of these events, by a law of nature; that is, that there is a law of nature that correlates the occurrence of events of the one description to the occurrence of events of the other description.[5] Davidson further claims that these must be strict, exceptionless laws. Tomorrow’s solar eclipse is no doubt covered by such laws, but these are not laws that “predict” Joan’s flying to Hawaii today. That pi has the value it has is a necessary truth, which is not covered by any laws of nature; likewise, that Max’s thesis (taking “Max’s thesis” as a rigid designator) is contradictory.
To maintain his causal analysis without compromising his nomological conception of causation,[6] 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. Thus he holds that a reason explanation is always, at least implicitly, a psychological explanation. When made fully explicit, it is always an explanation in terms of a property of the agent, specifically, the agent's having certain beliefs and desires. This is a point that has in recent years become dogmatically “obvious” to many philosophers. In the case of,
1. Sam is running because it’s raining (and if it is raining, running helps avoid getting drenched),
the more explicit version would be,
1B. Sam is running because he believes that it is raining (and that if it is raining, running helps avoid getting drenched).
Further, it is the standard view -- and Davidson’s -- that 1 and 1B are enthymematic. Reason explanations are said to portray the action as a reasonable thing to have done, given (a) what the agent believed to be the relevant facts and also (b) what the agent was aiming at. It is because of qualification (a) that mention is made of the agent’s beliefs; and it is because of qualification (b) that mention is made of the agent’s desires or, generically, pro-attitudes. Davidson thus identifies a “primary” reason with a pair of propositional attitudes, namely a belief and a pro-attitude, adding that typically in giving an explanation one would not need to mention explicitly both the belief and the pro-attitude.
The question, “Why is Sam running?” is for Davidson chiefly a matter of which of two belief-desire pairs actually caused his running: (a) the one consisting of his belief that he will get drenched unless he runs and his desire not to get drenched, or (b) the one consisting of his belief that he will be in danger unless he runs and his desire not to be in danger. Or, in a case of causal overdetermination, each pair a and b independently causes his action. In short, a reason explanation represents an action, that is, an instance or token of some general action type T, as caused by beliefs and desires that correspond to premises of a practical syllogism that provide a basis for performing an action of type T. Thus, even though reason explanations aim at a kind of understanding not sought in explanations of other events, showing the action to be rational from the agent's point of view, a causal component had to be present as well. Not only are rational and causal explanation compatible, according to Davidson: Reasons explain why an action is performed only if they -- that is, the relevant beliefs and desires -- are causes of the action. To assert that an agent acted because of a particular reason is to give a causal explanation of a special kind.
Summing up so far: An apparent advantage of analyzing reason explanations as belief-desire explanations rather than as simple nonmental explanations is that the former, but not the latter, seem to be the sorts of entities that might be held identical with events that are covered by laws. The reason for preferring entities that might be held identical with events that are covered by laws is that (given the nomological conception of causation) this is required for a causal relation. A causal relation, in turn, seems to give us a solution to the mixed motives problem.
It is not clear why Davidson insists on a causal relation between reasons and action, rather than merely a counterfactual relation. It would appear that a counterfactual difference between primary reasons a and b would be enough to fill the gap between having a reason to X, and X’ing for that reason. Suppose the following:
If Sam did not have belief-desire pair a, then he would not be running -- that is, he would not, even if he had belief-desire pair b, and other things were “equal.”
However, it is not the case that, if Sam did not have belief-desire pair b, then he would not be running -- that is, he would not, even if he had belief-desire pair a, and other things were “equal.”
If there is this counterfactual difference between the two belief-desire pairs, then it would seem that Sam is running for just one reason, namely, that it is raining. (In saying this, I am ignoring problems that are common to a causal account of reason explanation and a counterfactual account. For example, just as there is a problem of “deviant causal chains,” so there would be a problem of deviant counterfactual chains. Sam’s running may be counterfactually depend on a belief-desire pair, but in the “wrong” way, a way that does not make this a reason for which he is running.) Not only does the counterfactual difference seem to fill the gap just as well as a causal relation does; it is plausible--though I will not argue the point--that it is only because the causal relation has these counterfactual implications that we are inclined to give credence to a causal account of reason explanation. In other words, what makes the causal account seem compelling, at least in my example, is that it would yield the counterfactual difference I have indicated between a and b.
If counterfactuals suffice to solve Davidson’s mixed motive problem, however, then haven’t we eliminated the need to think of reason explanations as belief-desire explanations? One problem with understanding reasons as belief-desire pairs is the fairly obvious one that it is not Sam’s believing it is raining that is a reason of his. It is not sentences that ascribe a belief that argue for the action, but rather sentences that express the belief: for example, “It’s raining,” or “I will get drenched if I don't run.” It is the certainty of getting drenched, not the agent's being in a state of certainty about it, that is a genuine reason for running, a consideration in favor of running. Likewise, in the case of desires or other pro-attitudes: apart from special cases, it is the sentences that express them, not those that ascribe them, that are premises of practical arguments: for example, “It would be awful to get drenched now,” not, “I have a desire that I not be drenched now.” Can't we, then, let reasons be reasons, that is, considerations that might be raised in arguing for the action, rather than states of the agent for whom they are reasons? Suppose:
If it were not raining, then Sam would not be running -- that is, he would not, even if he were in the dangerous district, and other things were “equal.”
However, it is not the case that: If Sam were not in the dangerous district, then he would not be running -- provided it were raining, and other things were “equal.”
If this is so, then it would seem that Sam is running for just one reason, namely, that it is raining. (In saying this, I am again ignoring problems that are common to a causal account of reason explanation and a counterfactual account.)
If we assume that counterfactuals suffice for the mixed motive problem, then perhaps we need not abandon the “motley lot” I spoke of. We need not worry whether being in or entering a “dangerous district” can plausibly be identical with an event covered by a law of nature -- much less, whether a necessary truth can be. This may allow us to pick up all the fallen dominoes. If we don't need a causal relation to solve the mixed motives problem, then perhaps we don't need to limit the explanans in a reason explanation to something identical with an event that is covered by laws. And if this is so, then we don't need to insist that the real explanans in a reason explanation must always be a mental state rather than a reason proper, a consideration that might be raised in arguing for the action. Indeed, we may not even have to abandon causality if, as some philosophers think, causality itself may be understood in terms of counterfactuals: for example, that event c caused event e provided that, if c had not occurred, e would not have occurred.
There is a problem, however. For it is widely held that counterfactuals themselves rest on laws. Using a simplified possible-world approach to counterfactuals, Jaegwon Kim (Philosophy of Mind, 141-142) explains clearly why this is supposed so:
The counterfactual "If P were the case, Q Would be the
case" is true just in case Q is true in the [possible] world in which P is
true and that . . . is as much like the actual world as possible (to put it
another way: Q is true in the closest P-World) . . . Let's see how this works
with the counterfactual "If this match had been struck, it would have
lighted." In the actual world, the match wasn't struck; so suppose that
the match was struck . . . but keep other conditions the same as
much as possible . . . it was dry, and oxygen was present in the vicinity. Did
the match light in that world? In asking this question, we are asking which of
the following two worlds is closer to the actual world:
W1: The match was struck; it was dry; oxygen was present; the match lighted.
W2: The match was struck; it was dry; oxygen was present; the match did not light.
We would judge, it seems, that of the two W1 is closer to the actual world. But why do we judge this way? Because, it seems, we believe that in the actual world there is a lawful regularity to the effect that when a dry match is struck in the presence of oxygen it ignites, and W1, but not W2, respects this regularity. . . .
Consider a psychophysical counterfactual: "If Brian had not wanted to check out the noise, he wouldn't have gone downstairs." . . . Consider the following two worlds:
W3: Brian didn't want to check out the noise; he didn't go downstairs.
W4: Brian didn't want to check out the noise; he went downstairs anyway.
. . . So why should we think that W3 is closer than W4? . . . The only plausible answer, again, seems to be this: We know, or believe, that there are certain lawful regularities and propensities governing Brian's wants, beliefs, and so on, on the one hand and his behavior patterns, on the other, and that given the absence of a desire to check out a suspicious noise in the middle of the night, along with other conditions prevailing at the time, his not going downstairs at that particular time fits these regularities better than the supposition that he would have gone downstairs at that time. . . .
Again, the centrality of psychophysical laws to mental causation is apparent. . . . the laws involved in evaluating these counterfactuals . . . are rough-and-ready generalizations tacitly qualified by escape clauses . . . Laws of this type, often called ceteris paribus laws, seem to satisfy the usual criteria of lawlikeness: . . . they seem to have the power to ground counterfactuals, and our credence in them is enhanced as we see them confirmed in more and more instances. . . . it seems beyond question that they are the essential staple that sustains and nourishes our causal discourse.
The ceteris paribus laws Kim mentions are different from the strict laws Davidson requires, and they may seem to support psychological counterfactuals of the sort Kim mentions. But the choice is not just between two sorts of law. For laws are not the only generalizations that satisfy the criteria of lawlikeness: there are others that support counterfactuals and are confirmed by their instances. Moreover, contrary to the widely held conviction that Kim articulates, it appears to be these others, not laws, that sustain and nourish our explanations of action and in particular our determinations of the reason or reasons for which the agent acted. To illustrate my alternative view, I will use a historical example.
Mixed motive questions, concerning which of the agent's reasons are reasons for which the action was performed, are often of considerable practical, moral, and legal import; they are not just philosophers’ questions. In lawsuits concerning discrimination in employment, the crucial issue, typically, is whether illegitimate considerations of gender or race, for example, swayed a decision to reject a candidate for hiring or promotion, rather than just job-relevant considerations. Was candidate x chosen rather than candidate y because x was the better person for the position, or was this consideration one among others, such as race or gender?
The difficult cases are those in which a case is made that both types of consideration point in the same direction, and the possible influence of biases is masked by the possible presence of legitimate, job-relevant considerations. In a case decided by the US Supreme Court in 1989, a female employee of Price Waterhouse, a major accounting firm, highly regarded for her competence and intelligence, had been turned down for promotion to a partnership in the firm. She sued on the grounds that the decision was discriminatory.[7] The firm admitted that there had been widespread sex-stereotyping: She had been held "too assertive for a woman," her behavior too masculine. She had been advised that to improve her chances for partnership, she would have to become more “feminine” in the way she walked, talked, and dressed.[8] But the firm argued that there were also legitimate, non-discriminatory grounds for denying her the position, such as a “lack of leadership qualities,” and it claimed that the decision was influenced only by these latter considerations. They further maintained that it was the burden of the employee to show that the decision had been due solely to the unallowable, discriminatory consideration; that, “but for” the gender factor, the decision would have gone the other way.
The Court ruled to the contrary.[9] The relevant statute forbids an employer to “deprive any individual of employment opportunities ... because of such individual's ... sex.” Congress, the Court plurality notes, had specifically rejected a qualifying amendment that would have replaced “because of” with the more restrictive phrase, “solely because of.” Therefore, the employee plaintiff has only to prove that gender was one of the factors on the basis of which the decision was made, not that it was a “but-for” cause in the absence of which the decision would have gone the other way. It then becomes the burden of the defending employer to prove (by “a preponderance of evidence”) the contrary, that it is false that,
If it had not taken the illegitimate factor, the employee’s gender, into account, it would have decided in favor. [10]
Or, in other words, the burden is to prove that the decision was gender-independent, where the criterion of gender-independence is the counterfactual,
GI. The partners would have made the same decision irrespective of gender -- even if the candidate had been a male with like credentials.
For earlier and later decisions to count as evidence of the influence of particular reasons on the promotion decision under examination, a crucial assumption must be made: that the firm did not change its standards for promotion to partnership; or, more broadly, its policy with regard to such promotions. Price Waterhouse certainly had no official policy requiring that female partners act in a “feminine” way. But, given the behavior pattern discerned in other employment decisions and given internal communications, there appeared to be an operative policy, a de facto standard, permitting such considerations to count in evaluations of a candidate. Assuming that there was no change in policy, then one may infer that in the case under examination such considerations were again allowed to influence the decision. The charge of sexism in this case may be confirmed or disconfirmed by examining other decisions, given the assumption that the underlying policy does not vary.
The Court was quite clear about this assumption. Despite substantial evidence of a general pattern of discrimination in Price Waterhouse’s past promotion decisions, the Court reaffirmed an earlier decision that in adjudicating an individual claim, as opposed to a class action, it is not sufficient to establish that such a pattern existed. For the relevant statute
does not authorize affirmative relief for individuals as to whom, the employer shows, the existence of systemic discrimination had no effect.[11]
In an individual case, it was argued, the focus is on the influence of particular reasons on a particular hiring decision. Thus, Price Waterhouse was given the opportunity to show that, despite the general pattern, in this case the counterfactual GI was true: They would have made the same decision even if they had not taken the plaintiff’s gender into account. If there had been a general policy, it may not have been applied in this particular case. The Court evidently had an insight that was not granted at least some of the philosophical behaviorists: namely, that, where the concern is to explain a particular decision or action, the relevance of a general pattern of behavior is merely evidential. It is evidence, but not conclusive evidence, that the corresponding counterfactual conditional held in the particular case in question. They left it open to the firm to show that in the particular employment decision in question, the outcome was just what it would have been had they not been systemic discriminators.[12]
Assuming consistency of policy is crucial, too, as one moves between actual and counterfactual cases. Applying Kim’s possible world analysis, consider the following two possible worlds:
W5: The candidate was a male with like credentials; the partners decided in favor of the candidate.
W6: The candidate was a male with like credentials; the partners decided against the candidate anyway.
Why should we think that W5 is closer than W6? According to Kim’s view, the only plausible answer is that we know, or believe, that there are certain lawful regularities governing the wants, beliefs, and so on, of the voting partners, and their behavior patterns; and W5 better fits these regularities. A law, of course, covers any member of a set of individuals and ignores differences among these individuals. For a law to be applied to a given individual i, it must be complemented with specific parametric data concerning i. Thus, in the case of Price Waterhouse, the differences between a company that is that is influenced by sexist considerations and a company that is not would be counted differences in data, whether these concern propositional attitudes or physical or computational states; the laws themselves are invariant from one individual or company to another.
There is, however, another answer, and one that seems to me far more plausible. There are lawful regularities of a sort, but they concern, not relations among physically described events, nor the invariant formal relations between wants, beliefs, and actions, but rather substantive relations between situations and actions, mapping types or descriptions of situations onto types or descriptions of actions, and varying across agents and, within a single agent, over time. In the case of Price Waterhouse, for example, on the basis of prior decisions, internal correspondence, and other evidence, we know, or believe, that the firm, or a sufficient number of the voting partners, was in this particular decision applying a certain standard for promotion to partnership; or, more broadly, implementing a policy with regard to such promotions. And, once we take away the gender factor (with other conditions remaining as they were), it better fits this standard or policy to decide in favor of a candidate with such credentials than to decide against.
If the agent is successfully implementing the same policy in
a number of decisions, then there will be a corresponding correlation between
descriptions of the situations in which the agent acted and descriptions of
what the agent did in those situations. This will be a non-accidental
correlation. The statement that at a
particular time period a particular agent is implementing a certain policy is
of course not the statement of a law of nature. Yet one may express a policy
as a law-like principle: for example,
JG. If x is a candidate for promotion, x is promoted if and only if x is both job-qualified and gender-qualified.
JG would be confirmed by promotions of
job-and-gender-qualified candidates and by denials of promotion to candidates
who are either not job-qualified or not gender-qualified. JG is disconfirmed
by promotion of candidates who are either not job-qualified or not
gender-qualified, or by denial of promotion to a job-and-gender-qualified
candidate. In addition, if JG is the expression of a general policy, it clearly
supports counterfactuals, specifying whether purely hypothetical
candidates would have been promoted and also whether a particular actual
candidate would have been promoted if, counterfactually, she had been
gender-qualified.
The Court, as I mentioned, held that a record of systemic discrimination was not decisive if it could be shown that in the case in question the defendant would have made the same decision even if they had not taken the plaintiff’s gender into account. However, the Court failed to note that the counterfactual may be false for reasons having nothing to do with sexism. For example: Suppose that, if the partners had not considered gender, their decision would have been favorable-- but only because the meeting would have been shorter. Suppose it is a psychological law that the longer a meeting runs, the less likely it is that the discussants will vote in favor of the matter under discussion. If the partner’s meeting had been shortened either because gender was not considered or because other business was expedited, the decision might have been favorable. Likewise, the decision would have been unfavorable even if the gender issue had not been taken up, provided other business had extended the meeting. In such a case, where the relevant fact is not the issues discussed but the length of the meeting, what should the Court decide? Strictly, the partners would fail the counterfactual test: They would not have made the same decision if they had not taken the plaintiff’s gender into account. And yet, although the decision was influenced by an “extraneous” factor (meeting length) that was affected by a discussion of gender-related matters, the decision is not, intuitively, an instance of gender discrimination.[13] In short, what should be held determinative of a discriminatory decision is not the truth[14] of the counterfactual conditional but the truth of a gender-discrimination generalization such as JG, which in this case was true because a corresponding company policy was in force and being implemented in that employment decision.
However, if we are interested in a general characterization of what must be assumed constant when projecting from actual behavior patterns to future or counterfactual behavior, “policy” isn't the concept we want. Even systematic race or sex discrimination may reflect, not a policy but a general sense of what is appropriate or right, even morally right. Or it may simply be the unprincipled expression of a general attitude, socially sanctioned or not. And yet, given a general attitude or a general sense of what is right, and no countervailing reasons, one’s behavior may resemble that of a person carrying out a policy. There will be a corresponding correlation between situations and actions, and it will not be an accidental correlation. That is, the lawful regularity JG will characterize the substantive relations between situations and actions, and one would be warranted in projecting from actual behavior to future or counterfactual behavior. JG will be confirmed by positive instances and disconfirmed by negative.
Turning to my other example: Suppose that Sam does not start running until he enters a dangerous district, although it had been raining for some time before. Further, he continues to run well after it stops raining. He stops running only when he has left the district. The evidence regarding Sam may be summarized as follows:
Phase 1: Raining but not in dangerous district: Sam does not run.
Phase 2: Raining and in dangerous district: Sam runs.
Phase 3: Not raining but in dangerous district: Sam runs.
We want to discover the answer to the following question: In Phase 2, is Sam running because (a) it is raining (and he will get drenched if he doesn’t run) or is he running because (b) he is in a dangerous district (and he will be in danger if he doesn’t run) -- or is he running for both reasons? It seems reasonable to infer from the evidence that in Phase 2, Sam was running because (b) he is in a dangerous district, and that he was not doing so because (a) it was raining.
To warrant this inference, however, we must assume that in
the relevant respects Sam acts consistently in all three phases. Suppose
that Sam does not act consistently in the relevant respects throughout
the three phases. Suppose that in Phase 1 Sam was indifferent to getting
drenched, or even thought it a great idea. But, just as he entered the
dangerous district, he suddenly became (perhaps through a quasi-religious
experience) a rain-avoider. Consequently, he started running. Thus, in the
ambiguous Phase 2, Sam was in fact running because it was raining; this
was the sole reason for which he was running. Further, suppose that,
just as it stopped raining, he suddenly came to see, by an equally obscure
process, being in a dangerous district as a reason for running. So,
although his original reason for running -- the rain -- was no longer
applicable, he continues to run. Then, despite Sam’s behavior before and after
Phase 2, we were mistaken in inferring that in Phase 2 Sam was running because
he was in a dangerous district. Our inference is warranted only if Sam was not
inconsistent in the way described; if, on the contrary, it is lawful or
non-accidental that in all relevant respects he would have acted similarly
in similar situations in each of the three phases.
What I have been discussing are reason explanations in the strictest sense: explanations of action in terms of the reasons, the favorable considerations or arguments, for which they were performed. These are probably the most common form of explanation of action, and they should be of particular interest to friends of the simulation theory. For they are the explanations we give, typically, of our own actions, either current or in the recent past: Why am I running? Because it’s raining; because I’m late; because I’m training for a marathon; because I’ll live longer if I do.[15] These answers not only indicate something about the world that explains why one so acted, but indicate that it explains it because to the agent this was a consideration favorable to so acting.
Other people, of course, or the agent at another time, might not find the reason a good one; they might not view the consideration as favorable to this action. We have only to think of an employer explaining that an employee was turned down for promotion “because she wasn’t feminine enough.” One might recognize the explanation as successfully explaining the action, even if it fails to justify it, to show it in a favorable light, or even to make minimal “sense” of it. To be informed that Sam is running because, say, “there's a gibbous moon” is to learn why he is running, but such an explanation is likely to introduce a question that leaves one more perplexed than before. We needn’t be told this by Sam. By noting the occasions on which he runs without any other evident reason and by applying Mill’s methods, we may even discover for ourselves that (apparently) Sam is running because there's a gibbous moon. We surmise that there is a longer story to be told or unearthed that will make his running make sense at least from Sam’s point of view: that gibbous moons have certain astrological influences, for instance, affecting those who do not run.
It is here, within our commonsense thinking about action, that the traditional distinction between explanation and understanding appears to have its place. Simulation in the light of Sam’s background and his behavior at other times may help us overcome our perplexity, enabling us to answer the question, “What does a gibbous moon have to do with it?” But we have an explanation nevertheless, even if it is one that doesn’t make the action make sense. In the case of a perverse or evil action, in fact, one just might not want to make it make sense; one doesn’t want “understanding,” even in the context of pretend play or biography. So too in the case of an unobjectionable action done for a wrong or perverse reason, such as an act of “charity” performed solely because it will make the recipient feel inferior: one might not wish to make doing it for that reason make sense. I have no interest in showing that an explanation in terms of the agent's reasons is a genuine explanation in the absence of empathetic understanding, for I do not think the concept of explanation will bear the weight of a “genuineness” test. I do want to note, however, that it is sometimes very important to get such an explanation right. To know that Sam is running because the moon is gibbous may rule out other explanations, and, like an explanation in the physical sciences, it may help in the prediction and control of future behavior.
Although I have been stressing that reason explanations are explanations in terms of facts that count as considerations in favor of so acting, I fully appreciate that, where the explanans in a reason explanation is a fact, it must be a fact of which the agent is aware, a fact that is known to the agent; thus, the agent must have the corresponding belief. This distinguishes, for example, the explanation,
1. Sam is running because it’s raining,
from non-cognitive brute-cause explanations such as the following:
1a. Sam is running because he is on amphetamines.
1b. The sewer is running because it’s raining.
The accepted view (at least, the view I once accepted, taking it to be the accepted view) is that reason explanations and the corresponding counterfactuals are peculiarly complex sorts of explanations and counterfactuals. We understand from (1) that Sam’s running is counterfactually dependent[16] on the fact that it is raining, by way of Sam’s belief -- indeed, his reliable belief, constituting knowledge -- that it is raining. Likewise, the partners’ vote was counterfactually linked to the candidate’s being a woman, by way of their belief that the candidate is a woman. With this understanding, when we consider counterfactual possibilities, we restrict possible worlds to the agent's horizon of knowledge, excluding worlds in which the agent does not know or believe that the antecedent, or the explanans, is true. This account would shore up and be shored up by the general picture of reason explanations as, at least implicitly, belief-desire explanations.
Such an account seems natural, given the commitments I once shared. However, to restrict possible worlds to the agent's horizon of knowledge would seem to require, on the accepted account, considerable conceptual sophistication. If (1) exemplifies a relatively complex form of explanation, excluding worlds in which the agent does not know or believe that p, then why do young children -- three-year-olds -- seem to use and understand this form, even though they lack the concepts of knowledge and belief, or at least the capacity to use them explicitly in explaining actions? Must we conclude that young children really do not understand reason explanations as distinct from brute, non-cognitive explanations such as (1a) and (1b)?
I will briefly sketch an alternative account that shows how, even if one lacks the concepts of knowledge and belief, one can give and understand explanations such as,
Sam is running because it’s raining,
that implicitly confine the explanans to the agent's epistemic horizon. According to my account, even the young child explains actions in terms of full-fledged reasons, not in terms of a generic non-cognitive explanans that befits a flood sewer just as well as a person. What takes conceptual sophistication, requiring explicit use of the concepts of knowledge and belief, is inclusion of possible worlds beyond the agent's epistemic horizon. Only the older child is capable of giving and grasping such explanations.
My alternative account rests on the simulation theory. To introduce the account, I begin by considering first person present tense explanations. Suppose that among the partners at Price Waterhouse who voted against the female candidate there was a particularly scrupulous individual who asked himself,
Would I be making the same decision if the candidate had been a male with like credentials?
To answer this question, he need only imagine the decision to concern a male candidate with like credentials. (This requires adjustments to accommodate the new premise -- most obviously, denial that the candidate is a woman.). That is, he assimilates the if-clause into his own practical knowledge base, taking its content, or what it asserts, as a possible datum for decision-making. It becomes available as a possible reason, as something that may be taken into consideration. He follows the procedure I described in the first of my “simulation” papers:
To simulate the appropriate practical reasoning I can engage in a kind of pretend-play: pretend that the indicated conditions actually obtain, with all other conditions remaining (so far as is logically possible and physically probable) as they presently stand; then—continuing the make-believe—try to “make up my mind” what to do given these (modified) conditions.[17]
I am not suggesting that this is always a reliable way to answer questions about one’s actions in hypothetical situations. The crucial point for the present paper is this: If the fact that p is assimilated into one’s practical knowledge base, as something that may be taken into consideration, then it is not beyond one’s epistemic horizon. Where this is done “off-line,” as in entertaining a counterfactual possibility, it excludes possible worlds in which one does not know or believe that p. Whatever I introduce as a possible premise for decision-making, as a fact that may taken into consideration, I introduce as something of which I am aware, as something known to me and therefore believed by me. I implicitly restrict possible worlds to my horizon of knowledge simply by milking the counterfactual premise for its practical consequences: by thinking, for example,
Supposing the candidate to be a male, other things being equal: How shall I vote?
Purely theoretical inference from counterfactual premises does not restrict possible worlds in this way. One may consider, for example, what would happen if the food one is eating were heavily contaminated with salmonella. Using background theoretical beliefs, one might trace the natural sequelae in one’s body without supposing oneself to be aware of them as they progress.[18]
To milk counterfactual premises for their practical consequences is child’s play. Children's pretend play sets up counterfactual premises for the main purpose of acting on them -- unlike the pure mind-play of, say, hypothetico-deductive reasoning. A child engaged in pretend play -- a young child, at least -- always, unavoidably, restricts possible worlds to her projected horizon of knowledge. Simply by allowing the explanans or the counterfactual antecedent as input to practical reason, she treats it as something of which she is aware, thereby excluding possible worlds in which it lies beyond her horizon of practically available knowledge. She does so, whether or not she possesses the concepts of knowledge and belief. Thus, even a young child explains her own action in terms of a full-fledged reason, not in terms of a generic non-cognitive explanans that befits a flood sewer just as well as a person. What does takes conceptual sophistication is, rather, inclusion of possible worlds beyond one’s current horizon of practically available knowledge: that is, worlds in which, so far as theoretical reason is concerned, it is the case that p, even though, so far as practical reason is concerned, it is not (or, epistemically, may not be) the case that p. To consider such worlds, the pretend-player must engage in a compartmentalized double pretense. The child keeps two tracks going. One of these makes its premises available to practical as well as theoretical inference: for example, “These globs of mud are tasty pies.” The other confines its premises to a purely theoretical compartment: “These pies have secret magical powers: eating such a pie immediately transforms one into a donkey.” The children innocently “eat” their pies; then, acting out the sequelae of unseen forces, they begin to walk on all fours and bray.
It is the same with counterfactual thinking and reason explanation. A double pretense is required -- with a corresponding conceptual sophistication, I believe -- to include possible worlds beyond one’s current horizon of practically available knowledge: that is, worlds in which, so far as theoretical reason is concerned, it is the case that p, even though, so far as practical reason is concerned, it is not (or, epistemically, may not be) the case that p. In contrast, simply by not running a double pretense, we treat the explanans in a reason explanation as a fact that is known to the agent. We treat the explanans as something known, simply in treating it as something not limited to theoretical reasoning -- as something of potential practical significance.[19] This we may do in utter innocence, without using or even possessing the concepts of knowledge and belief. We do it, as I said, just by not running a double pretense, thus excluding possible worlds in which there is a divide between fact and belief.
What about explaining the actions of others? You know, of course, what a simulationist will say about that: In the third person case just as in the first person, one treats the explanans or the counterfactual antecedent as accessible to practical reason. If I conceive the other's situation in its practical aspect, as a situation in which one is to act, then I conceive it as a situation of which the other -- the individual I am simulating -- is aware. Thereby, I exclude possible worlds in which there is a divide between fact and belief -- worlds, for example, in which it is not raining but Sam thinks it is, or worlds in which the Price Waterhouse candidate is really a male candidate successfully passing as female. Not to exclude these possibilities calls for a double pretense: One track of the pretense (it is really not raining, the candidate for promotion is really a male passing as female) is milked only for its theoretical (i.e., non-practical) consequences: Because it is not raining, poor Sam, needlessly running, would not get wet even if he were not running; because the candidate is really a male, the employer has unwittingly exercised his gender bias against someone of his own sex. In short, in reason explanations such as,
1. Sam is running because it's raining,
the implicit ascription of a complex epistemic state (e.g., Sam’s knowledge that it’s raining) is a consequence of not performing a certain complex act, that of making some counterfactual premises available to theoretical reasoning but not to simulated practical reasoning.
What I have briefly put forward in this final section is a new account of the cognitive dimension of reason explanation, an account I will develop further in another paper.[20] The main thrust of the present paper has been to free ST from a stigma earned by some of its predecessors: their failure to recognize that actions depend on the reasons for which they are performed. Because this dependence pivots on general features, reason explanations must be implicitly general, a trait they share with explanations in the physical sciences. However, I argued, against Davidson and many other writers, that the generality that is pertinent to reason explanations is not the generality of a scientific law, which assigns common causal properties to all entities, states, or events of certain kinds. In the Price Waterhouse case, we assume the invariance, not of law, but of the partners’ general standard or policy. What is relevant, in general, is the nature of the situation and the nature of the act, and we assume consistency of content, situation types being linked to action types. This provides the rails along which projection rides, from past to future and from actual to counterfactual.[21]
Robert M. Gordon
[1] On Verstehen, see Schutz 1962, 1967; and von Wright 1971: ch. 1. On reenactment, see Collingwood 1946.
[2] Among the more prominent of these philosophers are Dray 1957, Winch 1958, Melden 1961, Anscombe 1963, Taylor 1964, and von Wright 1971. I avoid the more common philosophical term, “rational explanation,” to avoid both its congenital ambiguities and the legacy of refined or technical senses introduced by others. I also avoid Davidson’s term “rationalization.”
[3] Gordon 1986.
[4] Davidson 1963.
[5] Davidson further argues that there can be no psychological or psychophysical laws at all, on the grounds that the assignment of content to beliefs and desires obeys a normative constraint of maximal rationality, which has no place in the application of objective laws of nature.
[6] One might accept Davidson's view that reason explanations are a species of causal explanation but deny that all causal relations are backed by laws. I will not be taking a stand here on either question: I will not be denying that reason explanations are causal or that causal explanations are always nomological. (I do deny the conjunction, however: for, in a later section, I argue that the implicit generality that underlies reason explanations is not nomological.)
[7] Price Waterhouse v. Hopkins, 490 U.S. 228 (1989).
[8] A lower court judge had remarked, “It takes no special training to discern sex stereotyping in a description of an aggressive female employee as requiring "a course at charm school." Nor . . . does it require expertise in psychology to know that, if an employee's flawed "interpersonal skills" can be corrected by a soft-hued suit or a new shade of lipstick, perhaps it is the employee's sex and not her interpersonal skills that has drawn the criticism.”
[9] They write:
To construe the words "because of" as colloquial shorthand for "but-for causation," as does Price Waterhouse, is to misunderstand them. But-for causation is a hypothetical construct. In determining whether a particular factor was a but-for cause of a given event, we begin by assuming that that factor was present at the time of the event, and then ask whether, even if that factor had been absent, the event nevertheless would have transpired in the same way.
[10] This standard with respect to mixed-motive cases was later modified (to plaintiffs’ advantage) by the Civil Rights Act of 1991.
[11] Thus, the Court allowed neither the agent's self-understanding nor the objective behavior pattern to be decisive. Like Davidson, the Court appreciated the logical gap between having a reason one sees as justifying one’s action, and acting for or because of that reason. The fact that sexist considerations were allowed to enter into the deliberations may have been symptomatic of the attitudes of many of the partners, attitudes that probably affected the way they “saw” and justified to themselves their decision not to promote. But establishing this leaves open the question whether sexist considerations, or indeed gender considerations of any sort, actually made a difference in the decision itself; or whether, on the contrary, the legitimate, gender-blind considerations alone would have led to the same decision.
[12] I should add that they overstate this position. They say that if legitimate, non-discriminatory considerations were strong enough to have led to exactly the same decision, then “the existence of systemic discrimination had no effect.” As in other cases of overdetermination, it is not correct to say, “had no effect”; what would be correct to say is that it had no differential effect. For if it had no effect, then it would be equally correct to say that the legitimate considerations (lack of leadership qualities, for example) had no effect either; whence, from the conjunction, it follows that no considerations had any effect on the decision!
[13] Except on a broad construal, perhaps, encompassing what might be called “de facto” discrimination. That is, there might be a bias toward exclusion of females, or “unfeminine” females, just because a small minority of sexists draws out what would otherwise be an open-and-shut case.
[14] Or substitute whatever term is appropriate for counterfactuals, if not “truth.”
[15] Each of these responses may be put in terms of a desire: Because I want to avoid getting drenched/ to avoid arriving too late/ to be in good shape for the marathon/to live longer.
[16] For reasons similar to those given earlier, we should not think of this as a causal link.
[17] Gordon 1986.
[18] This needs qualification. I might infer, e.g., from the counterfactual premise that most automobiles are powered by fuel cells, that Sam believes that most automobiles are non-polluting. I may infer this by a purely theoretical (i.e., non-practical) inference from, “Most automobiles are powered by fuel cells,” to, “Most automobiles are non-polluting.” If this inference occurs within the scope of a simulation, then an ascent routine yields, “I believe most automobiles are non-polluting,” where “I” refers to Sam, the individual simulated. In that case, I am implicitly restricting possible worlds to Sam’s horizon of knowledge: that is, it is implicit that in all possible worlds considered, Sam is aware that most automobiles are powered by fuel cells. In general, a question of scope arises whenever a counterfactual or hypothetical supposition is combined with a simulation -- a topic I intend to take up in future work.
[19] A very naive animist might do the same, of course, in explaining the sewer’s behavior, but few people are likely to reverberate emotionally and motivationally to a sewer’s “situation.”
[20] To be published in the journal Protosociology.
[21] Comments by the following people helped me see the need to write the last section of this paper: Fred Adams, Larry Davis, Joel Anderson, Philip Pettit, Piers Rawling, Carol Slater, Dona Warren, and the editors of this volume.