1. INTRODUCTION
There is an intimate relationship between knowledge and reasons for action. Our folk appraisals make this much clear – it is natural to use ‘know’ to defend and criticize action. Some philosophers have taken our natural usage to suggest that a principle like the following is true:
KA: You are in a good enough epistemic position to treat p as a reason for acting iff you know that p.
KA entails two principles, which I list for ease of reference:
K-NEC: You are in a good enough epistemic position to treat p as a reason for acting only if you know that p.
K-SUFF: If you know that p, you are in a good enough epistemic position to treat p as a reason for action.Footnote 1
KA is also motivated through appeal to intuitive reactions to certain contrast cases – cases where two subjects with the same epistemic position for some proposition p vary with respect to practical matters. One subject is usually described as being in a low stakes setting and possessing knowledge that p, the other in high stakes and lacking knowledge that p. Advocates of pragmatic encroachment argue that the best explanation of these contrast cases is that KA is true.Footnote 2 In fact, they suggest that intuitions about these cases are evidence for KA.
In this way, KA plays an important role in the motivation of pragmatic encroachment – the view that practical (non-epistemic) factors, such as the cost of being wrong – can affect whether or not one has knowledge. According to pragmatic encroachment, one can gain or lose knowledge as the practical features of one's environment changes. Pragmatic encroachment offers us a picture of knowledge whereby knowledge is unstable. Contrary to this ‘shifty’ view of knowledge, I contend that the stability of knowledge is an important feature of knowledge – it is one reason why we value knowledge.Footnote 3 Thus, I find this unstable view of knowledge troubling.
But pragmatic encroachment is committed to more instability than most discussions suggest. One surprising result of the arguments in this paper is that pragmatic encroachment is not merely about changes in stakes. All sorts of practical factors can make for the presence or absence of knowledge on this picture – stakes are just one factor among many that are knowledge-depriving. In this way, the focus in the literature on ‘stakes-sensitivity’ is misleading. Furthermore, insufficient attention has been paid to the variety of ways in which on this view pragmatic factors affect knowledge: pragmatic factors are not merely knowledge-depriving but are also knowledge-inducing.
These considerations bear strongly against the idea that knowledge stands in a necessary and/or sufficient relationship to appropriate reasons for action.Footnote 4 This does not imply, however, that there is no intimate relationship between the two. Rather, the relationship is more complicated than we might have hoped it to be.
1.2 Preliminaries
Before I begin, one preliminary point is in order. There are multiple ways you could be criticized for treating a proposition as your reason for acting. These principles speak only to the epistemic warrant required for one to be in a good enough epistemic position to treat a proposition as one's reason for action; they are silent with respect to the relevance of a proposition to the proposed action. A proposition can be sufficiently warranted to treat as a reason for acting while failing to be relevant to that action. For example, suppose my epistemic position for the proposition I ate breakfast this morning is close to maximal. I shouldn't treat this proposition as my reason to carry an umbrella today because it bears no relevance to my decision of whether to carry an umbrella. Nevertheless, it is warranted enough for me to treat as my reason to carry my umbrella.
Moreover, it's clear that sometimes we do something inappropriate when we fail to have an appropriate epistemic position towards some proposition we treat as a reason, even when that proposition is connected in the right way to the action. That there is a fire in the building is an excellent reason to leave the building, but if I have no idea there is a fire in the building it is epistemically inappropriate (and very odd) for me to treat there is a fire in the building as my reason to leave. I do not have the right epistemic position for there is a fire in the building to appropriately treat it as my reason. (In this way the language of ‘treat as a reason’ can be used to describe settings where one is epistemically appropriate and settings where one is epistemically inappropriate to use a proposition as one's reason for acting.)
In what follows, I will advance five independent problems for KA. The first is directed at K-NEC alone; the remaining problems are directed at the combination of K-NEC and K-SUFF. I will later discuss to what extent some of the oddities could be avoided if one were to drop one of the two principles but maintain the other.
2. PROBLEMS FOR KA
2.1 Acting on Unknown Propositions
It is easy to generate cases where a subject seems to do something inappropriate if she treats p as her reason to act without knowing p. Arguably, in many situations knowledge is required. But if K-NEC is true, knowledge is required in all settings, not just in many settings. Here I argue that, contrary to K-NEC, we sometimes act appropriately on a proposition we are justified in believing, even when we lack knowledge.
Motivation for K-NEC typically involves cases where agents face some ‘high stakes’ decision or other. But if K-NEC is true, knowledge is required in low stakes settings as well as in high. Therefore, the most apt test cases for determining whether knowledge is always required or only often required will be cases where very little is at stake. But when the stakes are low it is far from obvious that knowledge, rather than some weaker level of epistemic warrant, is required to meet one's epistemic obligations with respect to reasons for action. Consider the following scenario:
RESTAURANT: Ana is on her way to a restaurant to meet some friends. She is not in a hurry and nothing depends on her arriving at a particular time. She comes to a fork in the road and thinks the restaurant is to the left (hereafter LEFT). In fact, she is correct. She doesn't know LEFT, but is justified in believing LEFT. Her justification is close to, but fails to be knowledge-level.
I suggest that Ana acts appropriately when she treats the restaurant is to the left as her reason to go left. She is in no way subject to criticism for treating LEFT as her reason for going left. But if it is appropriate for Ana to treat LEFT as her reason to go left, then K-NEC is false because Ana does not know LEFT.
John Hawthorne and Jason Stanley (Reference Hawthorne and Stanley2008) disagree with this analysis. They discuss a similar scenario – one in which a subject has a partial belief that p and does what is rational to do in the subject's situation (as given by expected utility theory). Hawthorne and Stanley draw a distinction between what is rational to do and what is appropriate to treat as one's reason for action. Their view delivers the result that in RESTAURANT it is rational for Ana to go left, but inappropriate for her to treat LEFT as her reason to go left.
This raises the question: what is the subject appropriate to treat as reason in a case of this sort? One suggestion they make is that it can be appropriate to treat knowledge of the chance that p as one's reason. “Chance,” in this context, expresses an epistemic probability, which Hawthorne and Stanley understand as the probability a proposition has on one's total body of knowledge. The idea is that when p is unknown, it is appropriate for S to treat the chance that p as a reason to act – though only if S knows the epistemic probability for her that p. In the case at hand, if Ana knows that probably the restaurant is on the left and treats this proposition as her reason for action, she does not violate KA.
I don't disagree that we sometimes appropriately treat propositions about chances as reasons to act, but this strategy won't explain every case of this kind. Our epistemic chances are not luminous – that is, we are not always in a position to know what our epistemic chances are. Consider this modification of the original case: Ana is driving to the restaurant, and she believes there is a .80 chance that the restaurant is on the left. We can imagine that the epistemic chance that the restaurant is on the left for Ana is .80, but that she fails to know that it is .80. (Suppose she could easily have believed it was .81 or .79.) K-NEC delivers the result that Ana does something inappropriate and is criticizable if she treats there is a .80 chance the restaurant is on the left as her reason to go left. But now it looks like Ana has a strong epistemic position for the restaurant is on the left yet, according to KA, she is incorrect to treat LEFT as her reason and is also incorrect to treat there's a .80 chance the restaurant is on the left as her reason. Knowledge of chances does not provide an explanation of what one is appropriate to treat as one's reason in settings where one does not know p and does not know one's epistemic chance that p.
An advocate of KA might respond that our knowledge of chances is rarely, if ever, so specific. We typically operate with general categories of chance: unlikely, more likely than not, likely, and so on. But this is beside the point. Change the example so that it is very likely, Ana believes it's very likely, but she doesn't know it is very likely and the result is the same: she acts on a belief that fails to be knowledge, yet it is not correct to criticize her.
Moreover, even if our epistemic chances were luminous, knowledge of chances would not provide a satisfactory response to the original objection. The problem in RESTAURANT is not that there is no proposition in the vicinity that Ana could have appropriately treated as her reason to go left; the point is that when Ana in fact treats LEFT as her reason to go left when she doesn't know LEFT, it seems inappropriate to criticize her. K-NEC delivers the counterintuitive result that she violates a norm, and thus is criticizable.Footnote 5 Knowledge of chances alternatives distract from this key point.
A second idea one might find tempting to use in reply to the kind of objection raised by RESTAURANT is as follows: Ana's action involves an excusable violation of the norm. An action can be excused when a subject violates a norm but doesn't know that she violates the norm; instead, she believes that she conforms to the norm.Footnote 6 Advocates of KA might excuse Ana for treating LEFT as her reason for going left in a case where she doesn't know that she lacks knowledge that the restaurant is on the left. If Ana believes that she knows LEFT and is thereby excused for violating K-NEC, then KA advocates can easily explain why it is inappropriate to criticize her.
The problem with this suggestion is that RESTAURANT is not a case where an excuse is needed. When Ana arrives at the restaurant, if her friends were to criticize her for treating LEFT as her reason to go left when she didn't know LEFT but justifiedly believed LEFT, she would be right to reject the criticism, but not because she believed that she knew LEFT. She can reject the criticism on the basis that LEFT was appropriate for her to treat as her reason: she doesn't need an excuse. If K-NEC were correct, we would expect her to accept their evaluation of her and then appeal to her belief that she knew LEFT as her excuse for violating the norm; but this is not what we find. Instead, she can defend the action itself.
Furthermore, if we add the assumption that Ana knows that she doesn't know LEFT, she cannot be taken to excusably violate K-NEC. Suppose Ana knows that her memory isn't good enough to constitute knowledge in this instance – she has made mistakes of this type in the past. Nevertheless, she has strong reason to believe the restaurant is to the left, and in fact treats this proposition as her reason to go left. Her action is clearly not in accordance with K-NEC and she does not have an excuse, yet I suggest that she does nothing epistemically criticizable. My suggestion, once again, is that the reason Ana doesn't deserve criticism is that she doesn't violate an epistemic norm.
The judgment that Ana acts appropriately when she treats LEFT as her reason in RESTAURANT strikes me as highly plausible, but not a conclusive case against K-NEC. Next we will look at what happens when we add K-SUFF to K-NEC. Some of the following concerns apply to K-SUFF alone. I will indicate, as needed, when an objection does not depend on K-NEC.
2.2 Knowledge and Stability
One result of KA combined with some plausible judgments about cases is that whether you know p can change with what is at stake. This makes knowledge unstable, and, specifically, less stable than we intuitively and traditionally expect it to be. Although some philosophers have accepted the idea that knowledge comes and goes as what is at stake changes, it has gone without recognition that according to KA, stakes are not the only pragmatic factor that will make knowledge come and go with ease. On the assumption that practical factors make an epistemic difference, we would expect the following principle to be true:
Modest stability: practical factors make a difference to what we know only through changes in stakes.
Modest stability is extremely plausible: if two subjects have the same stakes, and have the same strength of epistemic position with respect to p, it is strongly intuitive that they are in the same position to know p. But the view under consideration must reject modest stability. Consider the following scenario where a non-truth relevant and non-stakes relevant factor makes for the presence and absence of knowledge, if KA is assumed:
NEEDLE: Tiffany, Julie, and Jonny are hiking the Grand Canyon. Jonny falls and needs stitches. Luckily, Tiffany is a doctor and carries a needle in her backpack. But as she pulls the needle out, she rips the packaging ever so slightly. It is possible, though extremely unlikely, that the needle is contaminated. Despite the little tear, Tiffany thinks to herself, ‘it's clean; I'll use it.’ In fact, the needle is clean. Then Julie remembers that she may have a needle in her backpack. If its packaging is not ripped, it would be better to use that needle. Tiffany waits for Julie to look. False alarm – Julie has no needle. Tiffany continues with the procedure.
In this scenario, the stakes do not change – throughout the case the stakes are high: Jonny is in urgent (though not life-threatening) need of stitches.Footnote 7
Here are two intuitive judgments about the case: Tiffany initially knows the needle is clean (hereafter, CLEAN); and Tiffany is not appropriate to treat CLEAN as a reason once Julie suggests she look in her backpack for another needle.Footnote 8 Holding these judgments fixed the advocate of KA seems to be committed to the following analysis: before Julie indicates that she might have a needle, Tiffany knows CLEAN and thus it is appropriate for Tiffany to treat CLEAN as a reason to act, by K-SUFF. But once Julie suggests she might have a needle in her purse, Tiffany should wait. She shouldn't treat CLEAN as her reason to act. According to K-SUFF, if it is inappropriate for Tiffany to treat CLEAN as a reason to act, then she doesn't know CLEAN.Footnote 9 When Julie confirms that she does not have a needle, Tiffany can treat CLEAN as a reason to act and thus (by K-NEC) she knows CLEAN once again.
This result is extremely odd. Surely whether Julie has a needle in her backpack is irrelevant to whether Tiffany knows that the needle in her hands is clean. Although some find it plausible that stakes can make a difference to what you know, KA delivers the strongly counterintuitive result that too many factors can make a difference. If KA and our intuitive judgments about the case are true, knowledge comes and goes with more ease than has been recognized: if KA and these judgments are true, knowledge is radically unstable.
Although the instability in this case is raised as a problem for KA, dropping K-NEC does not remove the instability. On the assumption that Tiffany knows at the start of the case, K-SUFF combined with the claim that she shouldn't treat CLEAN as her reason to act while Julie looks for the needle results in loss of knowledge and the rejection of modest stability. Furthermore, although it is open to the advocate of K-SUFF to deny that Tifffany knows CLEAN once it is revealed that Julie does not have a needle, this results in an odd asymmetry between the start and end of the case.
I suggest that the best non-skeptical way of thinking about the case is to reject K-SUFF and affirm the intuitive judgments about when Tiffany is appropriate to treat CLEAN as her reason to act. On my picture, Tiffany knows CLEAN throughout the case, but she shouldn't treat CLEAN as her reason to act when Julie mentions she might have a different needle. Whether she is appropriate to treat CLEAN as her reason changes, but her knowledge is stable.
2.3 Double-Checking
As support for their position, advocates of KA often rely on intuitions regarding pairs of cases where two subjects are in the same epistemic position with respect to a proposition, but differ with respect to stakes – that is, the cost of being wrong is different for each.Footnote 10 It is often suggested that the subject in high stakes should double-check whether p prior to acting, but the subject in low stakes need not double-check. The apparent need to double-check in high stakes, but not in low stakes, is taken to be an indication that the epistemic position of the subject in high stakes is not strong enough for knowledge, though the same strength of epistemic position is strong enough for the subject in low stakes to know. For example:
TRAIN-1: Matt is a tourist leisurely making his way to Foxboro. He is about to board a train that stops in Foxboro and continues to Providence. It doesn't make a difference to him if he ends up in Providence. He looks at his train schedule, confirms that this is the right train, and boards. He knows the train stops in Foxboro (hereafter FOXBORO); he doesn't need to double-check.Footnote 11
TRAIN-2: Jeremy is at the train station about to board the same train. He is in the same epistemic position as Matt, but it is very important to Jeremy that the train stops in Foxboro. Jeremy should double-check.
According to KA, since Jeremy shouldn't treat FOXBORO as his reason to board, he doesn't know the train stops in Foxboro.
In these types of cases, it is usually assumed that double-checking is fast, easy, free, and available. But consider a case where double-checking is costly, in one of these ways:
TRAIN-3: Jeremy is in the same high-stakes scenario and epistemic position as TRAIN-2. But there is no one around – he has no way of double-checking. He should go ahead and board.
TRAIN-4: Jeremy is in the same high-stakes scenario and epistemic position as TRAIN-2. His only option of double-checking is to ask the train attendant, who is going to charge him $500 for the information.
Suppose that in TRAIN-4 the cost of double-checking is higher than the cost of getting on the wrong train. Jeremy should board without double-checking. Thus, in both TRAIN-3 and TRAIN-4 Jeremy should go ahead and board. Does Jeremy know that the train stops in Foxboro? Can he treat the train stops in Foxboro as his reason to board? It's not clear what verdict KA delivers here.
The most straightforward answer seems to be that Jeremy knows in both TRAIN-3 and TRAIN-4. After all, Matt knows the train stops in Foxboro, and Matt has the same strength of epistemic position for the proposition as Jeremy. Nothing epistemic stands in the way of Jeremy treating the train stops in Foxboro as a reason to board in TRAIN-3 and TRAIN-4. Thus, in TRAIN-3 and TRAIN-4 Jeremy passes the test that Fantl and McGrath offer to determine, in any practical setting, when weakness in your epistemic position for some proposition stands in the way of action. The test works as follows: if strengthening your epistemic position makes a difference to whether you ought to perform an action, something epistemic stands in your way. (On their account, when weakness of epistemic position for p stands in one's way, one is not appropriate to treat p as one's reason to act, and thus one does not know p.)
In TRAIN-2, Jeremy fails the test: conditional on having some stronger epistemic position for FOXBORO, Jeremy ought to board. Weakness in his epistemic position stands in the way of boarding in TRAIN-2. But in TRAIN-3 and TRAIN-4, Jeremy passes the test. So the explanation for why Jeremy shouldn't treat FOXBORO as his reason in TRAIN-2 is not available in TRAIN-3 and TRAIN-4. But if he can treat FOXBORO as his reason to board, by K-NEC, he knows FOXBORO. This is a strange result – since according to the KA advocate, Jeremy doesn't know in TRAIN-2, it looks as though he gains knowledge by not having the option to double-check, or by the station employee's attempt to over-charge him for information. Intuitively, whether you know does not depend on such factors.Footnote 12
One story the proponent of KA might be tempted to offer is that the absence of knowledge in TRAIN-3 and TRAIN-4 is a direct result of Jeremy's high stakes situation. But it should be obvious that this kind of strategy is not very promising: a blanket denial that we have knowledge in any high-stakes scenario will result in semi-skepticism. KA is usually advanced as part of an anti-skeptical epistemology. After all, in the usual high-stakes bank case, where one allegedly loses knowledge that the bank is open Saturday, we do not expect one to also lose knowledge that the bank is open on Friday. On the contrary, it tends to be assumed that the subject knows the bank is open on Friday. A direct link between high stakes and ignorance would have the result that we know very little in any high-stakes settings. The non-skeptical KA advocate must embrace the idea that Jeremy knows in TRAIN-3 and TRAIN-4 or she must provide an explanation for why Jeremy fails to know.
NEEDLE and the train cases make clear that according to the view under consideration the practical factors that make knowledge come and go are not limited to changes in a subject's stakes. Seemingly irrelevant changes in the environment – such as the contents of a friend's bag – can affect whether one has knowledge, on this view: the availability of alternative options and the cost of double-checking are just two examples of the many factors that make knowledge come and go. Although this position is usually presented as accommodating sensitivity to stakes, stake changes are just one kind of non-epistemic factor that can make for the presence or absence of knowledge on this picture.
2.4 Reversing the Stakes
Cases that involve a change in stakes almost always move from low-stakes to high-stakes. First we imagine a subject in low-stakes who has knowledge. Then we are asked to consider that her stakes have been raised. The expectation is that we will be inclined to think that she loses knowledge.Footnote 13 Without affirming these intuitions, I will merely point out that if we reverse the order of the stakes, the intuitive pull is much weaker.
Reverse BANK: It is Friday afternoon and Keith is on his way to the bank to deposit a check. Although he was at the bank two Saturdays ago, it is very important to him that his check be deposited by Monday and he reasons that there is a small chance that the bank has changed its hours. While on his way to the bank, he receives a message from his wife saying that it is no longer important that the check be deposited by Monday.
KA predicts that when his stakes are lowered Keith gains knowledge because he is then appropriate to treat that the bank is open on Saturday as his reason to go to the bank on Saturday. But it is implausible that knowledge is gained by lowering one's stakes. Yet if KA is true, knowledge is gained in this manner. This reveals an odd asymmetry: pragmatic encroachment strikes many as plausible only when practical factors are knowledge depriving. Indeed, most of the cases offered in support of pragmatic encroachment are cases where subjects lose knowledge when their circumstances change and the cost of being wrong increases. It is a strain to think that lowering one's stakes is knowledge inducing.Footnote 14 But the proponent of KA is committed to this counterintuitive result.
Moreover, the asymmetry calls out for explanation: what explains why is it relatively easy to get many into the frame of mind where, due to practical factors, the judgment that knowledge is lost sounds plausible, when at the same time it seems absurd to think that changes in pragmatic factors of one's situation could result in gaining knowledge? As the motivation for pragmatic encroachment relies heavily on our finding shifts in knowledge between high and low stakes cases plausible, this asymmetry indicates that pragmatic encroachment does not have nearly the intuitive advantage over non-shifty views that proponents of this picture suggest. The support that our intuitive judgments about cases give to this view is limited to a far more narrow sample of cases than is often supposed.
2.5 Multiple Actions
It has gone largely unappreciated that we are often in a position to undertake multiple actions at the same time, on the basis of the same reason.Footnote 15 Consider the following case:
Dinner
Alli tells her husband Tim that she is going to a coffee shop for the evening and won't be home until late. On the basis of her testimony, Tim considers two actions he might do: first, make pizza for dinner – Alli doesn't like pizza, so Tim only has pizza when she's not home; and second, invite his brother for dinner. Tim's brother recently had a huge disagreement with Alli and Alli made it very clear that she didn't want to see Tim's brother for a while. Tim decides to make pizza but not invite his brother over.
It seems that Tim acts correctly when he decides to do one action but not the other. Before he invites his brother over, he should call and double-check that Alli will be out late. Tim's epistemic position with respect to the proposition that Alli is coming home late (hereafter HOMELATE) is strong enough to justify him in making pizza, but not strong enough to justify inviting his brother over.Footnote 16 Thus, it is appropriate for Tim to treat HOMELATE as his reason to make pizza, but it is not appropriate for him to treat HOMELATE as his reason to invite his brother over. This result is irreconcilable with KA. According to K-NEC, if it is appropriate for Tim to treat HOMELATE as his reason to make pizza, he knows HOMELATE. But by K-SUFF, if Tim can't appropriately treat HOMELATE as his reason to invite his brother over, he doesn't know HOMELATE. The consequence is that according to KA, it seems Tim both knows HOMELATE and does not know HOMELATE. On pain of contradiction, the KA advocate must decide between the following: either Tim knows HOMELATE and is appropriate to treat HOMELATE as his reason to make pizza and to invite his brother over, or he doesn't know HOMELATE and should not treat HOMELATE as his reason to do either action. But this is implausible. Surely he can appropriately treat HOMELATE as a reason for one action but not the other. KA cannot account for the plausibility of acting appropriately on a proposition with respect to one action, when simultaneously it would be inappropriate to treat that proposition as a reason for another action.Footnote 17
3. CONCLUSION
There is reason to think that knowledge is neither necessary nor is it sufficient for rational action. Of course, this does not imply that there is no intimate tie between knowledge and reasons for action. Most of the time, when you know a proposition it is appropriate for you to treat that proposition as a reason for acting, and most of the time you ought to know a proposition in order to appropriately treat that proposition as your reason for acting. If knowledge is usually, but not always, the standard by which we are appropriate to act, a principled explanation of our folk appraisals of reasons for action can draw on this generic insight, without commitment to the difficulties raised in this paper. The result is that there is still a tie between knowledge and reasons for action; the tie just isn't as intimate as we might have hoped.Footnote 18
