Rational Belief and Fundamental Epistemic Value: What s the Connection?

Transcription

1 Rational Belief and Fundamental Epistemic Value: What s the Connection? Abstract Rational beliefs seem to have epistemic value even in worlds where rational beliefforming processes are unreliable. If this is true, can we explain why it is true? It is easy to see that we cannot if we buy the common assumption that all derivative value must be grounded via instrumental relations to fundamental value ( Instrumentalism ). But we should reject this assumption. Instrumentalism yields a generally flawed picture of derivative epistemic value. Moreover, it is widely rejected by value theorists outside of epistemology. By rejecting Instrumentalism in ways that nonepistemic value theorists have already explored, we can provide a unified and truthoriented picture of derivative epistemic value that captures rational belief as a special case. There are important upshots for epistemology. Many think that reliabilists have a unique ability to explain why we value justified belief. But it is actually harder for reliabilists to explain the value of justified belief if they accept Instrumentalism. And once we see that truth-connectedness need not be understood instrumentally, we can also see how internalists who equate the justified and the rational might preserve a truth-oriented account of epistemic value. 1 Introduction Consider a familiar pair of subjects. Alpha inhabits a world like ours. She knows just about everything there is to know about this world, and her beliefs and reasoning are flawless in all other epistemologically interesting respects. Now compare Beta. Beta s mental life is internally just like Alpha s. Any experience or apparent memory that Alpha has is equally had by Beta. But Beta inhabits a world whose only other occupant is a Cartesian demon, and this demon ensures that Beta s beliefs about extra-mental reality are radically false. While it is uncontroversial among non-skeptical epistemologists that Alpha is better off epistemically than Beta, most epistemologists will agree that there is some positive epistemic property that Beta s beliefs retain. Different epistemologists just have different labels for it. Some externalists will deny that Beta s beliefs about extra-mental reality are as justified as Alpha s. But they will have some other term of appraisal, like rational or blameless. 1 My term is rational, and I will call Beta s rational beliefs about extra-mental reality Beta s relevant beliefs to save words. In this paper, I am interested in the value that rational beliefs have from a purely epistemic point of view. It is plausible that even in radical skeptical scenarios (e.g., demon 1 See Littlejohn (2012) for the first option. See Littlejohn (2009) and Pritchard (2012) for the second option. Others have proposed that while Beta s beliefs are unjustified, Beta herself has some good epistemic property. This leads some to distinguish between doxastic and personal justification; see Bach (1986) and Engel (1992). 1

2 worlds), rational beliefs about extra-mental reality have some value from this point of view. For even in these worlds, it is natural to think that one is better off from an epistemic point of view if one believes rationally rather than irrationally. Call this the Value Datum. It would be nice if we could explain the Value Datum. It is not a brute fact, one might think. But two common assumptions make it hard to explain. To see the first, note that many epistemologists think there is a fundamental epistemic value (FEV). The usual candidates are true belief and knowledge. Many epistemologists think things like rational belief and justified belief are only derivatively significant relative to FEV. Hence the first assumption: Derivativeness: Rational beliefs are only derivatively epistemically valuable. There is a second common assumption often conflated with the first. Epistemologists tend to assume that derivative epistemic values only have epistemic value by standing in instrumental relations of two sorts: (i) by being things that tend to produce FEV (call this production value ), 2 or (ii) by being products of sources that tend to produce FEV (call this product value ). Thus, the second assumption: Instrumentalism: Derivative value is value of the production or the product kind. Given these assumptions, it is unclear how to explain the Value Datum. After all, in extreme skeptical scenarios, believing rationally will tend to thwart the goal of producing a high ratio of true to false beliefs (or of knowledge to ignorance). The only obvious strategies are to (A) hold that rational belief is a further FEV or (B) reject Instrumentalism. Several epistemologists pursue option (A). 3 But I would prefer to avoid (A) unless we have no other defensible option. I think it is clear that we care about rationality because we care about something else, like true belief or knowledge. I am not happy to sacrifice this intuition without considering the alternatives. Luckily, (B) is a defensible alternative. Although epistemologists have so far ignored this option, the rationale for it is simple. Instrumentalism yields a generally flawed picture of derivative epistemic value. Moreover, it is widely rejected by value theorists outside of epistemology, including some who are consequentialists about rightness (e.g., Thomas Hurka). 4 By rejecting Instrumentalism in ways that value theorists have already explored, we can provide the very same explanation of the value of rational belief that we should provide of other good kinds of belief, including justified belief of the externalist sort. If I am right, there will be major upshots for epistemology. It is often thought that reliabilists have a unique ability to explain why we value justified belief. This is supposed to owe to the instrumental link reliabilists erect between justified belief and true belief. 2 I don t use produce in a purely causal sense, since I am including constitutive means to FEV as producers. So things that tend to constitute the FEV count as having production value in my broad sense. Cf. David (2001). 3 See, e.g., Brogaard (2009), DePaul (2001), Kvanvig (2005), and Zagzebski (2004). 4 While Hurka s axiology will be my model, other axiologists who reject Instrumentalism include Anderson (1993), Herman (1993), Kagan (1999), Korsgaard (1983), Moore (1903), Parfit (2011), Rabinowicz and Ronnow- Rasmussen (1999), Rashdall (1907), Scanlon (1998), and Zimmerman (2001). 2

3 But as it turns out, Instrumentalism makes it harder to explain the value of justified belief even if justified belief is connected to true belief as reliabilists suggest. And if we reject Instrumentalism, we are left with no reason to think reliabilism has this advantage and some reason to think internalists who equate the justified and the rational can secure a truthoriented account of epistemic value. This undermines a major argument for reliabilism (though one might embrace reliabilism about justified belief for other reasons). With these claims in mind, here is the plan. In 2, I will explain further why it is hard to give a plausible Instrumentalist account of the value of rational belief, considering and rejecting two sophisticated stories. In 3, I will argue that Instrumentalism is generally illequipped to explain derivative epistemic value. I will explain at the same time why option (A) is unpromising: assuming Instrumentalism, the argument for adding rationality to the list of FEVs is equally an argument for inflating the list in ways no one should tolerate. I will then show in 4 how an alternative picture of derivative epistemic value allows us to provide the same explanation of the value of rational belief that we should provide of other epistemically important kinds of belief, including justified belief of the externalist sort. I will conclude in 5 by discussing the upshots for epistemology. Disclaimers. Before I proceed, a few disclaimers are in order. I will not be arguing that no derivative epistemic value is instrumentally grounded. Rejecting Instrumentalism is compatible with affirming that some derivative value is instrumentally grounded. Furthermore, rejecting Instrumentalism is compatible with the idea that some items might have both instrumental epistemic value and a different kind of derivative epistemic value. Truth-conducivity of a reliabilist kind, I agree, is a necessary condition for some epistemic virtues. I only deny that Instrumentalism can explain all derivative epistemic value. I will sketch a different model of epistemic value derivation. But I intend this model to supplement the instrumental model, not to supplant it. I will also stress that my focus is on epistemic value. I am not here defending a nonconsequentialist view about epistemic rightness. 5 Indeed, my inspiration in ethics is Hurka (2001) s recursive axiology, which was designed to provide a way for consequentialists about rightness to consistently uphold the non-instrumental value of virtue. 2 Instrumentalism and the Value Datum It might seem obvious that there is no Instrumentalist account of the epistemic value had by Beta s relevant beliefs. But this claim deserves argument. Here, I take it, is the most obvious argument: The Simple Argument 1. If Instrumentalism and Derivativeness are true, rational beliefs have epistemic value in a world w only if they tend to promote fundamental epistemic value in w or are products of types of sources that tend to do so in w. 5 For arguments against this view, see Littlejohn (2012) and Berker (2013a) and (2013b). 3

4 2. Beta s relevant beliefs do not tend to promote fundamental epistemic value in Beta s world and are not produced by belief-forming dispositions that tend to do so in Beta s world. 3. So, if Instrumentalism and Derivativeness are true, Beta s relevant beliefs lack epistemic value in Beta s world. Since the consequent of (3) entails the negation of the Value Datum, (3) amounts to: 3*. If Instrumentalism and Derivativeness are true, we should reject the Value Datum. This argument is valid. So if the argument fails, either (1) or (2) must be false. If one is both an Instrumentalist and a fan of Derivativeness, what can one say against (1) and (2)? Well, the following sophisticated pictures conflict with (1): A. Counterfactual Instrumentalism. Sometimes what makes a belief-forming disposition instrumentally epistemically good is that it would tend to produce fundamental epistemic value in favorable conditions. But Beta s belief-forming dispositions are epistemically good in that way and Beta s relevant beliefs are also epistemically good by manifesting those dispositions. B. Ingredient Instrumentalism. Even if X is not sufficient for promoting fundamental value, X can remain instrumentally good in an extended sense. Necessary ingredients for fundamental values are instrumentally good in a natural sense. But rationality is necessary for knowledge, and knowledge is a candidate for the FEV. One might complain that I have left out a third strategy. While there is no necessary connection between believing rationally and actually maximizing accuracy, one might hold that believing rationally is necessarily apt to maximize expected accuracy. Believing rationally might necessarily be apt to maximize expected accuracy. But this does not help to answer our question. Our question is about the value that rational beliefs have in skeptical scenarios, not about the value that they merely appear or can be expected to have from the believer s point of view. Expected value is perhaps relevant to what we ought subjectively to believe. But we are not interested here in whether what it is rational to believe is also what we ought subjectively to believe, but rather in whether there is anything epistemically good about believing in this way. Perhaps one will reply that it is obvious that Beta s relevant beliefs only have expected epistemic value. But this is implausible. We can look squarely at the fact that Beta s relevant belief-forming processes are unreliable and still find it plausible that there is epistemic value in the output beliefs. There is more epistemic value in these beliefs than there would be in the irrational doxastic attitudes that Beta might instead have formed. That fact cries out for an explanation. The appeal to expected value fails to provide one. At best it provides an error theory: we imagine ourselves into Beta s shoes and get muddled in our evaluation. But we are not guilty of that confusion. 4

5 2.1 The Failure of Counterfactual Instrumentalism Framed properly, Counterfactual Instrumentalism can look compelling. Consider Ernest Sosa s view about subjects like Beta. 6 For Sosa, the belief-forming dispositions that constitute epistemic rationality are competences. Competence here is broadly reliabilist: what makes a belief-forming disposition a competence is some instrumental relation to true belief. But Sosa s view differs importantly from other versions of reliabilism: by using competence as the core notion, Sosa can exploit the competence/performance distinction. Competences are dispositions. Like other dispositions, there is an understood set of favorable conditions for their manifestation. If you take a box of matches into a room without oxygen and strike them, none will light. That does not show that none is flammable. If you take an archer into a hurricane, she will not hit any targets. That does not show that she is an incompetent archer. The reliability of competence is relative to favorable conditions. Sosa invokes this point, suggesting that worlds like Beta s only illustrate that reliability of this sort is necessary for rationality. One might find this picture of the nature of rationality extensionally plausible but wonder how it explains why Beta s relevant beliefs have value from the epistemic point of view. Sosa s answer appeals to an analogy with a temperature control device: Suppose it is taken off the shelf in the display room for a demonstration, and a situation is simulated wherein it should activate the cooling trigger, and then a second situation is simulated wherein it should activate the warming trigger. In such a test the device might either perform well or not. But the quality of its performance is not to be assessed through how well it actually brings about the goods that it is meant to bring about in its normal operation. For in the display room it brings about neither the cooling nor the heating of any space... What we are doing is quite obvious: we are assessing whether it performs in ways that would enable it to bring about the expected goods once it was properly installed... 7 Sosa s analogy suggests that something can be derivatively good in a way worth calling instrumental in some situation even if the situation is not one where it is conducive to a more fundamental good. One might think that this yields a satisfactory Instrumentalist explanation of why Beta s relevant beliefs retain epistemic value. But this conclusion is incautious. The analogy with the temperature control device only works in mild deception examples. 8 Mild examples feature people in worlds that are otherwise favorable who become victims of deception e.g., people like us whose brains get envatted by evil neuroscientists. In the worlds of these examples, unreliability is not the norm. Here the temperature control device yields a helpful analogy. But the analogy ceases to be helpful when we consider extreme deception examples, where deception is the norm in the world and nearby worlds. Beta s world was supposed 6 See Sosa (1991, 1992, 2003) for this, and Sosa (2007, 2010) for discussions of competence. 7 Sosa (2003: 175 6). 8 Foley (2004) argues on similar grounds that Sosa s account of the nature of rationality is defective. I am arguing that Sosa s analogy fails to explain the epistemic value of Beta s relevant beliefs. 5

6 to be such a case. We could even imagine that there is a law of nature which ensures that Beta s relevant beliefs will be false: maybe our demon is the god who created this world and set up the laws of nature in this way. Sosa s strategy suggests that when we regard Beta s relevant beliefs as epistemically good, what we find good is that these beliefs are formed in ways that would be reliable if Beta were in a completely different, distant world. The analogy does not support this conclusion. Imagine an intrinsic replica of the temperature control device in a distant world with such different laws of nature that it could never reliably control the temperature of anything there. Imagine the same holds for all nearby worlds. Should we say that the device still exhibits a good feature in the world it inhabits because if it were suitably installed in a completely different, distant world it would reliably control the temperature there? This is implausible. If there are no conditions in w or nearby worlds in which a mechanism would reliably bring about success, it is no good feature of that mechanism in w that it would bring about such success in a distant world w*. It would be a good feature in w*. But we are asking what is good in w. The analogy with the temperature control device makes it harder to understand how the Value Datum could be true, not easier! This is not to deny that a belief-forming disposition is rational in a world like Beta s only if there is another favorable world where it is reliable. It is not to deny that being formed by dispositions that would be reliable in the actual world is necessary for a belief to be rational. It is just to deny that these necessary conditions explain the epistemic value of rational beliefs about extra-mental reality in bad worlds. The analogy itself suggests that they do not. There might be a helpful analogy between the temperature control device and people from an otherwise good world who become deception victims. But there is no good analogy between the temperature control device and Beta, who inhabits a distant, exceedingly bad world surrounded by similar worlds. 2.2 The Failure of Ingredient Instrumentalism The foregoing strategy is the obvious one for Instrumentalists to pursue if true belief is the FEV. But even if Instrumentalists accept Derivativeness, they might deny that true belief is the FEV. Knowledge is a contender after Williamson (2000). If knowledge is the FEV, Ingredient Instrumentalism is a potential strategy for explaining the Value Datum. 9 The plausibility of this explanation depends on the status of this argument: The Ingredients Argument a. Necessary conditions for more basic goods retain instrumental value qua ingredients for these goods even when the other necessary conditions are missing. b. If (a) is true, rational beliefs can retain instrumental epistemic value as ingredients for knowledge even when other necessary conditions (e.g., truth) are missing. 9 Ingredient does not mean independent factor, so this idea is consistent with Williamson s attack on composite analyses of knowledge. Ingredients for a dish can be blended together by cooking so that none can be factored out of the cooked dish. (In any case, I just use the term for vividness.) 6

7 c. If the consequent of (b) is true, rational beliefs about extra-mental reality have instrumental epistemic value even in bad worlds like Beta s. d. So, Beta s relevant beliefs have epistemic value in Beta s world. If successful, the Ingredients Argument undermines (1) in the Simple Argument: while Beta s relevant beliefs never yield knowledge and are not products of knowledge-conducive dispositions, they retain a kind of instrumental value anyway. But the Ingredients Argument does not really explain why Beta s relevant beliefs retain epistemic value. The flaw in the argument resembles the flaw in the attempt to support Counterfactual Instrumentalism by analogy with the temperature control device. To see this, consider when (a) looks plausible. (a) looks plausible when we imagine a world where the other ingredients stand a chance of existing. Imagine one has half the ingredients for a great dish. It is plausible that even if one is missing the other ingredients, one s ingredients retain instrumental value. This is because the other necessary ingredients stand a chance of existing in worlds like one s own (and presumably do exist). Suppose, by contrast, your friend Bob says of some stuff in the trash: If we were just to have a few further ingredients, we could use this stuff to make the best thing ever. Let s take it! You reply: The other ingredients can t even exist in this world! If Bob said Let s still take this stuff. It still has value, that would be absurd. There is a similar problem with using the Ingredients Argument to explain the value of Beta s relevant beliefs. By stipulation, Beta s relevant beliefs cannot be converted into knowledge: they are products of processes that the demon guarantees to be unreliable. By stipulation, this is true of nearby possible worlds. From an Instrumentalist point of view, Beta s relevant beliefs are like the stuff Bob wants to save from the dumpster. So, the Ingredients Argument fails. Ingredients for goods do retain instrumental value in worlds where the other ingredients stand a chance of existing. But this does not support the claim that Beta s rational beliefs retain instrumental value as ingredients for knowledge in her world. In such worlds, the other ingredients stand no chance of existing. Of course, Beta will have other rational beliefs that constitute knowledge (e.g., about her internal mental life). If so, there is a generic sense in which the other necessary ingredients for knowledge stand a chance of existing in her world. But this makes no difference to the argument by analogy. Suppose Bob and I know that although other stuff like the stuff we are discussing can be combined with stuff in our world to make something good, the particular stuff we are discussing cannot. It would be no less absurd for Bob to insist that we take the trash in this revised case. 2.3 Taking Stock Let s take stock. The core problem for Instrumentalists was simple. There are only two general Instrumentalist strategies that could explain why rational beliefs have epistemic value: 7

8 The Production Strategy. Rational beliefs have derivative epistemic value by tending to be instrumental to FEVs like true belief or knowledge. The Product Strategy. Rational beliefs have derivative epistemic values as products of types of sources that are instrumental to FEV. Beta s case makes both strategies look implausible. In Beta s world and nearby worlds, rational beliefs are guaranteed to be reliably false and rational belief-forming dispositions are guaranteed not to be conducive to FEV. We found it unhelpful to reply that Beta s rational beliefs are conducive to expected epistemic value. We also found it unhelpful to reply that Beta s relevant belief-forming dispositions would be conducive to FEV if Beta were in another world. This helps only if the analogy with the temperature control device usefully extends to extreme skeptical scenarios, not just mild ones. It does not. Since the Ingredients Argument also failed, it is hard to see how to resist the Simple Argument. So, it seems the appearance is not misleading: Instrumentalism and Derivativeness are inconsistent with the Value Datum. Since the Value Datum is what we want to explain, we should at least reconsider Instrumentalism. Perhaps a yet more sophisticated Instrumentalist story can be concocted, but we have good reason to consider the alternatives. 3 A General Problem for Instrumentalism But why think we should abandon Instrumentalism in order to explain the Value Datum rather than simply abandoning the Value Datum or Derivativeness? The first part of the answer is that Instrumentalism leaves us generally ill-equipped to explain derivative epistemic value. Indeed, even if rational belief were instrumentally linked to FEV as reliabilists suppose, we still couldn t explain the value of rational belief. Of course, we could go pluralist about fundamental epistemic value instead. But I will show and this is the second part of the answer that the motivations for going pluralist here overgeneralize, and support going pluralist about properties that have a paradigmatically derivative kind of value. 3.1 Swamping Problems Why does Instrumentalism leave us generally ill-equipped to explain derivative epistemic value? Because Instrumentalism is responsible for a generalized version of the swamping problem. Originally, the swamping problem was presented as a problem for a simple kind of reliabilism, owing to an argument by analogy for: (A) A belief s having been produced by a reliable belief-forming process does not as such make that belief epistemically better if that belief is already true. 8

9 (A) is made plausible by an analogy from Zagzebski (1999). The mere fact that some good coffee came from a reliable coffeemaker does not make that coffee better. But a reliably produced true belief is like a reliably produced cup of coffee. If so, then: (B) If knowledge = true belief produced by a reliable belief-forming process, knowledge is not as such epistemically better than true belief. But knowledge is as such epistemically better than true belief. Hence the problem. As people have increasingly recognized, 10 the swamping problem runs much deeper than this. It is not just a problem about knowledge or just a problem for reliabilists. To see the first point, note that simple reliabilists will identify a belief s being justified with its being formed by a reliable belief-forming process. Assume they are right for the sake of argument. We can use the same coffee analogy to argue that justification cannot as such add epistemic value to true belief. That is bad: justification as such does make a true belief epistemically better! This is not a restatement of the problem about knowledge. Thanks to Gettier, no reliabilist will equate knowledge with justified true belief. There is an even more general structure that makes the problem not just of limited interest to reliabilists. On any view on which justifiedness : true belief :: being made by a good coffeemaker : good coffee there is a worry that justification cannot as such add epistemic value to true belief. So, there is a worry for any view on which the epistemic value of justified belief consists in its being the mere product of something that is only instrumentally good for promoting true belief. Reliabilists are not the only epistemologists who accept this assumption. So, the problem is a broad one. Indeed, it is easy to construct a general argument against the Instrumentalist version of the view that true belief is the fundamental epistemic value viz.: (A: Instrumentalist Veritism) Items other than true belief are epistemically good iff they (i) tend to produce true beliefs or (ii) are products of a type of source with property (i). For the following is a natural generalization of the points about the coffeemaker case: (B: Swamping Premise) If X has its source in something that is only instrumentally good relative to property F and X already exemplifies F, the mere fact that X has that type of source cannot as such make X better. According to Instrumentalist Veritism, justifying and rationalizing sources only have instrumental epistemic value. So, given (A) and (B), we get incredible results like: 10 Cf. Pritchard (2010, 2011). Jones (1997) recognized this from the beginning. 9

10 (*) A true belief s being justified in the externalist sense cannot as such make that belief epistemically better. There is, accordingly, a general problem for Instrumentalist Veritism. The question now is whether we should reject the Veritist part or the Instrumentalist part. I will turn to show that the Veritist part is not the culprit: Veritists who reject Instrumentalism can avoid the swamping problem. 3.2 The Veritist Half Is Not the Culprit Why would it help Veritists to reject Instrumentalism? Because there are other forms of derivative value emphasized by non-epistemic axiologists that are swamping-proof. To bring this out, let s consider another form of derivative value. Moore said: [T]he proper appreciation of a beautiful object is a good thing in itself. 11 While the appreciation of beauty is a good thing in itself, it is implausible that it is fundamentally good. It is derivatively good: appreciating beauty is good because beauty is good. It is no good to appreciate trash, after all. But this because is not an instrumental because. Appreciating beauty does not reliably tend to produce beauty. Moreover, such appreciation is not often the product of anything beauty-conducive. How does the explanation proceed? On a more natural model, the instance of appreciation derives value from its object because (a) its object is good and (b) it is a way to value that good object. The result is attractive: appreciating beauty seems derivatively but also non-instrumentally good. Some ethicists have erected axiologies on the basis of this idea. Thomas Hurka, for example, holds that there are basic non-instrumental values such as beauty and pleasure, and suggests the following principle for deriving non-basic non-instrumental values from basic ones: Hurka s Principle: If V is a value (in a domain D), proper ways of valuing V and their manifestations have derivative non-instrumental value (in D). I now have two points to make: (i) the kind of value derived via Hurka s principle is immune to swamping by the presence of the relevant fundamental value, and (ii) Veritists can appeal to (i) to avoid swamping. To see (i), consider: Stronger Swamping Premise: For no type of derivative value and no sense of has its source in is it true that: if X has its source in something that only has derivative value relative to property F but X already has F, X s having that source as such makes X better. This is false. Suppose Alice performs an act of beneficence because she values beneficence, while Beatrice performs the same kind of action as a PR stunt. Alice s action seems to have greater moral worth than Beatrice s. Yet just as appreciating beauty is good because beauty 11 Moore (1903: Ch. VI, 114). 10

11 is good, so valuing beneficence and manifesting this valuing is good because beneficence is good. It is just that the because cannot be understood purely instrumentally. Here an act derives value from a source that is only derivatively good relative to another property the act exemplifies. The source manifested is (1) Alice s valuing of beneficence, and the property exemplified is (2) beneficence. Yet it is plausible that the sheer fact that Alice s action has its source in (1) makes that action worthier per se. In a picture: derives some value from act manifests has property of valuing of beneficence beneficence whose value is parasitic (Hurka-wise) on This would be impossible if the Stronger Swamping Premise were true. So it is false. This suggests that Veritism by itself creates no problem. At its core, Veritism says: True belief is the sole fundamental epistemic value, and other epistemic properties are only derivatively valuable in some truth-oriented way. This thesis is independent of Instrumentalism. If Veritists adopt a different model of value derivation, they can avoid swamping as it was avoided in Alice s case. On the view developed in 4, the beliefs we admire from the epistemic point of view gain epistemic value by manifesting ways to place value on accuracy in thought. The case of knowing parallels the case of Alice s beneficent action. Knowledge requires a belief whose accuracy manifests a following truth-oriented disposition: a disposition to believe something only if there is sufficient objective and subjective evidence that the belief is accurate. Having and manifesting this disposition just is a way to place value on accuracy in thought. Placing value on accuracy in thought is epistemically good because accuracy is epistemically good. But the because here is like the because in the claim that valuing beneficence is good because beneficence is good. Paralleling Alice s case: 11

12 derives some value from belief manifests way of placing value on accuracy in thought has property of accuracy whose value is parasitic (Hurka-wise) on I develop this picture further in 4. What matters now is that it cannot be the Veritist part of Instrumentalist Veritism that generates the generalized swamping problem. 3.3 A General Problem for Instrumentalists To put a nail in the coffin, I will now show that if Instrumentalism were true, other economical epistemic axiologies would face relatives of the swamping problem. To see this, note that there is another side to the coffeemaker analogy. We do not only think that coming from a reliable coffeemaker per se cannot improve good cups of coffee. We also think that coming from a reliable coffeemaker per se cannot improve bad cups of coffee. If one takes a sip of terrible coffee and cringes in disgust, it is no comfort to be told: But look: at least it was produced by a reliable coffeemaker! These observations support a more striking sibling of the original Swamping Premise: (Dud Principle) If X is produced by a source that is only instrumentally good for producing good Fs, that cannot as such make X better if X is otherwise a bad F. Carter and Jarvis (2012) took this to undermine the intuitions behind the swamping problem. For they thought the Dud Principle would imply, crazily, that non-factive epistemic properties most saliently justification are never epistemically valuable. But they were hasty: that crazy conclusion follows only if we grant Instrumentalism. Like the original Swamping Premise, the Dud Principle concerns the kind of value that supposedly attaches to mere products of instrumentally good sources. The conclusion would not follow from the Dud Principle if non-factive epistemic properties had a different kind of derivative epistemic value. Only if the Dud Principle applies to other kinds of derivative value is Carter and Jarvis s conclusion justified. The Dud Principle, however, does not generalize for the same reasons why the Swamping Premise did not generalize. Consider someone trying to perform a charitable act because she values charity and failing only due to bad luck. This person s efforts remain better than the failed efforts of someone just looking for a PR boost. What the Dud Principle suggests is that the Instrumentalist s model is an incomplete model of derivative epistemic value. After all, it is not as if Carter and Jarvis can convince 12

13 us that terrible coffee is better if it comes from an otherwise reliable coffee machine. Like the original Swamping Premise, the Dud Principle captures a fact about mere products of instrumentally valuable sources. It is thus easy to see that any modest epistemic axiology that embraces Instrumentalism will face a relative of the swamping problem. Suppose one views knowledge as the fundamental epistemic good. Given Instrumentalism, how could one explain the epistemic value of justified beliefs? One must claim that such beliefs are good by being products of knowledge-conducive sources or by constituting knowledge. The Dud Principle will make it puzzling why justified false beliefs are epistemically good. From an Instrumentalist point of view, they are duds just like bad cups of coffee from otherwise reliable coffeemakers. Adding more items to the stock of fundamental epistemic values will not help. Even if one adds justified belief, knowledge, understanding, and true belief to the stock of fundamental epistemic values, there remain epistemic values that (a) no one can reasonably take to be fundamental, but (b) admit of no Instrumentalist explanation. Consider the epistemic value of trying one s best to form one s beliefs accurately. This is a paradigmatically derivative value. Trying one s best to do something good is admirable partly because it is directed at something good. But if we accept Instrumentalism, it is hard to explain why it is derivatively epistemically good. Merely trying one s best to form one s beliefs accurately is not reliably instrumental to accuracy, knowledge, justification, etc. Yet there remains something admirable in one s best efforts. This is not evidence of a new fundamental value. This value is paradigmatically derivative. So, any modest epistemic axiology that endorses Instrumentalism faces some relative of the swamping problem, even including a pluralist one that counts justified belief, knowledge, understanding, and true belief as fundamental epistemic values. For there are derivative epistemic values that admit of no generally satisfying Instrumentalist explanation; trying one s best to believe accurately is one example, as we just saw. 3.4 Taking Stock and the Implications for Rationality We should conclude that Instrumentalism is mistaken. Non-epistemic value theorists have already drawn similar conclusions. But if we reject Instrumentalism, we can explain the Value Datum consistently with Derivativeness. As I will explain further in 4, we can say that rational beliefs are epistemically valuable because they manifest certain ways of placing value on accuracy in thought. We can explain why those ways of placing value on accuracy in thought are epistemically good by appealing to Hurka s Principle. A virtue of this story is that it is just another application of the kind of story needed to explain the epistemic value of knowledge, justified belief, and other kinds of belief we admire from the epistemic point of view. After all, we have learned from the swamping problem that no Instrumentalist story will do. 13

14 4 Explaining the Value Datum with a Unified Strategy 4.1 Placing Value on Accuracy in Thought My view treats the beliefs we admire from the epistemic point of view as manifestations of ways to place value on accuracy in thought. I understand these ways of placing value on accuracy in thought as different ways of honoring the following ideal of accuracy: (AI) It is correct to believe P iff P is true. Obviously, we can rarely conform to AI directly. But we can honor AI indirectly. By doing so, we place value on accuracy in our thinking. Coherent, rational, justified, and knowledgeable beliefs can be understood as epistemically good from a truth-oriented point of view by manifesting different ways to place value on accuracy in one s thinking: commitment, respect (which comes in two varieties distinguished in 4.2), and compliance. An upshot of my view is worth noting in advance. Many epistemologists believe we are forced to choose between valuing the evidence and valuing the truth. This belief is mistaken. Correctly responding to evidence just is a way to place value on accuracy in thought. It matters because accuracy matters. But this because is like the because in the claim that the appreciation of beauty is good because beauty is good, which is not instrumental. Epistemologists wrongly assume that the because can only be understood in an Instrumentalist way. Understood in that way, it is hard to see how it could be true that we value heeding the evidence because we value accuracy. But this just provides another reason to reject Instrumentalism Commitment and Coherence To see the first way to honor AI, consider other ideals, like this ideal of politeness: (PI) It is correct to say Q to some casual interlocutor only if saying Q would not make this interlocutor pointlessly uncomfortable. Imagine that Edward is trying to make you pointlessly uncomfortable by saying Q. He may fail: unbeknownst to him, Q may put you at ease. If so, he conforms to PI. Still, he is clearly reckless with respect to PI, and manifests a lack of commitment to PI. Being committed to an ideal is part of what it takes to honor it. Commitment to an ideal in φ-ing is grounded in a disposition to φ only if one takes there to be sufficient evidence that φ-ing is in conformity with that ideal. When we criticize people for recklessness with respect to some ideal, it is this we find lacking. Like any ideal, AI calls for commitment. Consider someone who takes himself to have conclusive evidence that he believes P inaccurately but believes P anyway. He does not place sufficient value on accuracy in his thinking. He might believe accurately anyway. But he fails to manifest commitment to AI in believing P, like how Edward failed to manifest commitment to PI in saying Q. 14

15 Commitment to AI lines up with one interesting epistemic value viz., coherence. Complying with the norms of doxastic coherence is the way to manifest commitment to AI. The coherent may fail to place value on accuracy in some ways, but not by lack of commitment to the ideal of accuracy Respect and Substantive Rationality Coherence is not the whole of epistemic rationality. Epistemic rationality also has a substantive side. Substantive epistemic rationality involves a stronger way to honor the ideal of accuracy than commitment, though it does not require conformity. To understand this, compare PI again. This time imagine Edna, who seems to succeed by her own lights with respect to PI. She is confident that her interlocutor likes talking about X and so intends to bring up X. But Edna neglects certain apparent evidence. She has a seeming memory that X makes her interlocutor uncomfortable but disregards it as misleading. This apparent memory might really be misleading. If so, Edna conforms to PI. But she is negligent with respect to PI, and fails to respect PI in one natural sense. 12 Respect is another way to honor an ideal. To respect an ideal in φ-ing is to manifest a disposition to φ iff the evidence indicates that φ-ing would be in conformity with the ideal. When we criticize people for being negligent with respect to an ideal, it is this we find lacking. Like any ideal, AI calls for respect. Consider someone who takes himself to have conclusive evidence for P by randomly disregarding obvious evidence against P as misleading. While he might believe something true, he fails to place sufficient value on accuracy in believing P, just as Edna failed to place sufficient value on politeness in bringing up X. In the sense of respect that turns on responsiveness to apparent evidence, respect for AI lines up with another interesting epistemic value viz., substantive epistemic rationality. Being substantively epistemically rational is the way to respect the ideal of accuracy, in one natural sense. Rational thinkers may fail to place value on accuracy in their thinking in other ways, but not by a lack of respect, in one sense of respect Compliance and Knowledge Even if we respect an ideal, we may not honor it fully. To honor an ideal fully, one must conform to it. Not all non-conformity manifests disrespect. Imagine someone falsely telling you that a topic does not make her uncomfortable. Perhaps she recognizes that you enjoy this topic and wants to oblige you. If you bring it up, you fall short with respect to PI. But not by disrespect. Of course, mere conformity is not a way to honor an ideal. Conformity can be the lucky product of negligence or recklessness. If you fall short because you do not fully honor PI, it is not just because of non-conformity. Rather, what you fail to do is something you are sadly in no position to do: namely, to comply with the ideal. 12 There are affinities here with Darwall (1977) s general notion of recognition respect. 15

16 What is it to comply with an ideal? It is to conform with it by respecting it. In cases like the one imagined, compliance is not open to you: only if it became apparent that the topic makes your interlocutor uncomfortable would it be open to you. But compliance sometimes is open to us. When it is, it is the most fitting way to honor the ideal. Like other ideals, AI also calls for compliance. Sometimes we can conform to AI by respecting AI, thereby complying with AI. When we do, we believe accurately by believing rationally. This way of placing value on accuracy lines up with another epistemic value viz., knowing. Knowing that P is the way to comply with the ideal of accuracy. 4.2 The Accuracy Ideal and Epistemic Reasons It is worth noting what these ways of valuing have in common. The ways to place value on accuracy in one s thinking are dispositions to hold beliefs only if the epistemic reasons suggest that they are likely to be accurate. Earlier I said that there are different ways to place value on accuracy in thought because there are different kinds of likelihood and epistemic reasons. We can now better understand what I meant. There is a familiar distinction between objective and apparent normative reasons. 13 Objective reasons are facts that count in favor of acts and attitudes by bearing on their correctness. These facts are not all ones to which we have privileged access. Examples in epistemology include the kind of evidence to which we refer when we say things like: There was evidence all along that Jones was the murderer. We just discovered it. Objective reasons are by themselves rarely relevant to rationality. Only apparent reasons matter as such for rationality, and not all apparent reasons are objective reasons. 14 In a famous case from Williams (1981), Bernie orders some gin and tonic from his favorite bar but by an unusual turn of events receives petrol. There is conclusive objective reason for Bernie not to drink. But Bernie would not be irrational if he drank. In the reasons and rationality literature, people also often regard apparent reasons as apparent facts which, if they obtained, would be objective reasons. Subjective reasons are not a kind of objective reason on this view. While I think this view has some flaws, I will set them aside here. What matters is that there are such things as apparent reasons. Indeed, there are several such things, since appearance talk can be used in several ways. Some appearances are belief-relative, others are not. The appearance at issue in the Müller-Lyer illusion is not essentially belief-relative. Call this kind of appearance seemings-relative. These appearances are non-factive, but there are also factive appearances, exemplified by seeing that P, remembering that P, intuiting that P, etc. Reasons that are apparent in this sense are objective reasons. Call them factively apparent reasons. They can include introspective appearances, like the appearance associated with being aware of some experience. So, even in cases of illusion in otherwise dependable worlds, one 13 See Parfit (2001), Kolodny (2005), Schroeder (2007, 2008, 2009), Lord (2010), Vogelstein (2012), and Whiting (2013). 14 Lord (2010) disagrees, but there are cases he doesn t address. In any case, nothing here turns on whether he is wrong. 16

17 has some factively apparent reasons to believe worldly propositions. 15 Facts about one s experiences are generally reliable indicators of external reality in those happy worlds. With these distinctions in mind, I suggest that for any norm N, commitment is grounded in a disposition to φ only if, conditional on the beliefrelative apparent epistemic reasons, φ-ing is subjectively likely to conform to N; weak respect is grounded in a disposition to φ only if, conditional on the seemingsrelative apparent epistemic reasons, φ-ing is epistemically likely to conform to N; strong respect is grounded in a disposition to φ only if, conditional on the factively apparent epistemic reasons, φ-ing is objectively likely to conform to N; compliance is partly grounded in a disposition to φ only if, conditional on the all objective epistemic reasons, φ-ing objectively likely to conform to N and fully grounded in conforming to N by manifesting that disposition When the relevant norm is AI, I propose the following connections: one s belief that P is structurally rational iff it manifests commitment to AI one s belief that P is substantively rational iff it manifests weak respect for AI one s belief that P is justified iff it manifests strong respect for AI one s belief that P is knowledge iff one complies with AI in believing that P There is truth-oriented unity here: rational belief, justified belief, and knowledge all manifest different ways of honoring the ideal of accuracy. 4.3 A Unified Pattern of Explanation Given this unity in the nature of rational belief, justified belief, and knowledge, we can provide a unified explanation of their epistemic value by appealing to Hurka s Principle. We can offer the following type of story, which I illustrate with our target in this paper: 1. Rational beliefs manifest a way to place value on accuracy in thought. 2. Accuracy has fundamental value in the epistemic domain. 3. Manifestations of ways of valuing fundamental values in the epistemic domain have derivative epistemic value in this domain. (From Hurka s Principle) 4. So, rational beliefs have derivative value in the epistemic domain. 15 This is how Williamson (2000) addresses the bad case. See Schellenberg (2013) for doubts about this story. 17

18 The same story can be told about the epistemic value of justified belief and knowledge. This shows that no new story is needed to explain the epistemic value of rational beliefs. Some such story needs to be told. That is the moral of the swamping problem. The moral is that not all derivative value is explicable in terms of instrumental relations. And once we see this moral and look to value theory, we find a unified, simple explanation. It does fall out of this explanation that coherent belief, rational belief, justified belief, and knowledge are non-instrumentally epistemically valuable. Does that mean that we have explained nothing? No. To think otherwise presupposes Instrumentalism. We should accept pluralism about non-instrumental epistemic value. But this is not unparsimonious. Parsimony only demands a simple, unified explanation of why everything is epistemically valuable in terms of some FEV. That is what we have. 4.4 Generalizing My framework is flexible. While I prefer taking accuracy as the fundamental epistemic value, a knowledge-firster could replace my ideal of accuracy with: (KI) It is correct to believe that P iff one knows that P. One could still use Hurka s principle and appeal to commitment to KI and respect for KI to explain the epistemic value of coherence, substantively rational belief, and justified belief. I am just not convinced by the standard arguments for putting knowledge first. Mark well what does not follow from taking accuracy or knowledge to be fundamental epistemic values. Value theorists who reject Instrumentalism stress that being valuable is not the same as being something we ought to produce. Being valuable is being proper to value, and there are ways to properly value something other than producing it. Valuing friendship involves more by way of commitment and loyalty to one s friends than producing new instances of friendship; as Scanlon noted: We would not say it showed how much a person valued friendship if he betrayed one friend in order to make several new ones, or in order to bring it about that other people had more. 16 Ways of placing value on accuracy and knowledge are more like this. The way to place value on accuracy or knowledge is not to try to amass true beliefs or pieces of knowledge, however trivial. The way to place value on accuracy or knowledge is to honor AI or KI in various ways. And notice the form of the ideal of accuracy. It does not say: (AI*) One ought to believe P iff P is true. Granting AI*, we would be obliged to amass trivial true beliefs. But AI calls for no such thing: AI only implies that if P is true, it is correct to believe that P. The claim that it is correct to believe truths does not imply that it is incorrect to not believe truths. 16 Scanlon (1998: 88-9). Similar points are made by Anderson (1993). 18