Keith DeRose has argued that the appropriateness of certain third-person denials of knowledge and the inappropriateness of now you know it, now you don't claims are the two greatest problems facing subjective-sensitive invariantism (henceforth SSI) (DeRose Reference DeRose2009: 185). In what follows, I defend SSI by addressing both of these problems. In Section 1, I argue that denials of knowledge pose no problem for SSI because the debate has made a crucial mistake about which types of denials are appropriate. After clarifying the linguistic data and considering our attributions and denials of knowledge more carefully, I provide, in Section 2, an SSI-friendly explanation of why most now you know it, now you don't claims are inappropriate to assert. Of course, the problems raised by these claims are just one of the many problems arising from SSI's rejection of purism – the thesis that only truth-relevant factors are relevant in determining whether or not a true belief counts as knowledge. Thus, a full defense of SSI requires some explanation of why purism seems so intuitive. In Section 3, I propose that the intuitiveness of purism is a side-effect of the narrow focus of previous epistemological inquiries.
1. THIRD-PERSON DENIALS AND RETRACTIONS
The problem cases for SSI are those in which the practical stakes salient to the conversation are high and the practical stakes relevant to the potential subject of knowledge are low.
Mary and John are going to the bank on a Friday afternoon. Seeing a long line out the door, Mary calls Paul to see if he knows whether or not the bank will be open the next day. Though it is of no significance to Paul whether the bank will be open, it is of crucial importance to both Mary and John that they deposit their paychecks by the close of business on Saturday.
Mary: [after calling Paul] Paul was here last week on Saturday. He knows that the bank will be open tomorrow.
John: Banks sometimes do change their hours. Remember, we must deposit this check by tomorrow or else the bank will foreclose on our house. Are you sure Paul knows that the bank will be open?
Mary: I completely forgot about that. You're right. Paul doesn't know.
It is generally agreed that once John informs Mary of their precarious situation, it is appropriate for Mary to assert that Paul does not know. Call these straightforward denials since they straightforwardly assert the negation of the proposition that Paul knows. The contextualist argues that Mary's straightforward denial is appropriate because what she says is true. And since Mary says of Paul that he does know when the practical stakes are low, the use of ‘knows’ suggests that the standards that a subject must meet in order to count as knowing depend upon the standards that govern the conversational context.Footnote 1 When the conversational standards are high, the standards governing knowledge attributions are high and when the conversational standards are low, the standards governing knowledge attributions are low.
In contrast, SSI claims that the standards a subject must meet in order to count as knowing depend upon the subject's practical context. As a result, SSI evaluates Mary's denial of knowledge as false. After all, Paul is in a low stakes situation and so his epistemic state seems strong enough for his true belief to count as knowledge in his low stakes practical context.
Keith DeRose has concluded that “these third-person cases provide a powerful objection – to [his] thinking, a killer objection – to SSI” (Reference DeRose2009: 65). Of course, it is open for the subject-sensitive invariantist to explain the appropriateness of Mary's denial by appealing to the warranted assertibility of Mary's false denial rather than by appealing to the truth of what is asserted. These proposals have opened up a debate about the validity of these pragmatic appeals.Footnote 2 But as I will argue, proponents of SSI, as well as proponents of classical invariantism, need not appeal to any controversial pragmatic explanations because the debate has failed to recognize a problem with these straightforward third-person denials. As I will show, there is a well-known pragmatic feature of factive verbs that makes straightforward third-person denials of knowledge in high stakes conversational contexts inappropriate. Let us turn to this pragmatic feature.
1.1 Factive Presuppositions and Third-Person Denials
Certain sentences, when uttered, trigger presuppositions. “Sentence S presupposes that P if and only if the use of S would be inappropriate in a context in which the speaker was not presupposing that P” (Stalnaker Reference Stalnaker1999: 7). The classic example of presupposition triggering comes from the use of definite noun phrases. “The King of France is bald” triggers the presupposition that there exists one and only one King of France. And if a speaker successfully asserts this sentence, it will be presupposed in the conversational context that there is one and only one King of France. Just as definite noun phrases trigger presuppositions, sentences with factive verbs trigger presuppositions as well (see Kadmon Reference Kadmon2001: 218–20).
(1) John regrets going to the picnic.
(1') John went to the picnic.
(2) Bill remembered that he already paid the bill.
(2') Bill already paid the bill.
(1) and (2) respectively presuppose (1') and (2'). Since ‘knows’ is a factive verb, (3) triggers the presupposition that its complement, (3'), is true. Call this the factive presupposition.
(3) Kate knows that she passed the test.
(3') Kate passed the test.
What is important for our discussion is that factive verbs also trigger a factive presupposition under negation.
(4) John does not regret going to the picnic.
(5) Bill does not remember that he already paid the bill.
(6) Kate does not know that she passed the test.
Successfully asserting (4), (5), and (6) respectively trigger the presupposition that (1'), (2'), and (3') are true. The fact that ‘knows’ triggers a factive presupposition under negation has important consequences for the appropriateness of third-person denials of knowledge.
In the high stakes bank case above, it would be inappropriate for Mary to assert that the bank will be open because she does not possess the evidence that would license both her and John to take for granted that the bank will be open the following day. And since asserting that Paul does not know that the bank will be open triggers the presupposition that the bank will be open, it would also be inappropriate for Mary to assert that Paul does not know. Given the presupposition triggering of factive verbs, straightforward third-person denials of knowledge are inappropriate in high stakes conversational contexts. Therefore, the contextualist's main argument against SSI is based upon a false premise.
1.2 Hedged and Unpronounced Denials
Though our discussion could stop here, it would be useful to have a more thorough discussion of the linguistic data that come from this type of case. After all, the debate between contextualism and SSI has generally assumed that straightforward third-person denials are appropriate when the stakes governing the conversational context are high while the stakes governing the subject's context are low. So it may seem a bit strange to simply insist that such denials are not appropriate. By carefully considering the linguistic data, I hope to clarify and precisify my objection. As I will explain, I am not arguing that all third-person denials of knowledge are inappropriate in these cases. Rather, my aim is to show first that straightforward denials are inappropriate and second that the linguistic data only supports the appropriateness of a more nuanced set of third-person denials. And as I will argue, these more nuanced denials are perfectly compatible with SSI.
To begin, let's return to some paradigmatic examples found in the literature. In most of these cases, the speaker is retracting a previous knowledge attribution. What is surprising is that the standard view of the retraction data as straightforward denials simply fails to take into account the way the cases were actually presented. Consider the following examples (emphasis mine): DeRose writes, “I guess I didn't know that they were zebras” (Reference DeRose1992: 926). Stanley writes, “Bill doesn't really know that the bank will be open on Saturday” (Reference Stanley2005: 5)Footnote 3 MacFarlane writes, “No, I suppose I didn't” (Reference MacFarlane2005: 210).
In each of these examples, the retractions are hedged in various ways. And it seems reasonable to interpret them as tantamount to asserting, “I might not know that P.” The use of the modal ‘might’ changes what is expressed in an important way. The hedged retraction expresses that P is epistemically possible in the conversational context. In the bank cases, this retraction achieves its aim by adding back into the common ground a subset of worlds in which the bank is not open the next day.Footnote 4 Thus, one appropriate type of denial is a hedged denial. Of course, the proponent of SSI is happy to acknowledge that one may appropriately assert that S might not know that P even though S does in fact know. After all, one may not be warranted in asserting the latter.
Alternatively, in some of the retractions, like the case we considered above, the whole verb phrase (e.g., “No, he doesn't”) or just the complement of ‘knows’ (e.g., “No, he doesn't know”) is left unpronounced.Footnote 5 Call these unpronounced denials. It has typically been assumed that the unpronounced element should be filled in by a knows-that phrase. Given the problems with this interpretation, we could interpret the assertion by filling in the unpronounced element with the subordinate clause, whether or not P.Footnote 6 On this interpretation, the denial seems perfectly appropriate since the assertion that Paul doesn't know whether or not P does not trigger the presupposition that P. As in the case of the hedged denial, the unpronounced denial achieves the aim of introducing back into the common ground worlds in which the bank is not open the following day. Thus, interpreted as a knows-whether claim, the unpronounced denial appropriately retracts the truth of P from the common ground.
Of course, these knows-whether claims need not function as retractions. But if such claims are not functioning as retractions, then they must be updating the common ground in some way or other. So what information does the claim that S does not know whether or not P convey? By considering a simple answer to this question, a problem arises for my appeal to knows-whether claims.Footnote 7 If Paul does not know whether or not the bank will be open, then this appears to entail that Paul does not know that the bank will be open. Perhaps, by expressing the denial as a knows-whether claim, the speaker simply avoids triggering the presupposition that the bank will be open while at the same time asserting that Paul does not know. If this is correct, then the proponent of SSI is left with the same old problem of trying to explain why it is appropriate to assert this new type of denial when it is true that Paul knows. Fortunately, there are good reasons to reject the entailment.
To provide a complete response, one would need to provide some account of the relationship between the meaning of knows-whether and knows-that claims. However, the debate about the general relationship between knowledge-wh (e.g., knows whether, who, when, etc.) and knowledge-that has grown quite large and sophisticated. So it would go beyond the scope of this paper to argue for one or another view. However, we can use some observations from the debate to highlight why one might reject the entailment from not knowing whether to not knowing that.
It is widely acknowledged that knowledge-wh requires one to know the answer to a question, and that what counts as a good answer to the question at hand depends upon the conversational context. For example, to borrow an example found in Parent (Reference Parent2014), suppose you assert that you are in Johannesburg in response to the question, do you know where you are? This response may count as a good answer in the context where you have just stepped off a plane but not count as a good answer in the context where you are stepping out of a cab in Johannesburg. Thus, while one knows that one is in Johannesburg in both cases, one knows where one is only in the former context. Since the truth-conditions of knowledge-wh claims clearly vary depending upon features of the conversational context, the proponent of SSI should acknowledge that the meaning of knowledge-wh claims is context-sensitive. However, the proponent of SSI is also committed to the claim that the meaning of knowledge-that claims is not context-sensitive. By holding both views, one may conclude that there are contexts in which the assertion “Paul knows that the bank will be open” is true while the assertion “Paul knows whether or not the bank will be open” is false. In the high stakes bank case, in order to answer the question of whether or not the bank will be open, Paul must know that the bank will be open having not changed its hours. However, in order to know that the bank will be open in his own practical context, Paul can simply meet the lower standards. And in his context, he need not rule out the possibility that the bank has changed its hours. Therefore, the proponent of SSI can reject the proposed entailment by adopting the view that the truth of knows-whether assertions depend upon the context of utterance but the truth of knows-that assertions do not.Footnote 8
To conclude our discussion of third-person denials, let us consider the example that DeRose has presented as a best type of case for contextualism (Reference DeRose2009: 4–5). The example is particularly interesting since it appears to involve a third-person denial that is not a retraction, has no unpronounced elements, and at first glance, does not appear to be hedged in any way.Footnote 9 If such a case existed, then my response to the contextualist's criticism may appear misguided since I have simply ignored the most difficult case for SSI. Since the details of DeRose's example will not matter for our discussion, let me simply restate the relevant denial to fit our bank case. Suppose that Mary possesses the same set of evidence as Paul regarding the bank's hours; they have both been to the bank the Saturday before. In response to John's question, we can imagine Mary asserting, “Paul has the same reason I have for thinking the bank will open but, like me, he doesn't know that the bank will be open.”
This type of denial avoids my initial objection since it does not trigger the factive presupposition. But why not? The presupposition seems to be canceled because of the phrase “like me” that precedes the denial. Thus, the key to understanding this type of denial comes from making sense of this phrase and the function it plays in the conversation. So in what respects is Mary asserting Paul to be like her? In this context, the phrase expresses that both the speaker and the subject of knowledge possess similar evidence with regards to the target proposition. This much is clear. However, is anything else expressed by the phrase? Intuitively, Mary's assertion does seem to provide some other information. If Paul was asked in Mary's conversational context whether or not the bank would be open, he wouldn't be able to say either. Of course, it seems somewhat implausible to think that Mary is identifying this somewhat complicated similarity between her and Paul. However, there is a simple type of similarity that does provide this information. In Section 3, I will argue that first-person denials of knowledge typically express a lack of confidence or belief. Since my overall defense of SSI will rely on this claim, let me suppose for the moment that it is correct. On this assumption, when Mary denies knowledge to herself, she is expressing a lack of belief. And so if she claims that Paul also fails to know and fails to know for similar reasons, then she seems to be expressing that Paul shares a similar lack of confidence or belief. On this interpretation, the appropriateness of Mary's assertion is perfectly compatible with SSI. For invariantists and contextualists alike, if you lack belief, you lack knowledge regardless of the standards in play.
In summary, it is important to distinguish those third-person denials of knowledge that are appropriate in high stakes contexts and those that are not. Having considered a variety of such denials, we find that in the cases found in the literature, the appropriate denials are hedged, contain unpronounced elements, or involve other complicated phrases. However, none of these denials pose problems for SSI. In addition, the presupposition triggering of factive verbs appears to show that straightforward third-person denials are inappropriate in high stakes contexts. Therefore, by cleaning up the linguistic data, we have undermined the contextualist's most powerful argument against SSI.Footnote 10
2. NOW YOU KNOW IT, NOW YOU DON'T
Now you know it, now you don't claims are “claims to the effect that matters were (or will be, or would have been) different with regard to whether a subject knows some fact when (or if) such-and-such was (or will be, or had been) the case, where differences in circumstances being imagined between the two situations being compared (one of which is often the actual, present situation of the subject) concern only some non-truth-relevant matters” (DeRose Reference DeRose2009: 194). Consider the following examples.
(a) I know, but what I said before – “I don't know” – is true.
(b) I don't know, but I previously did.
(c) Mary does know that P, but she wouldn't have known if more had been at stake.
(d) Mary doesn't know that P, but if less were at stake, she would know.
For stylistic purposes, let's call these now-you-don't claims.Footnote 11 While these now-you-don't claims are intuitively very odd, SSI entails that now-you-don't claims can be true. As DeRose notes, this consequence of SSI arises from the rejection of purism.Footnote 12 As noted, purism is the thesis that only truth-relevant factors are relevant in determining whether or not a true belief counts as knowledge.Footnote 13 Since now-you-don't claims are always false according to purism, purists have a straightforward explanation of their oddity. However, by rejecting purism, proponents of SSI admit that speakers could truly assert now-you-don't claims. While contextualism also entails that certain now-you-don't claims, such as (a), could be true, DeRose has argued – convincingly in my opinion – that contextualists can plausibly explain why now-you-don't claims are odd in virtue of the fact that one may not assert them. He concludes that “contextualism avoids endorsing the problematic [now-you-don't] sentences that plague SSI” (Reference DeRose2009: 199).
I will set aside DeRose's defense of contextualism since my aim is to respond to his indictment of SSI. In fact, even proponents of SSI have agreed with DeRose's conclusion. While these proponents have followed DeRose by providing their own explanations of why now-you-don't claims are infelicitous, they all agree that these explanations have their limits and have concluded that the felicity of now-you-don't claims is a price one has to pay for abandoning purism.Footnote 14 I believe that both critics and defenders of SSI have been too quick in their conclusions. So on behalf of SSI, I will offer an explanation of the infelicity of a broader array of now-you-don't claims. I will then show that the small set of remaining now-you-don't claims are not as problematic as they at first appear.
The four examples of now-you-don't claims (Table I) correspond to the four types of now-you-don't claims that are relevant for SSI. LOW to HIGH claims are claims to the effect that a subject knows in the current low stakes context but wouldn't if the stakes were higher. HIGH to LOW claims are claims to the effect that a subject doesn't know in the current high stakes context but would if the stakes were lower.
Taxonomy of now-you-don't claims
2.1 HIGH to LOW cases
It has typically been thought that employed by SSI, the appeal to warranted assertibility only works to rule out first-person HIGH to LOW cases. DeRose writes, “the problem with this escape is that it only helps with some of the problematic sentences that need to be handled, leaving other, equally troublesome examples untouched” (Reference DeRose2009: 196). And he states that LOW to HIGH cases along with third-person HIGH to LOW cases are not explained by SSI's appeal to unassertibility.
Though DeRose is right about the LOW to HIGH cases, he is wrong about third-person HIGH to LOW cases. For he has failed to recognize that every third-person now-you-don't claim triggers the presupposition that P is true. For example, suppose one asserts that Mary doesn't know that P, but if less were at stake, she would know. This assertion triggers the presupposition that P so a speaker is warranted in asserting (d) only if she is warranted in asserting that P. In HIGH to LOW cases, the speaker is not warranted in asserting that P and as a result, third-person HIGH to LOW claims are inappropriate to assert. Thus, SSI's appeal to the unassertibility of now-you-don't claims covers all HIGH to LOW cases.Footnote 15 All that is left to consider are first- and third-person LOW to HIGH cases. Explaining the oddity of these claims will require a more substantial discussion about attributions and denials of knowledge.
2.2 First-person LOW to HIGH cases
First-person LOW to HIGH cases are those in which a speaker asserts, “I know but I wouldn't know if the stakes were higher.” In order to explain why these assertions are inappropriate without appealing to purism or contextualism, we must clarify what first-person denials of knowledge express when expressed within these now-you-don't claims. Denials of knowledge have a disjunctive form. A subject may fail to know that P if the alethic, doxastic, or epistemic condition required for knowledge is not satisfied. But each of these distinct ways of not knowing are not all relevant for first-person LOW to HIGH cases. Since the speaker is also asserting that she knows, then in the cases of interest, the alethic condition must be satisfied. If the purportedly known proposition were false, then the now-you-don't assertion would also be false and inappropriate. In addition, the speaker may fail to know because of a change in the speaker's epistemic state when she moves from the LOW to HIGH case. However, since now-you-don't claims do not involve comparisons across situations where the subject's epistemic state changes, we can set this possibility aside. So the remaining cases of interests are ones in which the first-person denials of knowledge are true because were the subject in HIGH, the doxastic condition, the epistemic condition, or both would not be met.
Let me first summarize my strategy in accounting for these cases. I propose that first-person denials of knowledge typically express a lack of confidence. If this is right, then in first-person LOW to HIGH cases, now-you-don't claims express that the speakers knows but would lack sufficient confidence to count as knowing were the speaker in a HIGH case. I will then argue that on this interpretation speakers are not warranted in asserting these claims. In order to justify this explanation, we must first consider what first-person denials of knowledge typically express.
Denials of knowledge are most frequently found in first-person form, and the most frequently used first-person denials leave the complement unpronounced (e.g., “I don't know”).Footnote 16 In the vast majority of these cases, the speaker is responding to a question and the salient question fixes the content of the unpronounced complement. For example, if I assert that I do not know in response to your question about why John was angry, I am simply asserting that I do not know why John was angry. While this may imply that there is some proposition(s) that I do not know, I am not directly asserting that I do not know some particular proposition.Footnote 17 So it is worth noting that the majority of first-person denials of knowledge are not expressions of a lack of propositional knowledge but ways of expressing that one does not know the answer to a question.
Limiting our concern to explicit denials of propositional knowledge, what do these denials typically express? Do they express that the alethic, doxastic, or epistemic condition required for knowledge has not been met? The corpus data suggests that first-person denials of propositional knowledge are somewhat uncommon.Footnote 18 And if we look through the data, it becomes clear that one can replace almost every use of “I do not know that” with phrases such as “I do not think/believe that”, “I am not sure that”, or “I have doubts that” without any loss of meaning.Footnote 19 For example, consider the following two examples from the concordance data:
(7) I do not know that men will ever accept that.
(8) I do not know that he can do anything to secure himself the nomination.
(7) was asserted in response to someone's expressed opinion that it will be a long time before men can accept women wearing so-called men's clothes. This denial expresses that the speaker, in contrast, does not think that men will ever accept this. (8) was asserted in response to the question of what a politician could do to win a race. (8) expresses that the speaker does not believe that there is anything the politician can do. Thus, ordinary use suggests that first-person denials of propositional knowledge are typically used to express a lack of confidence. That is, first-person denials typically express that the doxastic condition required for knowledge has not been met.Footnote 20
In order to bolster the case for this interpretation, consider the difference between the following two assertions as they would be made in a high stakes bank case.
(9) I do not know that the bank will be open but I think that it will be.
(10) # I do not know that the bank will be open but I believe that it will be open.
Why does (9) appear perfectly appropriate but (10) seem somewhat strange? In fact, without further explanation, (10) appears inappropriate to assert.Footnote 21 Assuming that the bank will be open, these embedded first-person denials of knowledge are true if the speaker fails to meet either the doxastic or epistemic condition required for knowledge. Let's first consider the possibility that these first-person denials merely express that the speaker's belief fails to meet the epistemic standard required to know. If that is the case, then (10) is perfectly appropriate to assert for one is simply asserting that one has a belief that does not count as knowledge. And there is nothing unwarranted about asserting facts about one's doxastic states even if these doxastic states fail to meet the epistemic standards required for assertion. Therefore, if first-person denials only expressed the fact that the epistemic condition required for knowledge was not met, then we would be left without a good explanation of why asserting (10) seems inappropriate.Footnote 22
Next consider the possibility that first-person denials express a lack of confidence. On this interpretation, there is a fairly simple explanation of the difference between asserting (9) and (10). It is plausible that merely thinking that p requires less confidence than believing that p. Therefore, if first-person denials express a lack of confidence that is incompatible with believing but not incompatible with thinking then one should be warranted in asserting (9) but not (10).
While the linguistic data suggest that first-person denials of propositional knowledge typically express a lack of confidence, before we apply this interpretation of first-person denials, we must convince ourselves that these denials express a lack of confidence when embedded in now-you-don't claims. After all, first-person denials do not appear to express a lack of confidence when embedded in certain conditionals.Footnote 23 For example, consider the following cases:
(9) If I did not have hands, I would not know that I had hands.
(10) If I were in fake barn country, I would not know that I was looking at a barn.
As I noted, denials of knowledge have a disjunctive form. And I have proposed that as a default, first-person denials express that the doxastic condition required for knowledge has not been met. While the denials embedded in (11) and (12) do not express a lack of confidence, there is a simple explanation why the default interpretation should not hold. In (11) and (12), the antecedents respectively express that the alethic and epistemic condition required for knowledge has not been met. So the truth of the antecedents is sufficient for the speaker to lack knowledge. Therefore, in these cases, we do not need to appeal to the default that the doxastic condition does not hold. In contrast, in the relevant now-you-don't claims, the imagined scenario is one where the practical situation of the speaker is different. The default interpretation should apply in these cases because even if we did assume that the practical situation of the subject was different, this would not entail that either the alethic or epistemic condition did not hold. After all, just because the stakes are higher does not mean that the subject's epistemic state does not meet the standards relevant for the higher stakes. Of course, the practical stakes in the imagined scenario may be extremely high and in such a scenario, it may be clear that the standards required for knowledge are not met. In these cases, the default interpretation does not apply. Let me set aside such cases for the moment since I will consider them below.
Now that we have an interpretation of these denials in the relevant cases, we have the resources to explain the inappropriateness of first-person LOW to HIGH claims. To provide this explanation, consider a variant of the bank case in which it is Friday afternoon and John is advising Mary what to do over the phone.
Mary: I just arrived at the bank to deposit my paycheck but there's a line out the door.
John: Is there any reason you have to deposit your paycheck now?
Mary: Well, if it were next weekend, I would definitely have had to wait in line. I have an important payment due next week, and the bank would have foreclosed on my house if I hadn't made my deposit. But at the moment, there's no pressing need. I would, however, prefer to get this done before the weekend is over.
John: Well, do you know if the bank is open tomorrow? If it is, you can just come back then.
Mary: # You're right, I do know that the bank will be open. I was here two weeks ago on a Saturday and it was open then. I'll just come back. Of course, if this was next Friday, I wouldn't have known that the bank would be open.
While Mary's final assertions are conjointly inappropriate, we can offer two distinct but compatible explanations of why Mary's assertions are inappropriate. The first explanation applies in a special set of cases that are important for the debate between purism and impurism. DeRose has noted that the sensitivity of ‘knows’ to practical factors is best made in cases where the subject's confidence remains stable between LOW and HIGH contexts.Footnote 24 So if first-person denials typically express a lack of confidence and we are considering just those cases in which the subject's doxastic state remains stable, first-person denials would simply be false. Mary would be falsely asserting that her confidence would be lower in HIGH than in LOW. Therefore, in the cases that are relevant for adjudicating between purism and impurism, there is a simple reason why first-person LOW to HIGH claims are odd. They are simply false.
Even if we widen our consideration of cases, we can offer an alternative explanation of why Mary's now-you-don't claim seems odd and inappropriate. My suggestion is that they are inappropriate because they undermine the speaker's self-attribution of knowledge. Mary, if she is an epistemically rational agent, should assign a confidence that is proportionate to her evidence. In addition, since Mary possesses the same exact information and evidence in both the LOW and HIGH bank cases, if she were epistemically rational, she should not change her confidence between the two cases.Footnote 25 This entails that if Mary's confidence did vacillate, we would have reason to think that Mary's current belief about the bank's hours might not be rational. Under these circumstances, we might conclude that Mary is not sufficiently sensitive to the evidence suggesting that Mary should not count as a knower in the first place.Footnote 26 Thus, by asserting that she would not know were the stakes higher, Mary is undermining her initial claim to know. So first-person LOW to HIGH claims are generally inappropriate because they express an incoherent set of claims. First, Mary asserts that she knows and is thereby confident that the bank will be open. Next, on the interpretation of first-person denials that I have argued for, the subsequent counterfactual denial of knowledge expresses that a change in the practical factors alone would reduce Mary's confidence. But this latter claim suggests that Mary should not count as a knower.
2.3 Third-person LOW to HIGH cases
To account for the remaining third-person LOW to HIGH cases, we must differentiate two situations where LOW to HIGH claims are asserted in terms of the relative epistemic strength of the subject and attributor of knowledge.
Privileged Subject: The subject's epistemic state is strictly stronger with respect to P than the attributor's.
Privileged Attributor: The attributor's epistemic state is strictly stronger with respect to P than the subject's.
2.3.1 Privileged subject cases
Now-you-don't claims would be odd if asserted in privileged subject cases. In the bank case, we would not say of the bank manager that he fails to know that the bank will be open even if the stakes were fairly high. Though it may be very important to the manager that the bank be open – perhaps his job depends upon it – he may still count as knowing. If the subject has more expertise, we will not be in a position to judge the exact strength of the subject's epistemic position. We simply know that the subject's epistemic position is stronger with respect to the relevant proposition than our own. As a result, we would not be in a position to evaluate whether or not the subject's epistemic state is strong enough for her true belief to count as knowing in the high stakes context. At best, we could offer a hedged denial by asserting that the bank manager might not know. It might be objected that the stakes could be raised high enough that, according to SSI, the bank manager doesn't know. I will return to these extreme cases below.
2.3.2 Privileged attributor cases
We are now left with a final set of LOW to HIGH cases with two important features. First, the speaker is asserting in a conversational context with low standards and she must be warranted in asserting that P.Footnote 27 Second, the attributor's epistemic state must be strictly stronger than the subject's. Note that these features are exemplified in the cases that have traditionally been used to analyze knowledge. These traditional cases present situations in which we are asking from an omniscient point of view whether a subject's true belief counts as knowledge. My plan is to consider two types of now-you-don't claims that count as true and appropriate in these situations according to SSI. The first type involves extreme practical stakes, and I will argue that these are special cases for everyone and not just the proponent of SSI. The second and most difficult type of claim involves everyday practical stakes. By clarifying exactly which of these now-you-don't claims are true and appropriate, I will provide an explanation of why these claims seem odd and why their oddity poses no problems for SSI.
Let's begin by considering our original bank case but changing it slightly so that Mary is ascribing knowledge to herself and appealing to her past experiences at the bank as evidence. Before John reminds Mary of their precarious situation, Mary claims to know that the bank will be open. At this point, we can assume that Mary believes that the bank will be open and would act accordingly by returning to the bank the next day. Next, let's assume that we, as attributors, are in a privileged epistemic position. We have heard Mary and John's conversation and earlier in the day, talked to the bank's manager who informed us that the bank's hours have remained unchanged all year and will not change anytime in the near future. Just as the bank was open last Saturday, it will also be open this Saturday. Now, it seems that given our situation, we would say of Mary that she knows that the bank will be open. Of course, she can't rule out the possibility that the bank has changed its hours but, from our epistemically privileged perspective, that is not something that she needs to consider. Similarly, she need not consider a host of other highly unlikely and unactualized alternatives. For example, she does not need to rule out the possibility that the bank explodes between its closing on Friday and opening on Saturday. Thus, from our privileged perspective, it would inappropriate to assert that Mary knows but wouldn't if the stakes were higher.
However, what if we considered a scenario in which the practical stakes were inordinately high. What if Mary's life depended on the bank's being open the next day? Would we say that she knows? In this situation, I suspect that we would not claim that she knows. For in such a case, one would know only if one could be absolutely certain, and neither we nor Mary would be justified in being absolutely certain. As a result, SSI would allow speakers to assert, “Mary knows, but if her life were at stake, she wouldn't know.” To my ear, these extreme now-you-don't claims don't sound particularly odd. More importantly, there are plausible explanations for why they are appropriate and why any plausible account of ‘knows’ should allow one to assert them.
These now-you-don't claims appear to be somewhat unique. As DeRose notes, the subject's level of confidence can and, in fact, ought to change as she moves from a low stakes situation to an extremely high stakes situation. He notes that if the subject's confidence didn't change, this “wildly inappropriate attitude seems capable of resulting in the subject being, or at least seeming to be, a non-knower” (Reference DeRose2009: 193). So perhaps, if the subject is to count as knowing even in the low stakes case, she should be the type of subject whose confidence would be lowered in the extreme case.Footnote 28 And if the subject's confidence were sufficiently reduced, then she would not meet the doxastic condition needed for knowledge. On this explanation of the example, comparisons of the subject's knowledge between low and extremely high stakes situations are tantamount to asserting that Mary knows that she is at the zoo but if she failed to be believe that she were at the zoo, she wouldn't know.
We may offer a distinct but compatible explanation of why these now-you-don't claims are appropriate. Remember that we are limiting our consideration to cases in which our epistemic state is better than Mary's. However, even in that scenario, if Mary's life were at stake, our privileged epistemic position would not be strong enough to determine whether or not Mary knew. To be in that position, we would have to be (close to) infallible with respect to the bank's hours and since we are not, we should not say of Mary that she knows. This is confirmed by the fact that if our lives were at stake, we should not assert that we know.
Cases involving extreme stakes seem to be peculiar cases and as such, they do not appear to pose a special problem for SSI. However, there do remain a set of now-you-don't claims involving ordinary stakes that count as true and appropriate according to SSI. Suppose for the sake of concreteness that from our privileged perspective Mary is justified in adopting a 0.9 degree of confidence in the proposition that the bank will be open the following day. If Mary is in the practical situation where very little rides on the bank being open, she knows that the bank will be open. However, consider another scenario in which Mary is given the opportunity to bet on the bank being open the following day. By playing this bet, she would win $95 if the bank is open and lose $1000 otherwise. According to most versions of SSI, Mary would fail to know in this practical context. Therefore, SSI entails that it would be both true and appropriate to assert, “Mary knows that the bank is open, but if she were in a situation where she were deciding whether or not to accept this bet, she would not know.”
Thus, the proponent of SSI must concede that certain now-you-don't claims involving perfectly ordinary stakes are true and appropriate to assert. In response, I hope to first clarify the precise nature of these now-you-don't claims. By doing so, we will not only be able to explain why these claims seem odd but also explain why their oddity is not especially problematic.
Critics of SSI typically describe the problematic now-you-don't claims as those in which we make comparisons between situations that differ solely in terms of practical stakes.Footnote 29 However, this is a misleading description since scenarios that differ merely in terms of practical stakes often fail to make a difference in what the subject knows. For example, if we offered Mary a chance to play a bet where she would win $100,000 if the bank were open and lose $5000 if the bank were not, she may very well count as knowing. However, this is a scenario where the practical stakes are much higher than the one in which little rides on the bank's hours. So SSI does not entail that practical importance is the practical factor of crucial relevance for knowledge. Rather, most proponents of SSI propose that the crucial practical factor is the actionability of a proposition. And “p is actionable for a subject just in case … p is warranted enough to be the subject's justifying basis for action” (Fantl and McGrath Reference Fantl and McGrath2012: 695). Actionability can make a difference because there is a general connection between knowledge and action: if one knows, then one may act on the basis of P. While Mary is justified acting on the truth of the proposition that the bank will be open when little rides on it, she is not justified in acting on that proposition when deciding whether or not to play a bet on the proposition that the bank will be open at worse than 1:9 odds. The difference in actionability is what makes the crucial difference for knowledge.
Therefore, in order to be fair to SSI, we must limit our consideration of now-you-don't claims to those that compare situations that differ in terms of the actionability of the target proposition. The most obvious way to do this is to consider claims of the following form: Mary knows that P, but if the practical situation were such that she shouldn't act on P, then she wouldn't know. Such claims do not appear to be particularly strange, and I will explain why below. Of course, any now-you-don't claim that picks out an alternative scenario that has as a consequence a difference in actionability will also count as appropriate. Thus, some assertions of the form, “Mary knows but wouldn't know if the stakes were higher” will count as both true and appropriate when the difference in stakes amounts to a difference in actionability.
By limiting the types of now-you-don't claims that count as true and appropriate according to SSI, we can provide a plausible explanation of why they are odd and why their oddity is neither special nor problematic. As purists have noted, assuming that Mary does know, there is nothing odd about asserting that she wouldn't know if the target proposition were false. In contrast, given the same assumption, there is something very odd about asserting that if the stakes were higher, Mary wouldn't know. I take it that the reason why the former assertion is normal is because it is perfectly clear that the truth of the antecedent entails the truth of the consequent. In contrast, the latter assertion is odd because there is a disconnect between the truth of the antecedent and consequent.
The first thing to note is that the oddity of this latter assertion seems perfectly akin to the oddity of asserting that if Genghis Khan liked oranges, then Chile would have won the World Cup. This assertion is odd for the same exact reasons; there is no reasonable connection between the truth of the antecedent and consequent. Of course, there very well may be a complicated relationship, and were we convinced of this relationship, the assertion would seem to be both true and appropriate. If we consider the odd now-you-don't claims – those that compare situations in terms of practical stakes – proponents of SSI have a perfectly plausible and intuitive explanation of how the truth of the antecedent and consequent are related. Those now-you-don't claims that count as both true and appropriate according to SSI are those that compare situations that differ in terms of actionability, and actionability does make a difference to what we know.
While critics of SSI may not accept the claim that knowledge is related to actionability, there is no denying that the connection is, at first glance, intuitive. My modest proposal is that the intuitiveness of this connection explains why the oddity of now-you-don't claims is neither special nor problematic. To summarize, these claims are odd because there is no obvious connection between practical stakes and knowledge. However, as we noted, not all now-you-don't claims that compare situations that differ in terms of practical stakes count as true and appropriate. Only when a difference in practical stakes makes a difference in actionability are these claims true and appropriate. And in these cases, the proponent of SSI can appeal to the connection between action and knowledge to offer a perfectly systematic and intuitive explanation for why the truth of the antecedent entails the truth of the consequent.Footnote 30
3. THE INTUITIVENESS OF PURISM
As DeRose has noted, the debate about now-you-don't claims is very closely tied to the debate about the intuitiveness of purism.Footnote 31 As the default account of knowledge, purism is a view that we must have good reasons to abandon. Fantl and McGrath write, “[impurism] seems mad because it entails the falsity of purism. Purism – many will say – seems clearly true” (2009: 28). Proponents of impurism have thereby taken on the burden of trying to offer good arguments and reasons to reject purism. Stanley (Reference Stanley2005) argues that our use of ‘knows’ supports the denial of purism. Fantl and McGrath (Reference Fantl and McGrath2009) argue that we must reject purism if we want to remain anti-skeptical fallibilists while embracing the intuitive connections between knowledge and action. While the case against purism depends upon these arguments, it would also be beneficial for impurists to offer some story of why purism has seemed so intuitive. Thus, to conclude my defense of SSI, I will offer one such story. Though it is not meant to be the only or best story that can be offered, it has the advantage of being quite simple and straightforward. My proposal is that purism is intuitive because practical factors are simply irrelevant in the standard cases used to analyze knowledge. However, this does not entail that purism is true so long as these so-called standard cases turn out to be special cases.
What are the central features of the cases standardly used to analyze the concept of knowledge? First, the standard cases are those in which a subject fails to know. After all, the standard cases are those that are meant to be counter-examples to proposed analyses of knowledge. And we now have a repository of cases in which subjects have true beliefs, justified true beliefs, and so on but do not know. Second, the standard cases are those in which we, the attributors and evaluators of knowledge, are omniscient. In the very least, we know all the relevant facts.
If we simply look at individual cases that have these two features, then we find that these are cases in which, regardless of the practical context the subject is in, she fails to know. For example, consider cases in which the subject truly believes that P without any evidence. No matter how high or low the stakes involved, we would judge that the subject does not know. Next consider Jones' getterized but justified true belief that someone in the office owns a Ford. Whether or not Jones is betting on this proposition at the cost of $1 or his life, Jones fails to know. Therefore, in each of the counter-examples to the many failed analyses of knowledge, we possess a large repository of cases in which subjects fail to know regardless of the practical contexts they are in.
Proponents of SSI may grant that there are many examples in which the practical context of the subject is irrelevant in determining whether or not a subject knows. After all, these examples, by themselves, do not entail that the practical context of the subject is wholly irrelevant. To explain, consider an analogous proposal that evidence is irrelevant for knowledge. And now consider all the cases in which the purportedly known proposition is false. In each of these cases, we would judge the subject not to know regardless of the subject's evidence. So in each of these cases, the subject's evidence is irrelevant for determining whether or not a subject knows. Of course, these cases do not entail that the subject's evidence is irrelevant. And we can simply test to see whether the subject's evidence is relevant by considering pairs of cases in which everything is held fixed except for the subject's evidence. If we can find a pair of these cases where the subject knows in one case but not the other, then we will have reason to think that the subject's evidence is relevant for knowing. And here the right pair of cases are ones in which the subject maintains a true belief.
Of course, it may never have seemed intuitive to think that the subject's evidence was irrelevant to knowing. However, there are truth-relevant factors that may have intuitively seemed irrelevant. Prior to the introduction of Gettier cases, it may have seem intuitive to think that external factors unknown to the subject – other than the truth or falsity of the target proposition – were irrelevant in determining whether or not a subject knew. It was only after we tested this intuition by identifying the right pairs of cases that we came to recognize that these intuitions were simply a result of our philosophical near-sightedness. My suggestion is that the same type of near-sightedness explains the intuitiveness of purism.
In order to see whether the subject's practical context could be relevant, we should consider pairs of cases in which everything is held fixed except for the subject's practical context. Of course, these are exactly the cases that proponents of impurism have focused on. Fantl and McGrath (Reference Fantl and McGrath2009) have made it explicit that the argument for impurism depends crucially on the fact that there exists a pair of cases in which only the subject's practical context varies but the subject knows in one case and does not know in the other.
The intuitiveness of purism may result from the contingent historical fact that the cases typically used to analyze knowledge are those in which the practical context of the subject is irrelevant in determining whether or not a subject knows. And we have mistakenly inferred from these cases to the conclusion that practical factors are wholly irrelevant. After testing this intuition by considering the right pairs of cases, we may find that our intuitions are a product of our near-sightedness. Thus, combined with my response to the contextualist's main criticisms, the case for SSI appears at least as good as the case for contextualism.Footnote 32
