Knowledge, Assertion, and Invariant Standards

Assertion, Knowledge, and Invariant Standards

William S. Larkin

Epistemic contextualism is the view that the truth-conditions for knowledge attributions can vary across contexts as a result of shifting epistemic standards. According to Keith DeRose, the “chief bugaboo of contextualism has been the concern that the contextualist is mistaking variability in the conditions of warranted assertibility of knowledge attributions for a variability in their truth conditions” (167).[1] In “Assertion, Knowledge, and Context,” Keith DeRose attempts to hoist the proponents of this type of warranted assertibility objection on their own petard. His strategy is to use the independently motivated idea that warranted assertion requires knowledge to show that the variable conditions of warranted assertibility to which proponents of the warranted assertibility objection are committed actually imply epistemic contextualism. I will argue that DeRose’s attempt to use the knowledge account of assertion to defuse the warranted assertibility objection to contextualism fails, and for a reason that poses a general challenge for the contextualist. For there is a warranted assertibility maneuver to be made against the whole range of the contextualists’ most compelling cases that is effective and that remains unscathed by the argument from the knowledge account of assertion.

1. Contextualist Cases

The primary motivation for contextualism has been to provide a satisfying response to philosophical skepticism; but the primary support for contextualism needs to come from, as DeRose says, “how ‘knows’ and its cognates are utilized in non-philosophical conversation” (168). The contextualist needs to produce a pair of ordinary conversational contexts where a sentence of the form ‘S knows that P’ is true in one but false in the other, though there is no difference in the truth of P or the strength of S’s epistemic relation to P—i.e., S’s evidence for and/or reliability concerning P.[2] Adequate support for contextualism requires an intuitively compelling pair of cases in which the quality of a subject S’s epistemic relation to a proposition is sufficient for knowledge in one case but not the other. As DeRose puts it:

Following Austin’s lead, the contextualist will appeal to pairs of cases that forcefully display this variability: “Low standards” cases in which a speaker seems quite appropriately and truthfully to ascribe knowledge to a subject will be paired with “high standards” cases in which another speaker in a quite different and more demanding context seems with equal propriety and truth to say that the same subject (or a similarly positioned subject) does not know. (169)

Consider DeRose’s Bank Cases.[3] In these cases a subject S claims to know that the bank is open on Saturday morning. This proposition is in fact true and S’s belief is based on the ‘quite solid grounds’ that S was at the bank two weeks ago and found that it was open until noon on Saturdays. In the first case—Low Bank—S and his wife are deciding whether to deposit their paychecks on Friday or wait until Saturday morning, “where no disaster will ensue if (they) waste a trip to the bank on Saturday only to find it closed” (170). In the second case—High Bank—“disaster, not just disappointment, would ensue if (they) waited until Saturday only to find (they) were too late” (170). In Low Bank it does not seem reasonable to question the adequacy of the available evidence or require that one go to further lengths to determine whether the bank has changed its hours over the last couple of weeks. But in High Bank, it does seem reasonable to question the adequacy of the available evidence and require further efforts to make sure that the bank has not changed its hours. In Low Bank almost everyone would say that S does know that the bank is open on Saturday; but in High Bank almost everyone would say that S does not know that the bank is open on Saturday. So here we have a pair of cases where it seems that the truth conditions for a knowledge attribution, ‘S knows that the bank is open on Saturday mornings’, vary across contexts without a change in the subject’s epistemic relation to the relevant proposition.

The epistemic contexualist contends that the contextual sensitivity of knowledge attributions in such cases is the result of the contextual variability of the standards for knowledge. Given the relatively small risk involved in Low Bank, the available evidence is adequate; but given that disaster will ensue if S is mistaken in High Bank, the evidence that the bank was open on Saturday morning two weeks ago is no longer adequate. According to the contextualist, such cases as these show that as the practical stakes go up so do the standards for knowledge. Thus a knowledge attribution may be true in one context but false in another because rising epistemic standards have rendered what was once an adequate epistemic relation to a proposition inadequate.

The warranted assertibility objection (WAO) to contextualism contends that our intuition concerning at least one of the contextualist’s cases is in fact mistaken. The proponent of a WAO then attempts to explain that intuition away using a warranted assertibility maneuver (WAM) designed to show that the relevant attribution merely seems false because its assertion is not warranted. There are two WAO strategies available: First, one can claim that S does not really know that P in the purportedly low standards case and that ‘S does not know that P’ merely seems false there because its assertion is unwarranted. We can call this first strategy a Cartesian WAO, since it would be the approach taken by one who favored invariantly high epistemic standards. Second, one can claim that S really does know that P in the purportedly high standards case and that ‘S knows that P’ merely seems false there because its assertion is unwarranted. We can call this second strategy a Moorean WAO, since it would be the approach taken by one who favored invariantly low epistemic standards.

I aim to put forth a Moorean WAO that is effective against a whole range of the contextualist’s most compelling cases. I will argue that the attribution ‘S knows that P’ only seems false in certain purportedly high standards cases because its assertion generates a false implicature and is therefore unwarranted in that context. I will also show that my Moorean WAO is immune to DeRose’s argument from the knowledge account of assertion, to which we now turn.

2. The Argument from the Knowledge Account of Assertion

The knowledge account of assertion (KAA) claims that “one must know that P in order to be positioned well enough with respect to P to assert it” (179).[4] DeRose argues that the independently motivated KAA can be used to show that the variable conditions of warranted assertibility presupposed by the WAO actually imply epistemic contextualism.

“One of the most important recommendations” for KAA is that “it provides a nice handling of the knowledge version of Moore’s paradox and other troubling conclusions” (180). Moore noted that there is something extremely odd about sincerely uttering sentences of the form ‘P, but I don’t know that P.’ The contents of such claims are clearly not logically inconsistent—it is clearly possible that P is true but that I do not know that it is. Yet there seems to be something like a logical difficulty involved in asserting such sentences. Moore himself explained the oddity by saying, “By asserting P positively, you imply, though you don’t assert, that you know that P.”[5]

KAA explains why a sincere assertion of P would imply, in the sense of pragmatically implicate, that you know that P: There is a general conversational rule that states that one should assert only what one is in a position to know. Given that rule as a background presupposition, my sincere assertion of P generates the implicature that I know that P. So in the Moore-paradoxical utterance, ‘P, but I don’t know that P’, I implicate something by asserting the first conjunct that I immediately deny by asserting the second conjunct. This explanation does justice to the feeling that there is some kind of inconsistency involved, even though the content of the proposition does not involve a contradiction. There is an inconsistency between something pragmatically implicated and something semantically asserted.

After providing this glimpse at the independent motivation for KAA, DeRose continues as follows:

The knowledge account of assertion provides a powerful argument for contextualism: If the standards for when one is in a position to warrantedly assert that P are the same as those that comprise the truth-conditions for “I know that P”, then if the former vary with context, so do the latter. In short: The knowledge account of assertion together with the context-sensitivity of assertibility yields contextualism about knowledge. (187)

Moreover, DeRose claims that the proponents of the WAO are in fact committed to the context-sensitivity of warranted assertibility:

It is clear and uncontroversial that knowledge attributions are in some way governed by varying epistemic standards. What makes contextualism controversial are warranted assertibility concerns—the worry that these varying standards are only conditions for the warranted assertibility, rather than the truth, of the attributions. (187)

So, according to DeRose, even the opponents of epistemic contextualism accept varying epistemic standards. Invariantists oppose contextualists only in saying that the varying epistemic standards do not affect the truth-conditions of knowledge attributions but only their warranted assertibility conditions. DeRose concludes, then, that KAA is lethal to invariantism because “it turns against the invariantists the very context-sensitivity in warranted assertability that they themselves habitually appeal to” (188).

The argument from KAA to show that proponents of the WAO are after all committed to epistemic contextualism looks like this:

P1: The standards for when one is in a position to warrantedly assert that P are the same as those that comprise the truth-conditions for “I know that P”.

[from KAA]

C1: So, if the standards for the warranted assertibility vary with context, then the truth-conditions for knowledge attributions vary with context. [from P1]

P2: The standards for warranted assertibility do vary with context.

[from WAO]

C2: So the truth conditions for knowledge attributions vary with context—i.e., epistemic contextualism is true. [from C1 and P2]

In the next section I will reject P2 of this argument by presenting a particular WAO that does not presuppose that the standards for warranted assertibility vary with context—it accounts for context sensitivity in warranted assertibility without varying standards of warranted assertibility. It is neither clear nor uncontroversial that knowledge attributions are in any way governed by varying epistemic standards. But first let me point out a significant problem with P1 of the above argument as well.

The first premise claims on the basis of KAA that the conditions for warranted assertibility are the same as the truth conditions for knowledge attributions. However, KAA is committed only to the claim that knowledge is necessary for warranted assertibility, not that it is sufficient. Thus KAA does not imply that the standards for warranted assertibility are the same as the standards for knowledge; it implies only that the standards for warranted assertibility are comprised, perhaps merely, in part by the standards for knowledge. There may be other necessary conditions for warranted assertibility. If so, the context-sensitivity of warranted assertibility may be the result of the contextual variability of something other than the standards for knowledge.

While this point is devastating to DeRose’s argument as it stands, I will not pursue it. First, there may be a more subtle argument available from KAA to P1 of the above argument that shows that the context sensitivity of warranted assertibility must come from the variability in the truth conditions for knowledge attributions and not from some other condition on warranted assertibility. But second, there is a more interesting and more generally significant objection to be made against P2 of the argument.

3. The High Stakes WAM and the KAA

The most compelling contextualist cases crucially rely on raising what is practically at stake for a certain subject S on the truth of some proposition P.[6] I will argue that there is an effective WAO to any such pair of cases that is in fact immune to DeRose’s argument from KAA. In particular, I will propose and defend a Moorean WAM that explains away the mistaken intuition that S does not know that P in high stakes cases without presupposing that the standards for warranted assertibility of knowledge attributions rise as we move from low stakes to high stakes cases. I will call my strategy the high stakes WAM.

Take a pair of cases L and H where the assertion ‘S knows that P’ seems true in L but false in H because what is at stake on the truth of P has been raised considerably from X in L to Y in H. Any convincing pair of cases of this form crucially involves that S is willing to stake X but not Y on her epistemic relation to P. It is S’s reluctance in H that tells us just how high the stakes are; and it is the stakes being so high that makes it seem that S’s epistemic relation to P is not adequate for knowledge in H, though it was adequate in L.

A Moorean invariantist can object to any such pair of cases as follows: S really does know that P in H, assuming she knows that P in L, since epistemic standards are invariantly low. The attribution ‘S knows that P’ merely seems false in H because its assertion there is not warranted. For asserting that S knows that P in H generates the implicature that S is willing to stake Y on her epistemic relation to P. Since S is not so willing, the implicature is false and the assertion is unwarranted.

I will explain in the next section exactly how the false implicature is generated in H and defend the effectiveness of the WAM just made. But first let me explain why this high stakes WAM is not undermined by the argument from KAA.

The contextualist needs to show that the standards for knowledge vary across context. The argument from KAA, proceeding from the claim that the standards for warranted assertibility are the same as the standards for knowledge, thus needs the premise that the standards for warranted assertibility vary across contexts. This DeRose contends is something to which every proponent of a WAO is committed. But my high stakes WAM does not require or presuppose that the standards for warranted assertibility change as we move from low stakes to high stakes cases.

Any WAO is of course committed to the context sensitivity of the warranted assertibility of knowledge attributions—to the idea that a knowledge attribution may be warrantedly asserted in one context but not in another. This, however, implies the contextual variability of the standards for warranted assertibility of knowledge attributions only if those standards are the sole relevant factor that can vary with context. They are not. For the conversational implicatures generated by knowledge attributions are relevant to their warranted assertibility, and the implicatures generated by knowledge attributions can vary across contexts. Asserting a particular knowledge attribution can be warranted in one context but not in another because it generates different implicatures in those contexts. If an assertion generates different implicatures in different contexts, then an assertion can go from warranted to unwarranted without a change in the standards for warranted assertibility.

This is what happens in the WAO proposed above. Asserting that S knows that P is warranted in L but not in H because different implicatures are generated in those contexts. Asserting that S knows that P in L generates the implicature that S is willing to stake X on her epistemic relation to P; while asserting that S knows that P in H generates the distinct implicature that S is willing to stake Y on that same epistemic relation to P. By an invariant standard of warranted assertibility—one to the effect that an assertion is warranted only if it does not generate a false implicature—the same assertion is warranted in L but not in H. Asserting that S knows that P is unwarranted in H but not in L because it generates a false implicature in H but not in L; it is not because the standards for warranted assertibility have risen from L to H. The high stakes WAM is thus not committed in any way to varying epistemic standards.

The chief bugaboo for DeRose’s defense of epistemic contextualism in “Assertion, Knowledge, and Context” is the concern that he is mistaking the contextual sensitivity of warranted assertibility with the contextual variability of the standards for warranted assertion.

4. The effectiveness of the high stakes WAM

In reference to a kind of WAM that he deems to be effective, DeRose says:

Putting all of our criteria together, we can see how much [the effective WAM] has going for it. It starts out with the advantage of being aimed at an intuition that is a good candidate for being WAMed: an intuition that an assertion would be false that is opposed by an intuition that the opposite assertion would also be false in the relevant circumstances. It then explains away the relevant apparent falsehood by appeal to the fact that a false implicature would be generated by the assertion in the relevant situations—which, as we have seen, is less problematic than is tryin to explain away the appearance of truth by appeal to the generation of a true implicature. And finally, it explains how the false implicature in question gets generated by appeal to very general rules, rather than to ad hoc rules that attaché only to sentences of the type in question. (193)

So according to DeRose, the following are jointly sufficient for an effective WAM: (1) it is aimed at an intuition that some assertion is false that is opposed by an intuition that the opposite assertion would also be false; (2) it explains away an appearance of falsity by appeal to a false implicature without attempting to implausibly explain away an appearance of truth by appeal to a true implicature; and (3) it explains how the relevant false implicature is generated by appeal to very general conversational rules. I will now show that by these criteria my high stakes WAM is effective.

Concerning (1): The high stakes WAM tries to explain away the strong intuition that ‘S knows that P’ is false in a high stakes situation H. But there are equally strong intuitions that combine to support the claim that ‘S does not know that P’ is also false in H. First is the intuition that S does know that P in L on the basis of the very same epistemic relation to P that she has in H; and second is the intuition that the standards governing when an epistemic relation is adequate are invariant across contexts. The contextualist cannot deny the strength of the first intuition without denying the support for her view; and she cannot deny the strength of the second intuition without denying the significance of her view.

Concerning (2): The high stakes WAM does explain away the intuition that ‘S knows that P’ is false in H by appeal to the fact that asserting it generates a false implicature. In H, asserting ‘S knows that P’ generates the implicature that S is willing to stake some relatively high cost Y on her epistemic relation to P, and S is not so willing. Now the high stakes WAM does need to explain away the intuition that ‘S does not know that P’ is true in H; but it can, and indeed must, do so without appealing to the fact that the assertion generates a true implicature. It can do so as follows: ‘S does not know that P’ seems true in H simply because its negation ‘S does know that P’ seems false in H. It must do so, as we will see in the next section, because asserting that S does not know that P does not actually generate the implicature, which would be true, that S is not willing to stake Y on her epistemic relation to P.

Concerning (3): Asserting ‘S knows that P’ generates the implicature that S is willing to stake Y on her epistemic relation to P in virtue of the general conversational rule that one should assert only what one is willing to act on. When S asserts that P she is publicly endorsing P, and this public endorsement has the effect (and is often intentionally made precisely to have the effect) of inducing or inclining others to act on the truth of P. Since S should induce or incline others to act on P only if she herself is willing to act on P, it follows that S should assert P only if she is willing to act on P herself. We can call this the act account of assertion (AAA).

Now, when S* asserts in a context C that S knows that P, S* is semantically implying that S would (under suitable conditions) assert that P in C. For to say that S knows that P is to say in part that S (strongly) believes that P, which in turn is to say in part that S would assert that P. So when S* claims that S knows that P, she thereby implies that S would assert that P. That implication, given AAA, generates the implicature that S would be willing to act on P. In a high stakes context H, being willing to act on P means being willing to risk Y. So asserting that S knows that P in H generates the implicature that S is willing to stake Y on her epistemic relation to P; and it does so in virtue of AAA, which is a very general conversational rule that does not attach only to knowledge attributions.[7]

So, putting all of DeRose’s criteria together, we can see how much the high stakes WAM has going for it. It is aimed at an intuition that is a good candidate for being WAMed. It explains away the relevant apparent falsehood by appeal to the fact that a false implicature would be generated by the assertion in the relevant situations. And finally, it explains how the false implicature in question gets generated by appeal to very general rules.

DeRose at one point complains that:

No defender of invariantism has proposed an account on which general rules of conversation together with their proposed invariantist account of the truth-conditions of knowledge attributions predict the pattern of varying conditions of warranted assertibility of knowledge attributions that we encounter in natural language. (176)

Well, the high stakes WAM combines a broadly Moorean invariantist account of the truth-conditions of knowledge attributions with the general rule of conversation that one should assert only what one is willing to act on. And this combination does predict the pattern of varying conditions of warranted assertibility that we encounter in cases that involve raising what is practically at stake on the truth of a certain proposition.

5. AAA and Knowledge Denials

Given AAA, knowledge attributions of the form ‘S knows that P’ generate conversational implicatures to the effect that S is willing to act on P. Such attributions semantically imply that S believes that P and thereby that S would assert that P; and saying that S would assert that P implicates that S is willing to act on P. Knowledge denials, however, of the form ‘S does not know that P’ do not generate implicatures to the effect that S is not willing to act on P. This turns out to have a couple of significant consequences.

Before we look at those consequences, let’s see exactly why knowledge denials do not generate implicatures via AAA. To start with, given that there are other necessary conditions for knowledge besides belief, ‘S does not know that P’ does not by itself semantically imply that S does not believe P or that S would not assert P. But if we accept KAA, of course, then saying that ‘S does not know that P’ does pragmatically implicate that S would not assert that P. Still, this later claim does not then implicate that S is not willing to act on P. At least it does not do so via AAA. For the act account of assertion claims that one should assert only what one is willing to act on, not everything that one is willing to act on. So the fact that S would not assert that P does not implicate anything about S’s willingness to act on P. S might turn out to be some kind of epistemic daredevil, willing to act on all kinds of things she does not know, without violating any conversational rules.

Now let’s take a look at the consequences of the fact that AAA does not help generate any implicatures from knowledge denials. First, this fact explains why we must account for the apparent truth of ‘S does not know that P’ in a high stakes context H by appeal to the fact that ‘S knows that P’ merely seems false because its assertion is unwarranted. We cannot say, even if we wanted to, that ‘S does not know that P’ merely seems true in H because it generates a true implicature and is therefore a warranted assertion. Though it is true in H that S is not willing to risk Y on her epistemic relation to P, asserting ‘S does not know that P’ does not generate that claim as an implicature.

Second, the fact that AAA does not help generate any implicatures from knowledge denials shows why we must make a Moorean WAO and employ a high stakes WAM in response to contextualist cases. We cannot, on the basis of AAA, make a Cartesian WAO and employ a low stakes WAM. For, we cannot effectively explain away our intuition that ‘S knows that P’ is true in some low stakes context L by saying that it merely seems true because it generates a true implicature and is therefore a warranted assertion. That attribution does generate what would be a true implicature that S is willing to act on P in L. But as DeRose has pointed out, the general strategy of explaining away the appearance of truth by relying on a true implicature is problematic.[8] This leaves us with having to account for the merely apparent truth of ‘S knows that P’ in L on the basis of the merely apparent falsity in L of ‘S does not know that P.’ But we cannot account for the apparent falsity of that denial by claiming that it generates a false implicature and is therefore unwarranted. Though it is false in L that S is not willing to risk X on her epistemic relation to P, asserting ‘S does not know that P’ does not generate that clam as an implicature.

The act account of assertion thus provides the resources for an effective Moorean WAO to the contextualists’ most compelling cases but not for an effective Cartesian WAO to those cases. Thus it seems at this point that if we want to be invariantists, we ought to prefer a common sense epistemology with relatively low epistemic standards to a more philosophical epistemology with relatively high epistemic standards.

6. Conclusion

DeRose concludes “Assertion, Knowledge, and Context” by saying

Not only does [the warranted assertibility objection] fail the tests we have discerned, but, as we saw above…what a close look at the conditions for warranted assertibility actually leads to is a powerful argument in favor of contextualism” (194).

I have presented here, however, a particular WAO that passes those tests and avoids the argument from the knowledge account of assertion. The high stakes WAM explains away the intuition that ‘S knows that P’ is false in a high stakes context by showing how that assertion generates a false implicature in virtue of the general conversational rule that one should assert only what one is willing to act on. Moreover, the high stakes WAM accounts for the context sensitivity of warranted assertibility for knowledge attributions without presupposing a contextual variability in the standards of warranted assertibility for knowledge attributions. Since the argument from KAA to contextualism presupposes a contextual variation in the standards of warranted assertibility for knowledge attributions, that argument does not undermine the high stakes WAM.

The high stakes WAM poses a general challenge for the contextualist. She needs to produce a pair of cases where S knows that P in one but not the other because the standards for knowledge have gone up. The most compelling pairs of cases involve raising what is practically at stake on the truth of P to the point where S is unwilling to act on her given epistemic relation to P. Any such case however is susceptible to the high stakes WAM. A Moorean invariantist can say that S really does know that P in the relevant high stakes case, though it appears that she does not because saying so generates the false implicature that S is willing to risk some relatively high stake on her epistemic relation to P. Given that the high stakes WAM is effective and avoids the argument from the knowledge account of assertion, the contextualist is left with the task of finding an intuitively compelling pair of cases to support her view that do not crucially rely on raising what is practically at stake for some agent on the truth of a proposition.

[1] All parenthetical page references are to: DeRose, Keith. “Assertion, Knowledge, and Context”, The Philosophical Review, Vol. III, No. 2. (April 2002).

[2] I intend the notion of an epistemic relation to P to be neutral on the question of internalism vs. externalism.

[3] The Bank Cases originally appear in DeRose, “Contextualism and Knowledge Attributions”, Philosophy and Phenomenological Research 52 (1992).

[4] The knowledge account of assertion has been defended by G.E. Moore in his Commonplace Book: 1919-1953 (London: Allen & Unwin, 1962); by Peter Unger in his Ignorance (Oxford: Oxford University Press, 1975); and by Timothy Williamson in his Knowledge and its Limits (Oxford: Oxford University Press, 2000). The knowledge account of assertion can take the form of a claim that when one asserts a proposition one thereby represents oneself as knowing that P (as Moore would have it); or it can take the form of a constitutive rule of the practice of asserting that one should assert only what one knows (as Williamson would have it). DeRose takes these two different forms to be ‘two sides of the same coin’.

[6] DeRose says, “it makes the relevant [contextualist supporting] intuitions more stable if …the raise in epistemic standards is tied to a very practical concern” (169). And in “The Ordinary Language Basis of Contextualism”, forthcoming Philosophical Quarterly, he says that the best case pairs for “deciding between contextualism and invariantism in the first place…will be a low standards case where the stakes are low…and a high standards case where the stakes are high.”

[7] In using the ambiguous formulation ‘asserting that S knows that P in H’, I am avoiding certain complications concerning whether the assertion is made in the same context that it is talking about.

[8] He says, “By contrast, it seems much more problematic to claim that an assertion that seems true is in fact false by means of a claim that, though the assertion itself is false, it generates a true implicature and is therefore a warranted assertion. For, except where we engage in special practices of misdirection, like irony or hyperbole, don’t we want to avoid falsehood both in what we implicate and (especially) in what we actually say? So, it would seem that a false assertion will remain unwarranted despite whatever true implicatures it might generate.” (193)