Meta-epistemic defeat

Abstract

An account of meta-epistemic defeaters—distinct from traditional (first-order) epistemic defeaters—is motivated and defended, drawing from case studies involving epistemic error-theory (e.g., Olson 2011, Reasons for belief; cf., Streumer 2012, J Philos 110:1–25) and epistemic relativism (e.g., MacFarlane 2005, Oxford Stud Epistemol 1:197–233; 2011; 2014, Assessment sensitivity: Relative truth and its applications). Mechanisms of traditional epistemic defeat and meta-epistemic defeat are compared and contrasted, and some new puzzles are introduced.

1 Introduction

Most contemporary discussions of epistemic defeat focus on first-order epistemic defeaters. A defeater defeats at (merely) the first-order if it downgrades the epistemic status of a belief in a way that is neutral with respect to the meta-epistemic status of the belief. The meta-epistemic status of a belief is the status it has not in virtue of the particular belief it is, but in virtue of its being the kind of attitude that it is, whereas, the (first-order) epistemic status of a belief is just the epistemic status it has in virtue of the particular belief it is (e.g., formed in the fashion that it was, on the basis of the reasons for which it was held, etc.).

Examples of first-order epistemic status, the kind of status that stands subject to traditional (e.g., rebutting and undercutting¹) first-order defeaters, include knowledge and justification². Examples of second-order epistemic status are things like being objectively justified, being objectively known, viz., status that a given belief has in virtue of the kind of attitude it is, however, not in virtue of the particular belief it is (e.g., and so not in virtue of its being formed as it was, on the basis of the reasons for which it was held, etc.).

Obviously, first-order epistemic status matters. We care about whether our beliefs are justified, known, etc. But meta-epistemic status matters, too. As Paul Boghossian (2001; 2006) argues³, when we assess—in epistemology as well as in ordinary circumstances—whether a given belief is justified or known, we generally take it that that there is an objective fact of the matter, one that is potentially knowable⁴. And so we—most of us, anyway⁵—take it that the first-order epistemic status our beliefs enjoy is objective epistemic status, viz., objective justification and knowledge. Accordingly, just as we take it to be a kind of epistemic loss if (in the presence of a defeater) a given belief, p, is no longer justified or known, it is also natural to suppose we’ve suffered some sort of epistemic loss if, upon acquiring some new belief q, we regard the objectivity of our justification for believing p to be imperiled or vanquished⁶.

The rich literature on defeaters in epistemology is a testament to the fact that we care about the various mechanisms under which first order status can be downgraded, and potentially regained. But given that meta-epistemic status matters, too, it’s curious that the defeasibility of this kind of status has been comparatively overlooked. How exactly does meta-epistemic status get downgraded, when it does? What are the mechanisms by which this could happen, and how do these relate to the mechanisms by which first-order epistemic status gets downgraded? More generally: how does meta-epistemic defeat work⁷?

This paper will attempt to examine this general question through two case studies involving two very different positions in meta-epistemology: error theory and relativism. Here is the plan. Section 2 discusses what I will call error-theoretic meta-epistemic defeaters. Section 3 discusses relativist meta-epistemic defeaters. Though there are some interesting differences between how these two kinds of meta-epistemic defeaters work, there are also important commonalities. In Sect. 4, I’ll summarise these commonalities and taxonomise how meta-epistemic defeaters interface with familiar distinctions in the literature on (first-order) defeaters, raising some new puzzles in the process.

2 Error-theoretic meta-epistemic defeaters

Philosophers who are reluctant to agree that claims about knowledge and justification attain objective epistemic status generally agree on a much more basic point: that there are epistemic reasons for beliefs. For example, epistemic relativists such as Rorty (1980) countenance epistemic reasons. Rorty just thinks that the epistemic authority that epistemic reasons aspire to is confined to the social contexts giving rise to them, a point that absolutists like Boghossian deny. But neither denies that there are epistemic reasons.

Enter here the epistemic error theorist. Think of epistemic error-theory as analogous to moral error-theory⁸. The moral error theorist (e.g., Mackie 1977) draws our attention to the queerness of moral properties, which would be “qualities or relations of a very strange sort, utterly different from anything else in the universe” (1977, p. 38). We have, Mackie thinks, no reason to think moral properties are ever actually instantiated in the world, and thus, that there are no facts consisting in the instantiation of such properties. All statements that attribute moral properties (e.g., including attributions of moral reasons) are false.

The meta-epistemic analogue to this position—epistemic error-theory (e.g, Olson 2011) regards epistemic properties as in the same boat as moral properties. Thus, as Nishi Shah (2011, p. 96) puts it:

... Just as no persons have the property of being a witch, so [according to the epistemic error theorist] there are no facts or states of affairs that have the property of being a reason for belief. Any judgment that attributes to something the property of being a reason for belief is, therefore, false.

Epistemic error-theory is of course an odd view; if it’s true, it means that our practice of attributing epistemic reasons is fundamentally mistaken⁹. As Bart Streumer (2012) has argued, it might not even be possible to believe this view, regardless of whether it is true¹⁰. And, as Crispin Wright (1994) suggests, the truth of epistemic error theory would seem to call into doubt the sensicality of continuing to engage seriously with epistemological questions¹¹. But let’s set all of this aside.

There are currently arguments available, the most developed of which are due to Jonas Olson (e.g., 2011, 2014) which claim that epistemic error theory is in fact true. The arguments are sophisticated, though for our purposes, the details won’t matter¹². Let’s now consider the following case.

Lemonade: A student, A, sees what appears to be a glass of lemonade sitting on his meta-ethics professor’s desk, and, accordingly, forms the belief that there is. Call this belief lemonade. After some further discussion with the professor and two classmates, A then acquires three further beliefs, in addition to lemonade, which are: (i) mountain dew, (ii) hallucination and (iii) error theory. mountain dew is the belief that the meta-ethics professor drinks only Mountain Dew (which is different from, but looks like, lemonade). hallucination is the belief that A has recently ingested crystal DNT¹³, which causes inaccurate but highly believable hallucinations. error theory is the belief that Jonas Olson’s arguments for epistemic error theory (as outlined during the class by the professor drinking the yellow looking drink) are persuasive, and thus, that there are no reasons for belief.

Stipulate that A’s original belief, lemonade, is epistemically unimpeachable at the time of its formation. A, let’s assume, is justified in believing lemonade ¹⁴, prior to acquiring the further beliefs A acquires. By acquiring the belief mountain dew, A acquires (in the sense of Pollock 1986) a rebutting defeater for believing lemonade, which defeats A’s justification for lemonade by indicating that lemonade is false. By acquiring the belief hallucination, A acquires an undercutting defeater for lemonade. hallucination indicates that the process (perception) that issued A’s lemonade belief is unreliable in the present circumstances, given the ingestion of crystal DNT, and so therefore not sufficiently indicative of the truth of lemonade. But what about error theory? What implication does A’s acquiring error theory have for the status of A’s original justified belief that what is sitting on the professor’s desk is a glass of lemonade?

One line of response to this question is that A’s acquisition of the error theory belief is entirely orthogonal to the epistemic status of A’s lemonade belief. Call this orthogonalism. If orthogonalism is true, then this of course undermines the thought that A’s acquiring error theory might (in some way) defeat in any interesting way the epistemic status of A’s belief lemonade.

The reasoning in support of orthogonalism goes as follows. (i) A’s acquisition of error theory can potentially defeat the epistemic status of A’s belief lemonade only if A’s acquisition of error theory is relevant in some way to the truth of A’s belief that lemonade. But, whether (ii) A’s belief that error theory is true is irrelevant to the truth of A’s belief that lemonade; (iii) Therefore, it’s not the case that A’s acquisition of error theory can potentially defeat the epistemic status of A’s belief that lemonade.

Let’s simply grant the orthogonalist’s premise (i)—i.e., that A’s acquisition of error theory can potentially defeat the epistemic status of A’s belief lemonade only if A’s acquisition of error theory is relevant in some way to the truth of A’s belief that lemonade. Even if this much is conceded, we needn’t accept the conclusion of the orthognalist’s reasoning. This is because we should reject premise (ii). In short, this is because premise (ii) overgeneralizes in such a way that it has implausible implications concerning undercutting defeat.

Consider that undercutting defeaters are epistemically significant because they indicate that the grounds one has for one’s belief aren’t sufficiently indicative of the truth of the target proposition. After all, hallucination is epistemically significant for A vis-à-vis lemonade because it indicates that A’s grounds for believing lemonade aren’t sufficiently indicative of the truth of lemonade. But, A’s acquisition of the belief error theory also (like hallucination) indicates that A’s grounds aren’t sufficiently indicative of the truth of lemonade. Specifically, A’s acquisition of error theory indicates that A’s grounds aren’t sufficiently indicative of the truth of lemonade by indicating that A simply lacks the grounds which A originally thought A had which indicated the truth of lemonade.

What this means is that if orthogonalism is correct, then, it becomes at best mysterious why undercutting defeaters are epistemically significant. So if A’s coming to believe that Jonas Olson’s arguments for epistemic error theory are plausible can’t defeat in any way the epistemic status of A’s belief that the glass the professor is drinking from is filled with lemonade, it’s not because orthogonalism is true.

As I indicated in the previous section, I think there is an interesting kind of epistemic defeat at the meta-epistemic level. And that is indeed what I think is going on in the case described above. However, before engaging with this, I want to consider one further diagnosis of the Lemonade case according to which A’s acquisition of the error theory belief does count as a defeater for A’s justification for lemonade, but not by defeating the meta-epistemic status of A’s justification for lemonade. Rather, on the diagnosis I will now consider, the epistemic status that is defeated is just regular first-order (not-second-order) status: by acquiring error theory, A simply loses (on this envisioned diagnosis) A’s justification for believing that lemonade.

In order to get this assessment of the case in view, we need to introduce a different kind of epistemic defeater, called a no-reason defeater. According to Michael Bergmann (1997), a no-reason defeater for a belief, p, is a reason to believe that it’s no longer reasonable to believe p given that (a) one has no reason for believing p and (b) the belief that p is the sort of belief that it’s reasonable to hold only if one has evidence for p (Bergmann 1997, pp. 102–103)¹⁵. For example, suppose I believe that the President is in Boston. But then, on reflection, I can’t locate any reason I have for believing this, while maintaining also that this is the kind of proposition that’s reasonable to hold only if I have evidence for it. I thereby acquire on Bergmann’s view a no-reason defeater for my (would-be) justification for my belief that the President is in Boston.

Consider now the following question: Is A’s acquisition of the belief error-theory plausibly a no-reason defeater for A’s justification for believing lemonade? There’s an initially plausible story for why it is. And if this story is right, then the notion of meta-epistemic defeat needn’t enter the picture here.

The ‘no-reason’ diagnosis of the Lemonade case goes as follows: in Lemonade, A, in acquiring the belief error theory, is in much the same position as I am when on reflection I can’t locate any reason for believing the President is in Boston. In both cases, it looks as though one comes to believe¹⁶ that one lacks a reason for believing the target proposition, even though the target proposition would be reasonable to believe only if there is some evidence for it. In short, on this line of thinking, error theory is a ‘no reason’ defeater because believing it is tantamount to rejecting that you have a reason for believing the professor’s glass is full of lemonade.

The ‘no-reason’ diagnosis is, granted, an elegant way to think about the case. However, it is on closer consideration unworkable. The reason has to do with defeater defeaters, the acquisition of which neutralizes the original defeater and in the course of doing so causes the target belief to regain its previous epistemic status prior to having that status defeated.¹⁷

A defeater-defeater can be illustrated by a simple example. Let’s simply run a twist on our case used to illustrate a no-reason defeater. In short, I believe that the President is in Boston, but on reflection, I can’t locate any reason I have for believing this. Let’s now add to the story. Suppose that what accounts for why I actually hold the belief that the President is in Boston is that I overheard some people mentioning this. Because the circumstances under which I encoded this information were unremarkable (e.g., overheard on a train) I do not remember the reason for my belief when I attempt to locate it.

On the no-reason line, my inability to locate this reason defeats my justification for believing that the President is in Boston. As the recent epistemic situationist literature has indicated¹⁸, our cognitive abilities are to a significant extent influenced by environmental factors, including weather-induced negative mood¹⁹. Suppose I fly to Svalbard, and my mood deteriorates, improving my memory recall just enough so that I can now locate my reason for believing that the President is in Boston—viz., that I overheard it that morning on a train. Having now accessed this reason which was previously inaccessible to me, I acquire a defeater-defeater; it defeats my no-reason defeater. I no longer am such that I cannot locate a reason for my belief. The (previous) defeater is accordingly neutralized.

Let’s return now to Lemonade. If error theory were a no-reason defeater for lemonade, then, as such, it could itself be neutralized by the kinds of defeater-defeaters that characteristically defeat no-reason defeaters. But, as I want to suggest, error theory can’t in principle be defeated this way, i.e., via one’s coming to believe of some reason that that is one’s for believing the target proposition. Therefore, the conclusion we should draw is that A’s acquisition of the error theory belief cannot be a no-reason defeater.

To appreciate this point, the first step will be to run a ‘no-reason defeater’ variation on the original Lemonade case.

Lemonade*: A student, A, sees what appears to be a glass of lemonade sitting on his meta-ethics professor’s desk. Furthermore, A overhears other students talking about the lemonade on the professor’s desk, and even more, the professor announces that she is drinking lemonade. A accordingly forms the belief that there is lemonade on his professor’s desk. Call this belief lemonade. A then (due to a moment of confusion and blurred vision, due to anxiety) tries but fails to locate a reason for believing lemonade but cannot find one. The spell of anxiety subsides. A then reflects and is able to locate the various reasons he has for believing that there is a glass of lemonade on his professor’s desk.

In Lemonade*, A’s no-reason defeater for his justification for believing lemonade is defeated when, after the spell of anxiety subsides, A can locate several reasons A has for believing lemonade—viz., that A saw what appeared to be a glass of lemonade sitting on his professor’s desk, that A overheard other students talking about this, and that the professor herself announced that she is drinking lemonade. Any one of these reasons A locates suffices to defeat the no-reason defeater A acquired for lemonade when, during the anxiety spell, A tried but failed to locate a reason for believing lemonade. A thus regains his justification for believing lemonade.

But—and this is why A’s acquisition of the error theory belief in Lemonade cannot be a no-reason defeater—these kinds of defeater-defeaters cannot defeat A’s error theory belief. In order to bring this point into sharp relief, consider a final twist on Lemonade:

Lemonade**: A student, A, sees what appears to be a glass of lemonade sitting on his meta-ethics professor’s desk. Furthermore, A overhears other students talking about the lemonade on the professor’s desk, and even more, the professor announces that she is drinking lemonade. Over the course of the meta-ethics lecture, A comes to believe that that Jonas Olson’s arguments for epistemic error theory (as outlined during the class by the professor drinking the yellow looking drink) are persuasive, and thus, that there are no reasons for belief. A can on reflection locate what previously she regarded as her reasons for believing lemonade, viz., the perceptual evidence and the student and professor testimony.

Notice that, in Lemonade*, when A’s anxiety subsides and A subsequently recalls the perceptual and testimonial evidence A had acquired for believing lemonade, he was by reflecting on this able to defeat the no-reason defeater acquired during the anxiety attack. But reflecting on these things does not have the same kind of epistemic significance in Lemonade**. Having acquired (in Lemonade**) the belief that Jonas Olson’s arguments for epistemic error theory are persuasive, reflecting on such things as that A’s professor told him she was drinking lemonade does not by A’s own lights furnish A with a reason for believing there is lemonade in the glass.

Here, a quick parallel to scepticism is useful. Ordinarily, a hand-like appearance is evidence that there is a hand. However, as Crispin Wright (2007, pp. 26–27) puts it, if you replace the presumption that there is an external world with reason to believe you are instead a handless brain in a vat, whose experiences are being controlled by a highly sophisticated computer program, then the experience of seeing what appears to be a hand supports not the claim that there is a hand, but instead the claim that:

The computer is right now implementing a phase of its program which requires me to suffer the illusion of having a hand and holding it up in front of my face (2007, p. 27).

Likewise, if you replace the presumption that there are reasons for belief with the belief that (as the epistemic error-theorist tells us) there are not, then appreciating that you believe that the glass looks like it’s full of lemonade, and that a classmate and the professor herself told you it’s filled with lemonade, does not support your believing you have a reason for believing lemonade. Rather, it indicates to you that you have the sort of things that would ordinarily (by others) be called reasons to believe lemonade, but which you (taken in by the professor’s exposition of Olson’s arguments) now believe are not reasons to believe lemonade.

In short, no-reason defeaters are defeated by the production of reasons. A ’s error-theory belief however is not defeated by the production of the very kinds of reasons that defeat paradigmatic no-reason defeaters. And so A’s error-theory belief is not a no-reason defeater.

The line I want to advance, and which I will sharpen in §4, is that A’s acquiring the error theory belief in Lemonade is best understood as a meta-epistemic defeater. As such, it defeats meta-epistemic status, not first-order epistemic status, and it does so on a reflective level, and in a way that is subject to an altogether different form of defeat itself. I’ll unpack these ideas further. But first, it will be helpful to contrast the kind of error-theoretic defeater outlined in this section with a very different kind of meta-epistemic defeater, which is generated via the acquisition of beliefs about relativism and knowledge.

3 Relativist meta-epistemic defeaters

In a series of recent works, MacFarlane (2005, (2010, (2014) has argued that knowledge-ascribing sentences do not get their truth values absolutely, but only relatively²⁰. For example, suppose I say that “Rory knows that Paris is the capital of France.” Whether this is true, on MacFarlane’s relativist semantics for knowledge ascriptions, depends on the epistemic standards that feature in the assessor’s context, which is the context in which my knowledge ascription is being assessed as true or false. My knowledge-ascription is thus what MacFarlane calls assessment-sensitive.

Of course, when I say “Rory knows that Paris is the capital of France,” this can be assessed for truth or falsity from indefinitely many perspectives. And so what this means is that the sentence I utter when I say “Rory knows that Paris is the capital of France” does not get a truth-value absolutely, only relatively.

MacFarlane’s overarching argument for a relativist treatment of knowledge ascriptions is that, in short, it better explains our patterns of using “knows” than do other competitor views—e.g., traditional (insensitive) invariantism²¹, attributor contextualism²², and subject-sensitive invariantism²³.

Again, the details of MacFarlane’s argument won’t concern us here. What matters (for our purposes) is that this position is on the table, and that there are established arguments for the view.

Consider now, the following case:

Oven: A and B have just left for a holiday in Timbuktu. B asks A if A left the oven on. A says no, that he distinctly remembers turning it off. “I know the oven is turned off” he says, emphatically. B persists to doubt, as B did not see A turn the oven off. A thinks B’s doubts are foolish. Because it’s a long way to Timbuktu, and the conversation about the oven is stressful, A tries to lighten the mood by putting on an audiobook of  MacFarlane’s  (2014) monograph, defending a truth-relativist semantics for a range of different expressions, including for knowledge-attributions. A is impressed with MacFarlane’s careful arguments, and comes to believe that MacFarlane is right that “knows” is an assessment-sensitive expression, and so that knowledge-attributions do not get absolute truth-values.

In Oven, would we expect A to continue to purport to know the oven was turned off after acquiring the belief—call it relativism—that a truth-relativist semantics for knowledge attributions is plausible? Most likely. After all, MacFarlane’s semantics predicts that A will claim knowledge here. After all, from A’s perspective, where the context of use and the context of assessment are the same²⁴, A plausibly satisfies the relevant epistemic standard for knowing. Positioned within the framework of MacFarlane’s (2014) most recent presentation of the view, what this means is that A can rule out the alternatives relevant at the context from which he is assessing the knowledge claim for truth or falsity. And after listening to the MacFarlane audiobook, A remains able to rule out the same alternatives as before. However, there is an important respect in which, at least from A’s own perspective, his knowledge claim cannot be put forward with the same kind of authority as before²⁵.

In Oven, A’s acquisition of the relativism belief is best understood as a meta-epistemic defeater vis-à-vis A’s claim to know that the oven is turned off. It defeats—in a way I will articulate in more detail in the next section—the meta-epistemic status of A’s knowledge claim by undermining the reasonableness of A to continuing to advancing the claim as objective. It does so much as, in the previous section, A’s coming to believe error theory, in the Lemonade case, undermines the reasonableness of A’s regarding the evidence she locates as reasons for believing lemonade.

First, though, I want to consider two competing diagnoses of the situation. The first draws from Martin Kusch’s discussions of metaepistemology, and the second suggests that, if relativism is a defeater at all for A vis-à-vis A’s claimed knowledge that the oven is off, it is merely a first-order rebutting defeater.

Let’s consider first the argument that draws from Kusch (2010). Take as a starting point that that the acquisition of relativism for A doesn’t undermine the reasonableness of A advancing his knowledge claim as objective unless A already regarded, or took for granted, A’s knowledge claim as objective in the first place²⁶. Now Kusch denies the point Boghossian makes, and which I’ve argued for elsewhere²⁷, that ordinarily we take for granted that epistemic evaluations are being advanced as objective. According to Kusch, in ordinary epistemic discourse—i.e., outside of the philosophy classroom where meta-epistemic debates are taking place—individuals’ knowledge and justification assessments are made without any commitments one way or another to relativism or absolutism, or any other meta-epistemic position²⁸. To the extent that this line is right, we would have cause to wonder how A’s acquiring the belief relativism could undermine the reasonableness of A’s putting forward her knowledge claim as objective when we have no reason (absent positive reasons to think A has prior commitments in metaepistemology) to think A would be putting forward her knowledge claim as objective in the first place.

It’s debatable whether ordinary folk incur meta-epistemic commitments, not by articulating them explicitly (which ordinary folk do not do), but by revealing such commitments in other ways²⁹. But, even if Kusch’s point is granted, it can be pointed out that at least some individuals, even if not ordinary folk, do intend to put forward knowledge claims as objective. The thought that acquiring the relativism belief couldn’t undermine the reasonableness of putting forward a knowledge claim as objective is, for such individuals at least, by the by.

The second prima facie objection to the suggestion that, in Oven, A’s acquiring the relativism belief is best understood as a meta-epistemic defeater is that, by acquiring the relativism belief, A merely incurs a traditional rebutting defeater for A’s belief that he knows the oven is turned off. A rebutting defeater, recall, defeats first-order epistemic status by indicating that the target belief is false—i.e., by supporting its denial. There’s at least one indirect way of making the argument that when A comes to believe relativism, A acquires a rebutting defeater for A’s would-be knowledge that the oven is off in virtue of acquiring evidence that A does not know the oven is not off. Here is the reasoning: (i) Knowledge requires truth³⁰. (ii) A’s evidence for relativism is evidence that B’s assessment that A does not know that the oven is off, is true (as this is what MacFarlane’s semantics predicts). (iii) Given that knowledge implies truth, evidence for relativism is thus evidence that (assuming bivalence³¹) it’s not the case that the oven is off. Thus, (again, given that knowledge implies truth) evidence for relativism is evidence that A does not know that the oven is off.

The problem with the above line of reasoning is that it would be viable only if we unpack the first premise—that knowledge implies truth—as the objectivist unpacks it, but not as the relativist unpacks it. The relativist unpacks the slogan that knowledge implies truth differently from the objectivist, so that from “A knows that p” we cannot infer that “p is true for B”³². An argument aimed at establishing whether A’s acquiring the belief relativism is a rebutting defeater for A’s knowledge that the oven is turned off should be in principle neutral with respect to whether relativism is true or false; it shouldn’t require that the view be false (by depending on unpacking the slogan ‘knowledge requires truth’ in a way that is off limits to the relativist). Thus, the diagnostic story on which Oven is best interpreted as one where A acquires (by coming to believe relativism) merely a first-order rebutting defeater is not compelling. It is a story which, in short, begs the question against the relativist.

Yet, after A acquires the relativism belief, A really does incur some kind of epistemic loss vis-à-vis A’s purported knowledge that the oven is turned off, provided (to control for Kusch’s objection) that we suppose A is among those who had previously put forward the claim as objective. A’s acquired, on the view I’ll now turn to clarifying, a meta-epistemic defeater—the acquisition of the relativism belief is significant vis-à-vis the epistemic status of A’s claim to know the oven is off. But the significance is best appreciated at the second-order, not the first.

4 Meta-epistemic defeat

I’ve indicated in Sects. 2 and 3 that in cases like Lemonade and Oven, we have circumstances where an individual acquires a new belief which seems to downgrade the epistemic status of the target belief in each case, though not in the way that first-order defeaters downgrade epistemic status. I’ve indicated that the right way to think of these cases is in terms of meta-epistemic defeat. In this section, I want to consolidate some of the key features of meta-epistemic defeaters, sharpen their aspects of contrast with first-order defeaters, and in the process raise some new puzzles that I think future work on meta-epistemic defeat needs to address.

It will be helpful to begin by noting one very crucial difference between traditional first-order defeaters and the kind of meta-epistemic defeaters considered in Sects. 2 and 3, as this difference is relevant to how we should articulate a key point about the mechanisms of meta-epistemic defeat. To a first approximation, while both first-order and meta-epistemic defeaters defeat epistemic status (construed generally), meta-epistemic defeaters do so in a way that occurs entirely at the reflective level, but not the object level. For a given belief, p, facts about what other beliefs I have can cause first-order epistemic status simply not to obtain, vis-à-vis, p. Second-order status is different. The matter of whether this status actually obtains it is not sensitive to what I believe. And so therefore, lest we be radical constructivists about meta-epistemic status, for any belief q that I might form, if q in some way defeats the meta-epistemic status M of a target belief, p, it will not do so by actually causing p to lack M ³³.

But, as I’ve indicated in Sects. 2 and 3, there remains an important sense in which the overall epistemic status of certain particular beliefs is sensitive to the acquisition of new meta-epistemic beliefs I acquire, such as error theory and relativism. And this is so even though the acquisition of such beliefs does not defeat first-order status, nor does it cause second-order status to fail to obtain given that (again, lest we be radical constructivists) the obtaining of second-order status is insensitive to our acquisition of particular beliefs about second-order status.

The right way to make sense of the mechanisms by which meta-epistemic defeaters (like error theory and relativism) defeat the meta-epistemic status of a given belief is as follows: for some individual S, and first-order target belief, F, S’s acquisition of a meta-epistemic defeater M defeats the meta-epistemic status, $\alpha$, of F not by causing $\alpha$ to fail to obtain, with respect to F, but by reflectively undermining for S, $\alpha$, with respect to F. $\alpha$ is reflectively undermined for S, with respect to F, when $\alpha$ is caused to be rationally unavailable as a first-order mode of presentation, with respect to F, for S.

A mode of presentation (in the Fregean sense) is a way of thinking³⁴. One notable way of thinking about something is as oneself, as in the case of de se attitudes. When I think of something under this mode, certain dispositions follow, e.g., I am inclined to use first-person pronouns³⁵. Another much discussed mode of presentation is practical. Stanley and Williamson (2001), in defending the thesis that knowledge-how is a kind of propositional knowledge under a practical mode of presentation, put it this way:

Thinking of a person as oneself entails being disposed to behave in certain ways, or form certain beliefs, given relevant input from that person. Similarly, thinking of a place as here entails being disposed to behave in certain ways, or form certain beliefs, given relevant input from that place. Analogously, thinking of a way under a practical mode of presentation undoubtedly entails the possession of certain complex dispositions. It is for this reason that there are intricate connections between knowing-how and dispositional states. (2001, p. 429)

Likewise, thinking of one’s first-order epistemic status as objective plausibly entails certain complex dispositions. In the case of Oven, one such disposition (on the part of A) will be to regard B as making a mistake when B says that A does not know that the oven is off. Thinking about one’s knowledge claim as objective will also entail a disposition to regard things like local social contingencies and personal knowledge-standards to be irrelevant to discovering the truth of one’s knowledge claim³⁶.

I’ve said that meta-epistemic defeaters defeat (in their own idiosyncratic way) by reflectively undermining meta-epistemic status, and further that meta-epistemic status is reflectively undermined when is rationally unavailable as a first-order mode of presentation. This is not to say that it is not available at all as a first-order mode of presentation. For example, in the case of Oven, even after A comes to believe that MacFarlane’s arguments for a relativist treatment of “knows” are persuasive, A is nonetheless capable of continuing to think of A’s knowledge claim—i.e., that A knows the oven is turned off—under an objective mode of presentation, and so can potentially continue to manifest the complex dispositions that are entailed by thinking of A’s knowledge claim under an objective mode. But doing so is no longer rational by A’s own lights, given A’s having the belief relativism ³⁷. This is why being reflectively undermined is articulated in terms of a mode of presentation being rationally unavailable (as opposed to merely unavailable). The epistemic loss one incurs via meta-epistemic defeat is best understood in terms of this kind of rational unavailability of the objective mode, and hence not (implausibly) in terms of the non-obtaining of the objectivity of the first-order knowledge-claim.

I now turn to a question that was raised in Sect. 2, in the discussion of error-theoretic defeaters, though never satisfactorily resolved. In Sect. 2, it was shown that error-theoretic defeaters are not defeated by the production of the very kinds of reasons that defeat paradigmatic no-reason defeaters. And so the conclusion drawn in that section was that error-theoretic defeaters are thereby not no-reason defeaters.

That point was meant to dispel the idea that cases like Lemonade can be accounted for in terms of traditional (first-order) mechanisms of defeat. But this invites a more general question. Can meta-epistemic defeaters be defeated, and if so, how? As it turns out, this question generates some interesting puzzles, the solutions to which might—potentially, at least—be special cases of more general solutions to certain problems in the epistemology of disagreement (a point, I’ll return to).

Let’s begin by running an idealization on the Lemonade case, focusing on structure. Suppose A believes that p, on the basis of reason R. A then comes to acquire a belief, q, where q is the belief that epistemic error theory is true. On the diagnosis I’m offering, articulated in the language introduced in this section, A’s acquisition of q is a meta-epistemic defeater, vis-à-vis the meta-epistemic status of p, for A, by (in short) reflectively undermining its meta-epistemic status for A, by rendering rationally unavailable for A a realist mode of presentation of this epistemic status (i.e., that it is based on a reason) of A’s belief that p ³⁸. Now, how might this meta-epistemic defeater itself be defeated?

The most natural candidate answer is to suppose that A could accomplish this through the acquisition of a further belief q*, where q* is just the denial of error-theory. After all, in Lemonade, if I acquire beliefs that mountain dew and hallucination are false, both of these new beliefs are paradigmatic defeater defeaters for mountain dew and hallucination. So wouldn’t the acquisition of the belief that error theory is false suffice to defeat error theory in its capacity as a meta-epistemic defeater?

Perhaps not. In order to appreciate this point, first consider that various epistemically-irrelevant psychological factors could cause you to lose your error-theory belief, where these causal facts are not rational descriptions of how the defeater is defeated³⁹. But once the aim in play is clarified as that of giving a rational description of how a meta-epistemic defeater might itself be defeated, it becomes much less straightforward to see how the acquisition of a belief that is the negation of the meta-epistemic defeater could itself defeat the meta-epistemic defeater in question⁴⁰—at least in the error-theoretic and relativist cases envisioned.

Consider, after all, once you believe error-theory, it’s rationally unavailable for you to think of your newly acquired reasons for believing the denial of error theory as reasons for believing the denial of error theory, just as it is rationally unavailable for you to regard your reasons for believing there is lemonade in the professor’s glass as reasons for believing there is lemonade in the professor’s glass. But then, it’s not clear how, even if you come (on the basis of believing a new proposition, which is the denial of error theory) to no longer believe the error theory belief, that your doing so qualifies as rational by your own lights. Accordingly, it’s not clear how the acquisition of a belief that is the negation of error theory would suffice to defeat error-theory in its capacity as a meta-epistemic defeater. I’m using the case of an illustrative example, though a parallel kind of problem arises in the case of Oven, with respect to the matter of how to rationally defeat the belief that knowledge-ascriptions are relative, by coming to form a new belief which by your own lights could aspire only to relative knowledge⁴¹.

I have, to be clear, not suggested that meta-epistemic defeaters are indefeasible. Surely meta-epistemic defeaters can be defeated. More weakly, I’ve suggested that the mechanisms by which meta-epistemic defeaters stand subject to defeat themselves is not straightforward, and in an important respect, puzzling. Here, I think, is the source of conundrum: the acquisition of a new meta-epistemic belief (such as error theory and relativism) can reflectively undermine (in the sense I’ve articulated) not merely the meta-epistemic status of beliefs we already have. But, moreover, the acquisition of such a belief can influence which modes of presentation are rationally available and which are rationally unavailable vis-à-vis new beliefs we might accept. And this is why defeating a meta-epistemic defeater isn’t as simple as just coming to believe its negation, at least, in so far as we’re interested in how meta-epistemic defeaters could be rationally (by the lights of the agent) defeated or neutralised.

To the extent that the general account of meta-epistemic defeat I’ve proposed is on the right track, the matter of articulating the mechanisms by which meta-epistemic defeaters are themselves defeated raises important questions that work on meta-epistemic defeat needs to address. I want to conclude by gesturing toward a potential way forward, by drawing some parallels with some of the literature on disagreement, and in particular, cases featuring deep disagreement.

Let’s call a disagreement ‘deep’ if it has the following features: (i) there is a disagreement about the truth of some target proposition, p; and (ii) there is a disagreement about what kind of epistemic standards are relevant to adjudicating whether p ⁴². One very famous such disagreement is the early 17th century dispute between Galileo and Cardinal Bellarmine. The “p” in this dispute can be represented as the claim that heliocentrism is true (a claim Galileo accepted and Bellarmine denied). What’s epistemologically noteworthy about this case is that, given Galileo’s and Bellarmine’s very different background beliefs about what kinds of evidence are relevant to adjudicating matters of the cosmos (viz., scientific or scriptural), it looked to one another, and to some commentators, that neither was in a position to rationally persuade the other⁴³. In short, neither was in a position to think of what the other regarded as evidence (vis-à-vis p) as evidence⁴⁴.

The matter of whether this kind of deep dispute could potentially be rationally resolvable has been a notable point of contention⁴⁵. For our purposes, we needn’t take a stand. What’s relevant is that the philosophical problem of determining how meta-epistemic defeaters might themselves be defeated has features which closely mimic some of the well-established difficulties that feature in cases of deep disagreements.

In particular, consider that—in Bellarmine vs. Galileo—neither was in a position to think of what the other regarded as evidence (vis-à-vis p) as evidence. This can be appreciated as a social-epistemic variant of our own problem, of accounting how, for a given individual, a meta-epistemic defeater might itself be defeated through the acquisition of new beliefs. In both cases, the acquisition of such beliefs (which seem like they would be best candidates to, respectively, induce rational persuasion, and to rationally defeat a meta-epistemic defeater) are reflectively unavailable under the kind of mode of presentation which would make these objectives possible, in light of the agents’ background beliefs.

In sum, a satisfying account of how rational persuasion might be possible in cases of deep disagreements would very plausibly shine light on how we could model the defeasibility of meta-epistemic defeaters. I indicated that a solution to the latter might in fact be a special case of the solution to the former, which is why I think future work on meta-epistemic defeat would benefit from interfacing with social-epistemic debates about disagreement.

5 Conclusion

In this paper, I’ve attempted to locate a kind of epistemic defeat—i.e., meta-epistemic defeat—which is epistemically interesting in its own right, and which can downgrade epistemic status through mechanisms that are importantly different from more traditional kinds of epistemic defeaters. Meta-epistemic defeaters reflectively undermine meta-epistemic status, by rendering such status rationally unavailable as a first-order mode of presentation. Further, it’s been shown that meta-epistemic defeaters raise difficult puzzles when it comes to accounting for how meta-epistemic defeaters themselves can be defeated. I concluded by gesturing toward a kind of epistemic parallel between this problem and a related social-epistemic problem in the literature on deep disagreement; future work on meta-epistemic defeat could accordingly benefit by exploring further points of connection between meta-epistemic defeat and deep disagreements in social epistemology.