Anti-Skepticism Under a Linguistic Guise

Abstract

In this paper I consider the plausibility of developing anti-skepticism by framing the question in linguistic terms: instead of asking whether we know, we ask what falls within the extension of the word “know”. I first trace two previous attempts to develop anti-skepticism in this way, from Austin (particularly as presented by Kaplan) and from epistemic contextualism, and I present reasons to think that both approaches are unsuccessful. I then focus on a recently popular attempt to develop anti-skepticism from the “function-first” approach associated with Craig, which I also show to be problematic. I then argue that the apparent prima facie plausibility of fighting skepticism on linguistic grounds is due to a methodological spill-over from linguistics. Once we recognize this, it becomes clear that the skepticism debate should not be conducted in linguistic terms.

1 The Initial Thought

Skepticism is the view that we know nothing or very little in some domain. For the purposes of this paper, that domain will be the external world. Anti-skepticism is the rejection of skepticism: we know plenty of things about the external world. Questions regarding the truth of skepticism are sometimes phrased in linguistic terms. Rather than asking whether we know anything or much, we can instead ask whether the extension of “know” (perhaps understood as pairs of agents and propositions) is non-empty or very sparse. When phrased in terms of the semantic (truth-conditional) properties of the word, this is known as semantic ascent. As we will see, we can identify in the history of anti-skeptical thought several attempts at defending anti-skepticism by framing the motivating question in linguistic terms (whether semantic or otherwise). Dogramaci (2019) notes that there is a prima facie plausibility to this strategy:

The point I want to make right now is a psychological point. I think it’s simply a lot easier, psychologically, to appreciate how anti-skepticism could be true when we formulate it semantically. All it takes for anti-skepticism to be true is for our epistemic terms to have picked up anti-skeptical extensions. […] Sure, maybe it could be that we’ve ended up using words with skeptical extensions, but, equally surely it would seem, it could be that we are, and have been all along, using our words with anti-skeptical extensions. Why not? Can’t some words have anti-skeptical extensions, and why can’t our words have them? (Dogramaci, 2019, pp. 880–1)

I agree with this point – there does seem to be something initially more plausible about an anti-skeptical outcome when our motivating question is framed linguistically – about our knowledge terms rather than about knowledge.¹ Let’s call this the Initial Thought. The aim of this paper is to identify why this would be the case and whether it can form the basis of a successful anti-skeptical position.² The reason for doing so is that although I take the Initial Thought to have served as a motivation for a number of different anti-skeptical positions, I do not think the true explanation behind the Initial Thought has yet received sufficient attention. However, in terms of the prospects for a linguistic approach to anti-skepticism, my conclusion will be negative: once the explanation for the Initial Thought is given, we see that it does not provide a good basis for anti-skepticism.

The plan for the paper is as follows. In the next section I will partially develop the kind of skepticism that is at issue, considering both its subject matter (i.e. the kind of epistemic state that it is claimed we lack) and its motivation. In Sects. 3 and 4, I will present two previous attempts to develop an anti-skeptical position via the Initial Thought: in Austin’s anti-skepticism (particularly as presented by Kaplan (2008, 2018)) and in the contextualist response to skepticism. In doing so, I will diagnose where exactly such attempts to build an anti-skepticism falter. In Sect. 5, I will then turn to the recently popular “functional turn” developed by Craig and others (Hannon, Dogramaci etc.). Focusing particularly on Hannon’s (2019a) recent attempt to develop anti-skepticism, I will argue that this approach is also problematic. In Sect. 6, I will then introduce a new explanation for the Initial Thought – that in framing the question linguistically, our implicit methodological assumptions shift towards the kind of approach typically found in linguistics, and that this effectively turns the dial further in favour of anti-skepticism (although not conclusively). Nevertheless, I will argue that this methodological shift does not provide an appropriate basis for anti-skepticism. Thus I will conclude that the Initial Thought should effectively be ignored when considering skepticism.

2 Cartesian Full-Blooded Skepticism

Before an anti-skeptical position can be developed, we need some idea of the kind of skeptical position that is being rejected. I will not seek to develop any novel form of skeptical argument, but will instead just identify the form of skepticism of concern in terms of its subject (i.e. the kind of epistemic state it is skeptical about) and in terms of its motivation. Regarding the latter, a useful initial distinction can be made between Agrippan skepticism – which attempts to establish a regress in our chains of justification – and Cartesian skepticism – which appeals to skeptical scenarios (demons, dreams, brains in vats) in order to show that our beliefs are inadequately supported by whatever grounds we have available (Williams 2008, p. 22).

This paper is focused only on Cartesian skepticism. There are a number of ways in which Cartesian skepticism has been argued for. One way is to claim that in order for a belief that p to count as knowledge, the subject must be infallible regarding p. Skeptical scenarios are then used to show that we never stand in such a relation regarding propositions about the external world. Call this form of skepticism high standards skepticism (Kornblith, 2000). As both Kornblith and Pasnau (2015) note, however, a natural response to this kind of skeptic is simply: what does it matter if we fail to meet a perfect standard? While we may not achieve a perfect standard in the epistemic quality of our beliefs, this form of skepticism seems consistent with the idea that our beliefs are of sufficient quality to meet some lower standard, and we may then wonder whether it is a lower fallibilist standard that is reflected in our ordinary practice. This form of skepticism is then vulnerable to the challenge that it has mischaracterized the knowledge standard as unreasonably high. For this reason, it is common to put high standards skepticism to one side and consider whether there are alternative Cartesian skeptical arguments that may spell greater trouble for the prospects of knowledge about the external world. Indeed, both Kornblith and Pasnau have argued that the interesting form of Cartesian skepticism is not the high standards skeptic; it is (what Kornblith calls) the full-blooded skeptic who claims that we lack knowledge because we possess no justification (reasons, evidence, warrant, etc.) for our beliefs whatsoever. For the full-blooded skeptic, skeptical scenarios are used as part of an underdetermination argument to show that our beliefs are completely lacking in justification (Pritchard, 2021). How exactly this form of underdetermination argument is supposed to proceed is itself a matter of controversy. Following Nozick (1981) and others, many take an appeal to some form of closure principle to be the strongest form of skeptical argument. Others, such as Pryor (2000), have criticized the closure-based approach and presented alternative forms of underdetermination argument. For the purposes of this paper, the details of the argument in favour of full-blooded skepticism will not prove too important. What is important is the idea that skeptical scenarios play a crucial role in the argument for Cartesian skepticism, and that the skeptical conclusion is not that our beliefs may enjoy some justification that falls short of an infallibilist standard, but that our beliefs possess no justification whatsoever.

As mentioned, it is also important to be clear on the epistemic state that the skeptic claims we lack. In the first instance, our interest here will be in skepticism about knowledge. Of course, as we have just seen, in considering the full-blooded skeptic, we are considering a form of argument that runs monosyllabically as: no J, thus no K. But it is nevertheless important that we take our primary opponent as the skeptic over knowledge because the kind of anti-skeptical positions that we will consider make appeal to our ordinary practices of knowledge attribution and denial. In doing so, they can be seen as replying in the first instance to skepticism about knowledge.

To summarize, the skeptic we have in mind is focused on whether we can know anything about the external world. They claim that we cannot know because we have no reason or justification for believing what we do. And while we will not detail the skeptical argument in full, we can say that they make use of skeptical scenarios as part of an underdetermination argument.

3 Austin and the Skeptic

We are now in a position to consider how an anti-skeptical position could be developed that uses the Initial Thought as its basis. Perhaps the most famous anti-skeptical position of this kind can be found in Austin’s “Other Minds” (1979). Austin develops his anti-skepticism via an investigation of our ordinary linguistic practices, particularly with regard to our ordinary use of the question “how do you know?” At the heart of Austin’s position is a form of relevant alternatives view, understood as the claim that in order to know p, one must be able to rule out the relevant alternatives to p but does not need to rule out irrelevant alternatives. He considers in which circumstances it would be appropriate to say “but that’s not enough” in response to someone’s reasons for believing what they do. He notes first that in making this challenge there must be some determinate alternative that you have in mind to be ruled out, and second that you are not free to select any alternative you please (Austin, 1979, p. 84). It is this second condition that the skeptic falls afoul of, according to our ordinary practices. Dreaming hypotheses, brains-in-vats, and deceptive demons are, like the stuffed bird possibilities Austin considers, alternatives that we simply do not entertain in ordinary life and indeed it would be considered conversationally odd to raise.

This relevant alternatives approach can be challenged in a few different ways. First, any relevant alternatives approach owes some account of relevance that goes beyond gesturing towards the alternatives that those we take to be knowers can in fact rule out. Otherwise, we don’t seem to get far beyond a mere description of our knowledge intuitions in particular cases.³ Secondly, and more importantly for our purposes here, we might wonder whether the skeptic would care about what our ordinary linguistic practices suggest. Skeptical arguments don’t appear to be concerned with how epistemic agents ordinarily conduct themselves, they are concerned to establish that contrary to the way the ordinary person thinks or behaves, we in fact possess little or no knowledge. So isn’t this attempt to gesture towards our linguistic practices beside the point when considering skepticism?

Kaplan (2018) presents a response to this second worry on Austin’s behalf. On this view, philosophical methodology is already shot-through with appeal to ordinary practice. Any epistemologist that points to examples or counterexamples in the justification or rejection of a theory is already appealing to ordinary linguistic practices – they are appealing to what we would ordinarily say regarding whether someone knows.⁴ It seems that in order to even get started in considering what would or wouldn’t count as knowledge we need to appeal to what is widely or ordinarily thought about knowledge. Without that, it is hard to see how to proceed epistemologically without running the risk of any subsequent theory being idle insofar as it bears no relevance to our epistemic lives (Marušić, 2010). On the other hand, if we are to take our ordinary practices and our ordinary conception into account when producing a theory of knowledge, we need to take the full range of ordinary evidence into account. Austin’s point is that this includes the anti-skeptical practices that one clearly finds in ordinary conversation. So the dilemma for Austin’s opponent is as follows: we either do or do not take into account the ordinary epistemic practices when producing our theory of knowledge. If we do, this must include the anti-skeptical practices Austin highlights. If we don’t, our epistemological theory is robbed of a crucial source of evidence and looks impoverished as a result.

But even if we come this far with Austin, another famous objection looms, due to Stroud (1984, pp. 57–77). Stroud draws the distinction between an utterance being assertible and an utterance being true. We can understand this distinction in light of Gricean pragmatics, by distinguishing between the semantic matter of what is said in uttering a sentence and the pragmatic matter of what is conversationally implicated.⁵ In judging utterances, we don’t merely focus on whether what was said was true or false, we also focus on what was indirectly communicated. In the case of knowledge attributions, this opens up a possible skeptical-pragmatic account of our ordinary knowledge-attributing behaviour. Perhaps the skeptic is right that we rarely know anything, but in saying in ordinary contexts that we do know, we conversationally implicate something that is true. Of course, Grice’s own pragmatic theory and his particular conception of conversational implicatures has been the basis of a huge amount of critical discussion. But all Stroud requires here is some distinction between the semantic content of the sentence and the further pragmatic content that is communicated by uttering that sentence.⁶ The skeptic can then explain the anti-skeptical behaviour that Austin emphasises by claiming that we often attribute knowledge to one another in order to achieve some further pragmatic effect, perhaps to pragmatically communicate some further proposition. This kind of skeptical-pragmatic account of our ordinary linguistic practices has been defended by Unger (1975), who effectively argues that ordinary knowledge ascriptions can be viewed as cases of speaking loosely, in line with other absolute terms like “flat” or “straight”, and also Schaffer (2004), who previously argued that ordinary knowledge ascriptions can be viewed as instances of hyperbole.⁷ The very possibility of this kind of pragmatic explanation throws into question whether we should place much weight on the fact that our ordinary practices are anti-skeptical.

Kaplan (2008; 2018) responds to this worry on behalf of Austin. He argues that the cases where pragmatic considerations lead us to withhold from making some claim, or even to make some claim that is false, are cases that we are easily able to put to one side. In an earlier paper, he puts this in terms of speaking frankly:

The offending cases – the cases that no one can possibly want an epistemology to capture – are cases in which we dissemble. They are cases in which our knowledge (ignorance) attributions are, for reason of kindness or politeness, not frank. We need only say (on Austin’s behalf) that it is only to what we say and do and think right to say and do by way of frank knowledge (and ignorance) attributions that an epistemology must be faithful. Grice’s observations are thereby accommodated and disarmed. (Kaplan 2008, p. 360)

In his later book, he makes a similar point in different terms:

And so it is a completely ordinary thought that, if we are going to assess (for the purposes of crafting an epistemology that satisfies Austin’s requirement) whether a philosophical doctrine squares with what we would ordinarily say, we will need to set aside those things that we would (and wouldn’t) ordinarily say solely out of politeness, or kindness, or out of a desire not to mislead, and such.

 If that amounts to our needing to keep two sets of books, they are two sets of books we already keep in ordinary life. (Kaplan, 2018, p. 25)

However, the problem with this kind of response is that it puts far too much pressure on our ability as ordinary speakers to be able to distinguish between semantic and pragmatic phenomena. What exactly of a linguistic act is to be considered pragmatic rather than semantic is not a datum but is an outcome of substantive theorizing. Chomsky has given voice to this point well:

We may make an intuitive judgment that some more linguistic expression is odd or deviant. But we cannot in general know, pretheoretically, whether this deviance is a matter of syntax, semantics, pragmatics, belief, memory limitations, style, etc., or even whether these are appropriate categories for the interpretation of the judgment in question. It is an obvious and uncontroversial fact that informant judgments and other data do not fall neatly into clear categories: syntactic, semantic, etc. (Chomsky, 1977, p. 4)

We have seen Chomsky’s point play out in previous debates surrounding radical contextualism and what is said. For instance, pace Chomsky, Recanati has previously defended the following principle as a way of distinguishing what is said by an utterance from what is implicated:

Availability Principle: In deciding whether a pragmatically determined aspect of utterance meaning is part of what is said, that is, in making a decision concerning what is said, we should always try to preserve our pre-theoretic intuitions on the matter. (Recanati, 1989, p. 310)

Recanati, as a radical contextualist, would claim that what is said in uttering a sentence is often or typically enriched by certain pragmatic procedures, and that what we are aware of is the output of such procedures. Furthermore, as the above principle indicates, we are aware of the difference between what is said and what is implicated by uttering a sentence.⁸

It should first be noted that it is questionable whether anyone attempting to claim that we can gain access to the literal or standing meaning of our knowledge terms would do well to align themselves with a radical contextualist view like Recanati’s. A core part of Recanati’s view is that we don’t have conscious access to the semantic meanings of the terms used in an utterance; we have access instead to the output of pragmatic processes (e.g. saturation, enrichment, loosening) on the standing meanings of the expressions used. But even if we put this issue to one side, it is instructive to consider the criticism that Recanati’s Availability Principle has faced, for it illustrates the difficulty of relying on the ability of the ordinary speaker to intuitively distinguish between various levels of content (whether semantic or pragmatic). In particular, Carston (2002, pp. 167–170) has criticized the principle as insufficiently general and of no help particularly in the cases that are often of concern to semantic and pragmatic theorists. It is not hard to select well-discussed cases to illustrate this point:

• a) It is raining.

• b) Smith’s murderer is insane.

• c) John is a lion.

In (a), is an appeal to location part of what is said or what is implicated? Does the referential use of a definite description like (b) figure in what is said or what is implicated? The answers to these questions cannot be found in our pre-theoretical intuitions. Carston even reports considerable variation in the metaphorical uses of sentences such as (c), with there being considerable disagreement on whether what is said by an utterance of (c) would be true or false in an instance where John has been brave, majestic, powerful, etc. Whatever the correct treatment of these cases end up being, we cannot pretend that pre-theoretical intuitions provide the answer. Considering that there is a similar level of theoretical disagreement about the meaning of knowledge terms (and this is often reflected in the pre-theoretic judgments that one finds for instance in epistemology classes) we should not hope that the pragmatic content of knowledge attributions can be easily set to one side. While that is the case, Stroud’s skeptical defence against Austin remains viable. As such, the Austinian form of anti-skepticism remains unsuccessful.

4 Contextualism and Skepticism

A second form of anti-skepticism that is framed in linguistic terms is found in epistemic contextualism: the view that knowledge sentences vary in their meaning across different contexts of utterance. The basic contextualist account will be familiar to many and only requires a brief outline (Cohen 1988; DeRose, 1995; Lewis 1996). The contextualist claims that we very often speak truly when we attribute knowledge to one another in ordinary contexts. For instance, if I have been to the bank the week prior on a Saturday, that is a good basis to claim that I know that the bank is open on Saturdays. But what a knowledge claim means varies across contexts, and can be affected by one or both of the following: how important it is that you are right in your knowledge attribution and the conversational manoeuvres that have been made prior to the utterance. The effect can be that what it takes to truly attribute knowledge becomes a more demanding matter. For example, if my wife raises the possibility that banks do change their hours, and if it becomes important that the bank is open, then talk of knowing a proposition would then be talk of standing in a more demanding epistemic relation to that proposition.

One of the key supposed benefits of contextualism is that it is able to make sense of the apparent efficacy of skeptical arguments (particularly closure arguments) while also showing that knowledge attributions in ordinary contexts can nevertheless be true. According to the contextualist, the introduction of skeptical scenarios raises the epistemic standard that must be met to count as knowing. But in ordinary contexts where no such skeptical scenarios have been introduced the epistemic standard remains at a lower point such that it would be true to claim to know the kinds of things we typically do say we know.

I said in Sect. 2 that we are interested in full-blooded skepticism rather than high standards skepticism, as the latter is widely thought to be a relatively uninteresting position. However, Kornblith (2000) has argued that contextualism only serves to respond to high standards skepticism, leaving the more interesting full-blooded form untouched. The contextualist characterization of the skeptic is that they change the meaning of the word “know” such that it becomes incredibly demanding. However, this idea - that the plausibility of the skeptical view lies in the fact that the skeptic raises the epistemic standard to a point that our ordinary beliefs cannot meet - seems to allow that a belief that fails to meet this higher standard while still meeting some lesser, non-skeptical standard has some positive epistemic quality. Yet this is precisely what the full-blooded skeptic aims to deny. So the form of skepticism considered by the contextualist takes it for granted, pace full-blooded skepticism, that there is something epistemic to be said in favour of our ordinary beliefs.

Perhaps there are ways for the contextualist to respond here. For instance, perhaps a deeper form of contextualism (in the sense described by (Greco, 2017)) can respond to Kornblith’s objection. The idea of depth can be captured as follows. A shallow form of contextualism claims that knowledge attributions are context-sensitive, but claims that the underlying phenomena that knowledge attributions are used to talk about are invariant phenomena. For instance, a shallow contextualist that claims that justification is necessary for knowledge will claim that the level of justification a given belief has is a context-invariant matter, even if the threshold of justification required for knowledge is not. By contrast, a deeper form of contextualism will claim that knowledge attributions are context-sensitive because the extent to which a belief is justified is context-sensitive.

It may be that a deep contextualist solution can be given that accommodates the full-blooded skeptic in the following way. In skeptical contexts, the full-blooded skeptic is right that we don’t know anything and that nothing speaks in favour of our beliefs. However, this is because the notion of something speaking in favour of our beliefs is itself a context-sensitive notion. In other non-skeptical contexts, something may well speak in favour of our beliefs. Neta’s (2003) contextualism about evidence embodies this kind of deeper contextualism. He argues that within skeptical contexts, the notion of evidence that is in play is so restrictive that there is no evidence in favour of any of our beliefs. But in more ordinary contexts, a less restrictive notion of evidence is in play, such that one’s evidence base is wider. In this way, Neta’s contextualism could be seen as accommodating the full-blooded skeptic.

However, even a deeper form of contextualism could be criticized in terms of its treatment of the skeptic. First, one might worry that this deeper form still does not really address the full-blooded skeptic as originally conceived. After all, even if the notion of evidence in play in the current context is so restrictive that no beliefs bear evidence, there is still a distinction to be drawn in the current context between those beliefs that would count as having evidence in other contexts vs. those that would not. We may not be able to distinguish between these two sets of beliefs by appealing to whatever epistemic notion the contextualist claims is context-sensitive, but we can still say that the former set of beliefs are candidates for knowledge in other contexts, and this seems to be precisely what the full-blooded skeptic wants to deny. This speaks to a more general worry that many have had that the contextualist will inevitably mischaracterize the skeptic as making a weaker claim than they in fact make insofar as it is restricted to a skeptical context (Feldman, 2001).

Second, and more importantly for the purposes of this paper, even if this is a viable approach against the skeptic, it is not clear that this would be motivated via our ordinary linguistic practices and thus via the Initial Thought. Contextualism about knowledge has support in our ordinary knowledge-attributing practices, something which has been tested for experimentally (https://onlinelibrary.wiley.com/doi/full/10.1111/mila.12196). This is possible because knowledge talk is incredibly prevalent. On the other hand, notions like epistemic justification, evidence, epistemic reason, are far less prominent in ordinary discourse and so being able to justify such a view via appeal to our ordinary practices looks to be an uphill challenge. In this respect, we see an inherent limitation brought about by the Initial Thought insofar as it requires that we focus only on the epistemic practices of ordinary discourse. So perhaps the appeal to a context-sensitive standard could be part of a solution against the skeptic, but it looks doubtful that this kind of deeper context-sensitivity could be properly motivated by the Initial Thought and consideration of ordinary linguistic practices.⁹

5 The Functional Turn

A distinct approach to thinking about ordinary epistemic discourse can be traced back to Craig’s Knowledge and the State of Nature (1990). He argues that rather than engaging in conceptual analysis, a more fruitful way of investigating the concept of knowledge is to first consider what function or purpose the concept of knowledge plays in our epistemic community. If we didn’t have the concept, how would our lives be harder? Craig’s own view is that we have a need to identify reliable informants so that we can easily pool our true beliefs. The knowledge concept helps us meet this need by providing us with a way of identifying reliable informants. He then gives an account of how the knowledge concept would have developed into the knowledge concept we have today given the basic pressures that befall a concept performing this role.¹⁰ In turning focus to the function or purpose that a concept plays, Craig can be seen as initiating a functional turn in epistemology (McKenna, 2013), which many have since followed (Henderson, 2009; Grimm 2015; Dogramaci, 2019; Hannon, 2019b; Kusch and McKenna, 2020).

How could such an approach be used to defend a form of anti-skepticism? The general move will be to claim that once we take into account the function of the knowledge concept, we see that it cannot have a sparse or empty extension (Dogramaci, 2019; Hannon, 2019a). Craig himself doesn’t actually endorse anti-skepticism using the genealogical method. In a more conciliatory fashion, he instead attempts to show how both skepticism and anti-skepticism are motivated insofar as there are conflicting considerations in favour of both views that are grounded in the practical use of the knowledge concept. In doing so, he raises the following pragmatic reasons in favour of anti-skepticism:

i) The truth of skeptical scenarios would seem to make no practical difference to our lives. (Craig, 1990, p. 115)

ii) “Survival calls for action and action needs belief, so having false beliefs is no worse than not having any beliefs at all.” (Craig 1990, p. 118).

iii) There is no practical purpose in talking about an epistemic state no-one can have. (Craig, 1990, p. 116)

However, (i) and (ii) have little force. As Craig (1990, p. 115) notes in response to (i), while the skeptical hypotheses typically appealed to are designed such that they make no difference to our lives, it is not hard to conceive of many that would. For instance, consider the skeptical scenario that you are actually a BIV in a factory that is going out of business next year. As with the usual skeptical scenarios, you have no way now of ruling this out, but it will make a massive practical difference to your life if it is the case. Similarly, against (ii) the bare claim that having no beliefs is worse than having false beliefs is difficult to defend, as so much would depend on the practical situation involved in order to make any kind of assessment. Given that we are considering the possibility that the external world is radically different to the way it appears, for (ii) to be true it needs to be the case that we are not actually in a world where we benefit from ignorance and delusion.

Even (iii) is not without its problems. Note first that (iii) is a negative existential, which often come with a heavy burden of proof. Indeed, all the skeptic has to do is to provide some explanation as to why we would have a concept in circulation in our community that refers to a relation that no agents stand in. One natural explanation the skeptic could appeal to is that knowledge serves as a regulative ideal, that even if it ends up being the case that no-one knows anything, an epistemic community that consistently appeals to the notion, and even talks as if they realize the notion, is epistemically or practically better off than a community that does not. Gardiner (2015) briefly considers this possibility but dismisses it:

There is the role of picking out the mental state, evidence base, or epistemic relation to the world that is unobtainable for people in the human epistemic condition. There is flagging an ideal or perfect epistemic position. Perhaps there is the role of describing an epistemic haven that angels and gods can enjoy, but lesser beings could never enter. These kinds of functions for the concept of knowledge would result in a skeptical conclusion: we do not possess much, if any, knowledge. But these candidate roles for the concept of knowledge are not compelling; our behaviour and thought indicate the concept does not fulfil these roles. (Gardiner 2015, p. 42)

Merely appealing to our ordinary behaviour and thought as a way of dismissing this kind of skeptical position is unsatisfactory, however. Hannon (2019a, pp. 153–5) has developed two forms of response to this view of knowledge as a regulative ideal. First, he argues that this form of skeptic could only be a high standards skeptic, which we have already seen is widely thought to be an uninteresting view. This is because if knowledge is to serve as a regulative ideal then it must be something that we can approximate towards. To allow this much is to allow that our beliefs have some positive features insofar as they approximate towards the epistemic ideal in some relevant respect. In justificatory terms, the thought would be that knowledge serves as a regulative ideal in that it encourages us to form beliefs that approximate knowledge insofar as they have some level of justification, even if that level falls short of the level required for knowledge. But this account of knowledge serving the role of a regulative ideal clearly requires that our beliefs have some positive epistemic features, contrary to full-blooded skepticism.¹¹

However, this reasoning is problematic, and there are two kinds of response available to the functional skeptic (understood as the skeptic who claims that the function of the knowledge concept is that of a regulative ideal). First, we should note that nearly anything can be approximated towards in some loose sense. For instance, many take truth to be a notion that does not admit of degrees, and yet it is commonplace to talk of a claim being approximately true. Crucially, in order for some notion to be approximated towards, it is not required that the notion be partly constituted by some scalar element in the sense that this objection would require. France is approximately the shape of a hexagon, but it doesn’t follow from this that being hexagonal is a scalar notion. We are able to introduce scales on the fly with which we are able to make judgments of approximation in this manner. Something can also be approximated toward in one respect, but not in others. With this in mind, the functional skeptic could respond in the following way. While we have no justification for any of our beliefs about the external world, having knowledge serve as a regulative ideal brings other kinds of epistemic benefit insofar as our beliefs tend to approximate knowledge in respects other than justification. As an example, suppose that justification is necessary for knowledge and also that there is a minimal epistemic benefit to having one’s beliefs be consistent with one another. The skeptic could claim that having knowledge play the role it does in our epistemic community is a particularly effective way of encouraging consistency across a large number of beliefs. This would be a way of viewing knowledge as a regulative ideal insofar as we would be worse off if we did not have the knowledge concept playing the role it does while also maintaining full-blooded skepticism by claiming that, because we are unable to rule out skeptical scenarios, we completely lack justification for our beliefs.

A second way for the functional skeptic to respond would be to claim that knowledge does not really serve as an epistemic ideal – it does not have a positive epistemic effect on a community. It does, however, bring some other kind of benefit, such that we can make sense of why it enjoys wider circulation within a community. Perhaps having knowledge play the fictive role it does ends up being beneficial to a community’s survival. Perhaps acting as if we all do have a good grasp on what the external world is like regulates our behaviour. If I take myself to be in a community of knowers, I am less likely to try to deceive, for a start. So in thinking about the function of the knowledge concept, it must be acknowledged that the function may not be straightforwardly epistemic, and this is a point that the full-blooded skeptic could exploit.

The second objection that Hannon raises is that even if we allow that there is some idealized epistemic state that plays the role of a cognitive ideal, why would we think that this is identical to knowledge, as ordinarily spoken of? If we are envisioning some kind of epistemic ideal, why think knowledge has this exalted status? Here again I think the functional skeptic has responses available. In particular, one way (perhaps the only way) for an idealized notion to have an effect on a community is for it to be in wider circulation. A notion in wider circulation is clearly going to have a greater influence than a notion that is not, even if the majority of things that are said and thought with regard to the notion are in fact false. In this respect, knowledge stands alone above all other epistemic notions as enjoying incredibly wide circulation in terms of the extent to which it is appealed to across discourse of all kinds.¹² It is then well-placed to play the role of a regulative ideal.

It is also important to keep in mind that if we do view knowledge as playing the role of an ideal, we are not thereby committed to the idea that it is an epistemically perfect state. Knowledge may stand as a kind of ideal given certain limitations that we possess as humans, and in envisioning epistemic agents of other kinds, we could perhaps envision states of even greater epistemic value than knowledge. That knowledge might be a regulative ideal while not being an epistemically perfect state can be made good sense of. An ideal that is of such perfection that it is too far removed from our epistemic situation will serve as a poor guide. Instead, knowledge may sit in a sweet spot of serving as an ideal by always encouraging us to approximate towards it while also not being too alien to our own epistemic situation so as to be no guide at all. So Hannon’s question – why, if we allow that there is an epistemic state that plays the role of a regulative ideal, should we think it is knowledge? - admits of a clear answer.

We find then that the functional turn does not, contrary to what has been recently claimed, provide a basis for anti-skepticism. This is simply because there are stories available to the skeptic as to the functional role of the knowledge concept even if it is something that we nearly never have. If the functional turn is the best way to understand the Initial Thought, we find that initial appearances are misleading: there is no clear route to anti-skepticism here. In fact, I think the Initial Thought has its roots elsewhere, and I’ll seek to develop this in the final section.

6 A Methodological spill-over

Let us recap. We have considered three ways in which the Initial Thought could provide the basis for a form of anti-skepticism. Austin makes sense of the Initial Thought by appealing to conversational manoeuvres that are felicitous in a discussion of our beliefs and their reasons. Similarly, contextualists make sense of the Initial Thought by appealing to the flexibility of our epistemic terms across different conversational contexts. Craigians make sense of the Initial Thought by focusing on the purpose of the knowledge concept. However, I have argued that each approach is unsuccessful. In this section, I will argue that while these three approaches have each attracted a great deal of attention in the epistemological literature, they do not quite capture why the Initial Thought – that anti-skepticism seems more plausible when the topic is set in linguistic terms – is true. In this final section, I want to outline the proper basis for the Initial Thought, which lies in a kind of methodological spill-over from linguistics. However, I will argue that this does not provide support for anti-skepticism. Ultimately, then, in responding to skeptical arguments, the Initial Thought should be ignored and the temptation to phrase the issue in linguistic terms should be resisted.

Philosophical questions often get phrased in linguistic terms. This is due to a combination of factors including implicit theoretical commitments, historical philosophical developments, as well as the fact that the philosophical contribution to linguistics has been significant (particularly in semantics and pragmatics). As a result, it is considered normal practice for philosophers to switch back-and-forth between object-level language and meta-language. The Initial Thought is just one example of this practice.

But this habit can hide the fact that there are really two domains of inquiry here and that there are methodological differences in the way that each should be approached. When a philosopher theorizes about knowledge, they are concerned with the nature of knowledge, whether we really have any knowledge, what value knowledge has etc. These considerations all feed in to a philosophical theory of knowledge. A linguist that is concerned with epistemic terms (e.g. “know”) does not draw upon the same considerations. Their primary concern is in communicative acts, and the role that the epistemic terms play across the set of communicative acts it appears in. We can start to drive a wedge between these two projects from the linguistic side, by considering the following quote from the linguist Emmon Bach:

Metaphysics I take to be the study of how things are. It deals with questions like these:

What is there?

What kinds of things are there and how are they related?

Weighty questions indeed, but no concern of mind as a linguist trying to understand natural language. Nevertheless, anyone who deals with the semantics of natural language is driven to ask questions that mimic those just given:

What do people talk as if there is?

What kinds of things and relations among them does one need in order to exhibit the structure of meanings that natural languages seem to have?

Questions of the latter sort lead us into natural language metaphysics.

(Bach, 1986, p. 573)

Bach’s primary concern here is with the kinds of entities that need to be posited by a semantic theory in order to capture the meaning facts of a language (e.g. individuals, truth values, worlds, functions, events, etc.). But in outlining the linguistic project in this way, he acknowledges that our best linguistic theory (e.g. our best semantic theory) may not have a metaphysics that is consistent with the actual metaphysics of the world. When it comes to knowledge and knowledge terms, the concern of the linguist is to capture what is communicated by using such terms, and they will draw upon available linguistic data (intuitions, experimental surveys, corpora, psychological findings) in order to do so. And as soon as we do start to prioritize what people talk as if there is in this way, we will find that people talk as if knowledge is widespread. As an analogy, consider how other concerns about the true state of the world will not enter as considerations into linguistic theorizing. For instance, the weight of an object is relative to the strength of the gravitational field. Does it follow that the word “weight” has an implicit parameter for gravitational field that is assigned a value when the word is used? It does not. Similarly, the theoretical considerations that lead us to think that we rarely, if ever, know anything about the external world, are considerations that do not plausibly enter as considerations in linguistic theorizing.

Along similar lines, Chalmers notes the following differences in methodology between semantics and philosophy:

Where semantic premises deriving from linguistics are concerned, it is especially relevant to observe that linguists’ semantic claims are often subject to fewer constraints than philosophers’ semantic claims. For example, linguists’ analyses often aim to capture inferences that are held to be valid by most or all competent users, whereas a philosopher might instead aim to capture inferences that are in fact valid. Likewise, a linguist might freely invoke abstract objects in their analyses, while some ontologically scrupulous philosophers might not. A philosopher might also give a role to empirical facts about science and naturalness in giving their analyses, while a linguist might not. (Chalmers 2012)

Consider also the following quote from Lønning in discussing the construction of model-theoretic semantics for mass nouns:

Our interest lies in the natural language itself, not in the world it describes. This means that the models we build are not necessarily “true” models of the physical world, but means to understand the language, in particular, to give valid forms to intuitive true sentences and inferences. (Lønning, 1987, p. 7)

To borrow a distinction from Chalmers, the linguist’s project does not attempt to provide worldly truth conditions, that is, truth conditions that are consistent with the way that the world actually is, given all of the truths that philosophy is usually sensitive to. Instead, truth conditions play a narrower role of capturing the relations between meaning and truth that speakers recognise as part of their competency in a language.

Given this account of the linguistic project, and of the role that truth conditions play within it, it is natural that the linguistic project is tilted towards an anti-skeptical outcome. If the explanandum of the linguistic theory is speaker competence of a language, with truth playing some role in that insofar as the linguistic intuitions generated by competent speakers are partly concerned with how the meaning of a sentence/utterance relates to the truth of a sentence/utterance, then it seems that the linguistic project will want to capture the anti-skeptical behaviours that are typical of a competent speaker. Here we start to see what is correct about the Austinian and contextualist approaches. Both place great focus on the intuitive or acceptable moves of the ordinary speaker.

We saw earlier that Austinian anti-skepticism faces Stroud’s objection from pragmatics – that it is theoretically possible that a term is often used to make claims that are false in order to generate some further pragmatic effect. This is also an issue that has arisen in the contextualist literature, with skeptical forms of invariantism (such as Unger’s or Schaffer’s) remaining a theoretical rival to the contextualist position.¹³ In principle this option remains on the table for the linguistic project. The linguist could attempt to adopt a skeptical semantics and explain our knowledge-attributing behaviour via further pragmatic effects. But the more important point here is that the motivating reasons behind the skeptical position do not enter into the linguist’s considerations when they are attempting to capture the meaning of an expression like “know” as part of an ordinary speaker’s competence. While it is a theoretical possibility that we have a skeptical semantics supplemented with a suitable pragmatics that explains the behaviour that Austin was interested in, within the linguistic project this proposal looks unmotivated and out of step with the way that the linguist is going to attribute truth conditions for related expressions.

So anti-skeptical extensions for epistemic terms look plausible under the linguistic project. Does this provide a basis for philosophical anti-skepticism? It does not. In drawing out this separation between the linguistic and the philosophical project – where the latter but not the former will take into account considerations to do with epistemic value, skeptical arguments, ontological commitments etc. – we have to acknowledge that the results of the two projects may not align. The Initial Thought does speak to something true in another domain – that the linguist’s project of capturing the proper use of knowledge terms is naturally tilted towards anti-skepticism insofar as our ordinary practice suggests it and further because reasons in favour of skepticism do not get considered. But this does not help us with the philosophical project of considering whether knowledge is something that we in fact have. In that respect, the Initial Thought should not be pursued, and the fight against skepticism should be taken up on other fronts.

Of course, in relying in this final section on a distinction between philosophical and linguistic methodology, it needs to be the case that there is a well-defined philosophical project to be pursued. One may wonder what else there is to investigate about knowledge once one has captured the linguistic properties of knowledge attributions. This is obviously not the place to develop a complete philosophical methodology, but it is important to emphasize that philosophical methodology plausibly overlaps with linguistic methodology insofar as the two projects share some of the same data. Our linguistic intuitions about knowledge attributions are a source of data for epistemology and linguistics. We can see this by the simple fact that it would be a benefit, other things being equal, that our epistemological theory captures the way in which we talk about knowledge.¹⁴ But epistemological theories are usually more ambitious than just trying to capture the way in which we in fact talk, and it is here that skeptical arguments can enter as something that we hope to capture, dissolve, or respond to. Other considerations plausibly enter too, such as epistemic value considerations or genealogical considerations.

To conclude, it is no wonder that many have been tempted to pursue the Initial Thought, for it speaks to something true in a project closely related to the epistemological project as traditionally understood. But the Initial Thought is made true not merely by the fact that our ordinary linguistic practices suggest an anti-skeptical extension for “know”, but also by the fact that in the linguistic domain of inquiry, skeptical considerations are not given any weight. Once we recognize this, we see that the Initial Thought does not provide a suitable basis to respond to skeptical argumentation.¹⁵