2. Fumerton on easy knowledge and externalism

Richard Fumerton brought the problem of easy knowledge into clear focus in his book Metaepistemology and Skepticism (Fumerton Reference Fumerton1995). There, Fumerton objects to reliabilism as a paradigmatically externalist theory as follows:

Fumerton's point is that reliabilism is problematic because it makes it too easy for us to come to know that perception is reliable. Why too easy? Fumerton says: because the reasoning employed is circular. Perception is used to confirm the reliability of perception. Here is a choice quote:

Fumerton, then, rejects the easy way in which reliabilism allows for knowledge of perceptual reliability because it is inflicted with an unacceptable kind of source circularity. I believe that Fumerton's objection to reliabilism is deep and important, but I don't agree with his diagnosis of why reliability knowledge cannot be acquired in the easy way reliabilism permits.

3. Cohen on easy knowledge and the KR principle

In “Basic Knowledge and the Problem of Easy Knowledge,” Stewart Cohen adds two important insights to our understanding of the issue.Footnote ³ First, the problem of easy knowledge afflicts not only externalist reliabilism, but also internalist theories such as James Pryor's dogmatism and Michael Huemer's phenomenal conservatism.Footnote ⁴ Second, the bootstrapping-type of reasoning Fumerton rejects as defective is just one type of easy knowledge. Another type is unacceptably easy knowledge that one's current perceptual experiences are not deceptive.

What licenses both bootstrapped reliability knowledge and easy knowledge of non-deception is the rejection of what Cohen calls the “KR principle”:

It is important to distinguish between the key idea underlying KR and specific versions of the principle. It seems to me that making the key idea explicit does not require articulating the principle in terms of knowledge. Those who endorse KR typically do so from an internalist point of view, according to which a belief source is a knowledge source by virtue of being a justification source. Thus we can move from KR to:

Furthermore, we might worry that KR is too demanding if the kind of knowledge that figures in KR's consequent is fully explicit, thus entailing occurrent belief. In ordinary life, perceptual experiences result in spontaneously formed perceptual beliefs, without being accompanied by occurrent beliefs about the reliability of these experiences. Therefore, KR_a implies that, when perceptual beliefs are spontaneously formed, perceptual experiences are not a justification source. The following modification of the consequent avoids this problem:

But what is it to be in a position to know? Different theories will offer opposing answers. From the internalist approach that motivates KR, what puts one into a position to have knowledge is the possession of evidence. Thus we arrive at:

A further question of interpretation arises when we ask exactly what the intended relation is between the antecedent and the consequent. On one approach, it is an entailment relation: necessarily, if the antecedent is true, then the consequent is true. I believe that interpretation fails to capture the key thought that motivates KR. When epistemologists advocate KR, the question they are concerned with is: When a belief source is a justification source for you, by virtue of what does it enjoy this status? What makes it a source of justification for you? Epistemologists who endorse the key thought motivating KR-type principles would answer this question thus: whenever a belief source, S, is a justification source for you, it has that status because of your reasons or evidence in support of the proposition that S is reliable. This evidence is what makes S a justification source for you.Footnote ⁶ Accordingly, I will understand KR_c as follows:

In what follows, I will understand the KR principle in just this way.Footnote ⁷ Obviously, there are additional ways of interpreting KR resulting from alternative construals of the epistemic statuses that feature in the principle's antecedent and consequent. Exploring this territory more deeply would be a topic for a different paper.

Is KR_d sufficiently plausible to spell trouble for theories that reject it? I believe it is. What motivates the principle is the basic thought that it's irrational to rely on a source without having reasons to trust it.Footnote ⁸ Both reliabilism and dogmatism reject that thought. According to these views, for a belief source to give you knowledge, you needn't know, nor be in a position to know, nor have any kind of justification at all for believing, that the source is reliable. That's why they face the problem of easy knowledge.

4. Bad deductions: easy knowledge of non-deception

Fumerton points out that reliabilism allows for knowledge of reliability that is suspiciously easy. Cohen expands Fumerton's point by showing that any theory that denies KR suffers from this problem, and from the additional problem of permitting surprisingly easy knowledge of non-deception. Let's see how, according to Cohen, denying KR allows for the latter kind of easy knowledge. The example Cohen describes features his son who wants to know whether the red-looking table he and his dad might purchase is really red. Reliabilism and dogmatism allow resolving this doubt in the following way. Suppose, first of all, the situation is entirely normal. The table is illuminated by daylight and looks red to Cohen and his son because their color vision works reliably and the table is in fact red. Now suppose reliabilism is true. Then Cohen and his son know the table is red because their color vision is reliable. Suppose dogmatism is true. Then Cohen and his son know the table is red because it looks red to them. In each case, they come to know that the table is red without needing any justification for believing that their visual experiences are reliable.Footnote ⁹ Equipped with such knowledge, they can deploy the following deduction:

What are the grounds for calling this deduction ‘bad’? After all, logically there is nothing wrong with it. And, ex hypothesi, the premises are true. The deduction, then, is both valid and sound. Therefore, the reason for rejecting it as bad is not that the argument is logically flawed or has a false premise. Rather, the point is that the deduction fails to do what arguments are supposed to do, namely either track pre-existing or generate new justification for accepting the conclusion. When an argument is presented to resolve a doubt or justify a claim, its job is to answer a certain question, namely: How do we know the claim in question? For a proffered argument to answer this question, its premises must track the evidence in support of the claim that has been questioned. The argument from (1) and (2) to (3), then, is supposed to identify the evidence on the basis of which Cohen and his son know that the red-looking table isn't white with red lights shining at it. If Cohen were to offer this argument as a response to his son's worry, he would be telling his son that knowledge of (3) comes about by deducing it from (1) and (2). The point of Cohen's example is that, when his son doesn't buy this argument, he recognizes that (3) is not known, and cannot come to be known, by deducing it from (1) and (2). Therefore, the argument fails to generate justification for its conclusion. That is the sense in which it is a ‘bad’ deduction.

Let's focus on easy knowledge reasoning in the hands of dogmatists. According to dogmatism, a perceptual experience as of p can by itself justify the belief that p. Its justificational force does not derive from justification for taking it to be reliable, or from justification for rejecting deceptive scenarios. So, if dogmatism is true, when the table looks red to Cohen and Cohen has no defeaters, Cohen thereby comes to know, solely on the basis of the table's looking red to him, that the table is red. Consequently, the Bad Deduction reduces in essence to the following bit of reasoning:

In the same vein, dogmatism licenses arguments like the following:

It's difficult to not have the feeling that these are epistemically defective arguments, not because of any logical defects but rather because they fail to track the grounds on which the propositions in questions are actually known. That is what makes them bad deductions. Dogmatism and reliabilism license bad deductions. Therefore, they face the problem of easy knowledge.Footnote ¹⁰

5. Why bad deductions are bad: transmission failure

Martin Davies and Crispin Wright have argued that bad deductions suffer from transmission failure: they fail to transmit justification from the premises to the conclusion.Footnote ¹¹ No Red Light fails in this regard because perceptual knowledge is not as easily acquired as dogmatism would have it. According to dogmatism, the table's looking red can by itself give you knowledge that the table is red. The competing, more demanding view says that the table's looking red can give you knowledge that the table is red only if you have, to begin with, justification for rejecting the white table and red lights alternative. Knowing the argument's first premise (that the table is red) requires prior justification for believing the argument's conclusion. No Red Light, then, fails to transmit justification, and thus fails to generate knowledge, because it reverses the direction of justification.

Consider Non-Envatment. If hand experiences are a source of hand knowledge for me, I must already have justification for rejecting any skeptical scenario in which I have hand experiences without having hands. Therefore, justification for the premise “I have hands” does not transmit to the conclusion “I'm not a BIV.” Instead, I know that premise only if I have, to begin with, justification for rejecting the BIV scenario. Analogous points apply to No Painted Mules and No Hallucination.

Regarding No Red Light, Sosa puts the point very nicely in terms of what one must presuppose if one's color experiences are to give one knowledge. He points out that, even if KR is false,

Next, Sosa emphasizes that this presupposition must not be arbitrary. It must be epistemically appropriate or justified. That's the rub, for then it follows that

This seems essentially right to me, although I myself would prefer to state the relevant point in terms of evidence: for visual experience to have justificational force for you, you must have evidence that the present conditions don't undermine the trustworthiness of your experience.

The key thought motivating the rejection of easy knowledge is that of epistemic priority. Perceptual knowledge has a directional structure: having justification for rejecting reliability-undermining scenarios is epistemically prior to, or a presupposition of, the acquisition of perceptual knowledge. Therefore, we cannot first acquire perceptual knowledge and then deduce that reliability-undermining deceptive scenarios don't obtain.

6. Attempts to defuse the easy knowledge challenge

Since dogmatists deny KR-type principles, they face the problem of easy knowledge. To defuse it, they have two options.Footnote ¹⁴ First, they could claim that bad deductions aren't really bad. Second, they could claim that, initial impressions notwithstanding, dogmatism does not actually license bad deductions. Since the bad deductions listed above are paradigmatic examples of begging the question, the first option doesn't seem very attractive. James Pryor and Peter Markie have suggested a hybrid strategy. In a context of doubt, the bad deductions are indeed bad because they beg the question. But that's merely a dialectical defect, a defect that is limited to the sphere of rational persuasion. Dogmatists, the response goes, can consistently concede this defect and nevertheless insist that the allegedly defective arguments can generate knowledge of their conclusions.Footnote ¹⁵

This response assumes that an argument's begging the question is one thing and its capacity to produce knowledge another. I don't buy this assumption. It seems to me that, when a proffered argument is rejected as question begging, the point is that the rejected argument fails to explain how the claim in question could be known. For the response to go through, the defect in question would have to be merely psychological: a failure to persuade someone who harbors doubt about the argument's conclusion. But when we consider the full range of bad deductions and find them wanting, the thought is not that what's wrong with these arguments is merely that they are impotent means of persuasion and thus ill suited to placate pesky doubters. Rather, the worry is that they all fail as accounts of how the conclusions in question are actually known.Footnote ¹⁶

7. The puzzle of reliability knowledge

Views that reject KR-type principles are problematic because they allow for knowledge that's too easy. But views that accept such principles are also problematic. To see why, let us focus on perception. Applied to perception, KR principles tells us that perceptual knowledge requires prior knowledge (or at least evidence) of perceptual reliability. The problem is that there is a second principle, at least as plausible as KR if not more, namely RK: knowledge of perceptual reliability requires prior perceptual knowledge. The point is obvious enough: unless it's possible to deduce perceptual reliability a priori, you have to use your perceptual faculties to find out whether perception is reliable. Now, it would seem that, if both KR and RK are true, neither perceptual knowledge nor knowledge of perceptual reliability is possible. To have one, you must have the other one first: a catch 22 situation. The puzzle, then, arises because KR and RK are both individually plausible while their conjunction seems to form a vicious circle. Spelled out, the puzzle takes the following form:

Here are the options for escaping the puzzle:

1. 1. Reject KR. That's what externalists and dogmatists recommend. To make this response palatable, they must find a way to defuse the charge of easy knowledge.

2. 2. Reject RK. For this move to be workable, there must be an a priori argument for the reliability of our perceptual faculties.Footnote ²⁰ It's not easy to see what that argument might be.

3. 3. Reject NP. Advocates of this option face the challenge of explaining why the conjunction of KR and RK does not constitute a vicious circle.

I believe that the third of these options is the most promising, and I'll provide some support for this view at the later stages of this paper.

8. Dogmatism, conservatism, and credentialism

The puzzle of reliability knowledge is closely related to three main options to consider when we ask what it takes for a perceptual experience as of p to be a reason – a source of justification – for believing p and thereby to be a source of knowledge of p.Footnote ²¹ We have already come across the first option:

Conservatives don't agree that perceptual experiences are a knowledge source no matter what. They say the following:

Conservatives agree with dogmatists that, for Ep to be a reason for you believing p, you need not have evidence that Ep is reliable. But they reject the following implication of dogmatism: Ep is a reason for believing p even if you have evidence that Ep is unreliable. Advocates of KR go even further. They reject the conservative claim that Ep is a reason for believing p even if you don't have evidence that Ep is reliable. Instead, they hold that, for Ep to be a knowledge source for you, you need to have positive evidence that Ep is reliable. Thus we arrive at the third view:

I label this view ‘credentialism’ because it tells us that a perceptual experience is a source of justification only if it is properly credentialed by evidence of its reliability.

Here is how these views relate to the puzzle of reliability knowledge. Dogmatists reject KR. They must somehow find a way to defuse the charge of easy knowledge. Credentialists accept KR. If they accept RK as well, they must explain why the conjunction of KR and RK is not an epistemic debacle. Vis-à-vis these two positions, conservatives enjoy a couple of advantages. Unlike dogmatists, they do not claim, implausibly, that perceptual experience is a justification source no matter what. Rather, they hold that, when a perceptual experience is undermined by evidence of reliability, it ceases to be a justification source. Unlike credentialists, they do not go as far as saying that a perceptual experience is a source of justification for you only if you have evidence that it is reliable. Conservatives, then, do not accept KR principles, and therefore avoid the circularity problem that results from conjoining KR and RK.Footnote ²⁵ However, these advantages notwithstanding, since conservatives reject KR, they face the problem of easy knowledge no less than dogmatists do.Footnote ²⁶

9. Two kinds of basicness

Cohen calls ‘basic knowledge’ the kind of knowledge that is possible if KR is false: knowledge based on a body of evidence that doesn't include any evidence that its source is reliable. For example, when a table's looking red to me by itself – unaided by any evidence of reliability – gives me knowledge that the table is red, then my knowledge that the table is red is basic in Cohen's sense. Views that deny KR are, in Cohen's terminology, ‘BKS theories’: theories with a basic knowledge structure. This terminology suggests that traditional foundationalism, since it postulates the existence of basic knowledge, is a BKS theory. But it would be a mistake to think that basic knowledge as Cohen defines it is the same as the kind of basic knowledge that foundationalists say must exist if there is any knowledge at all. Rather, we must, as Sosa suggests, distinguish between the following two kinds of basicness:

1. (i) Kp is inferentially basic iff Kp is not based on inference, so its sources are not inferential.

2. (ii) Kp is hierarchically basic iff Kp is unaccompanied by knowledge that its basis (or, more generally, its source) is reliable.

Sosa adds: “Thus, a bit of knowledge could be inferentially basic without being hierarchically basic, and a bit of knowledge could be hierarchically basic without being inferentially basic.”Footnote ²⁷

Sosa's distinction between inferential and hierarchical basicness is important. It will be worth our while to focus on it. In addition to the two possibilities Sosa mentioned, there are two further ones. Here is a list of all four:

1. 1. I-basic & H-basic

2. 2. I-basic & ~(H-basic)

3. 3. ~(I-basic) & H-basic

4. 4. ~(I-basic & ~(H-basic)

I will review all four cases. But before I do so, I wish to provide a bit more theoretical background about the two types of basicness at play.

10. Justification and meta-justification

The kind of basic justification dogmatists have in mind is possible only if the KR principle is false. Pryor calls such justification ‘immediate.’ I do not think using the label ‘immediate’ is entirely fortuitous because it blurs the distinction between the two types of basicness we must separate. If your belief that the table is before you is red is justified solely by its looking red to you, then, I agree, it is properly described as being immediately justified. But a belief is also properly described as immediately justified if its justification does not come from any additional belief and thus is non-inferentially justified. These two types of immediacy must be distinguished because, as Sosa points out, non-inferential justification need not coincide with the kind of immediacy dogmatists have in mind.

The label ‘non-hierarchical’ is not entirely free of ambiguity either. Inference creates a hierarchy of premises and conclusions. A belief that's justified via inference from premise-beliefs is hierarchically justified in a way that non-inferentially justified beliefs are not. Hence, it's necessary to spell out in exactly which sense the kind of basicness that's incompatible with the KR principles is non-hierarchical. Here is how I think this can be done. There is a kind of hierarchical structure – marking two different types of justification – that is the central concern in the dispute over the KR principle, namely 1st order vs. meta or 2nd order justification. For example, a table's looking red gives you 1st order justification for believing it is red. Evidence indicating that your visual experience is reliable gives you 2nd order justification or meta-justification for believing that the table is red. In general terms, meta-justification is justification for believing that the things that give you 1st order justification are reliable or trustworthy. KR principles express the thought that there is no justification, and thus no knowledge, without meta-justification. Opponents of the KR principle believe that, if there is any knowledge at all, there must be justification without meta-justification.

Using the distinction between 1st order and 2nd order justification, we may now describe the two types of basicness as follows:

Here is what's important about this distinction: inferential basicness does not require hierarchical basicness. It doesn't because, as we will see momentarily, it's possible that I have non-doxastic perceptual justification for believing p and, at the same time, also have meta-justification for believing p.

11. I-basic and H-basic

Knowledge that is both inferentially and hierarchically basic has these two features: it does not result from inference from another belief, and its source does not include any kind of meta-justification: no evidence indicating that its source is reliable. Here are a couple of examples:

1. (i) You know that the table before you is red solely because it looks red to you. At the same time, you are not in a position to know that, under the circumstances, looking-red is a reliable indicator of being-red because you have zero evidence that the table is not illuminated by red light.

2. (ii) You know that the animals in the pen are zebras solely because they look like zebras to you. At the same time, you are not in a position to know that, under the circumstances, looking-like-a-zebra is a reliable indicator of being-a-zebra because you have no evidence allowing you to know that the animals in the pen are not cleverly disguised mules.

Dogmatism should not be understood as the view that these two cases describe our ordinary situation when a table looks red to us or animals in the zoo look like zebras. Normally, the presence or absence of red light is perfectly obvious. And normally we do have very good evidence for rejecting the cleverly-disguised-mule scenario. In ordinary life, therefore, it's not so easy to find counterexamples to KR. Dogmatism does, however, allow for the possibility of cases (i) and (ii) above. The view allows for this possibility precisely because it denies KR. If KR is false, if perceptual knowledge does not require evidence of perceptual reliability, then a perceptual experience that p, if in fact reliable and resulting in a true belief that p, is sufficient for coming to know that p is true. From knowledge of p, it can then be deduced that any deceptive ~p situation does not obtain. According to the objection from easy knowledge, coming to know one isn't deceived isn't quite as easy as that.

12. I-basic and ~(H-basic)

I consider this category the main take-away of Sosa's distinction. Its significance is that foundationalism is not the same as dogmatism. Why? Consider the standard definition of the kind of belief foundationalists take to be basic:

A belief that is non-inferentially justified in this way has a type of justification that is non-doxastic, one that does not come from further beliefs. So, the answer to the question,

is: because evidence of perceptual reliability need not come in the form of beliefs about perceptual reliability.

Imagine yourself in front of a red-looking table. You are going to have introspective and perceptual evidence regarding the present circumstances. This evidence, were you to consider it, would tell you the following: You have not taken any vision-distorting drugs, you don't suffer from a brain tumor causing perceptual malfunction, the table is illuminated by ordinary daylight coming through a large window. Your memory, if you were to consult it, would tell you that in the past, your color vision has, under such conditions, been nearly always accurate. This is a rich body of evidence for the belief that, under the present circumstances, the table's looking red to you is reliable. It is evidence that you can trust your visual experience. The crucial point is this: if we understand KR as requiring not knowledge but merely evidence of reliability, then KR does not require that, for the table's looking red to be a knowledge source for you, you must form explicit beliefs on the basis of the reliability evidence I just described. Having this body of evidence is sufficient for satisfying what KR demands.

The language of seemings is useful to press this point. Consider again the example of the red-looking table. You have introspective seemings about the content of your visual experience. You have perceptual seemings about the size of the table, about how far away it is from you, and about the lighting conditions. Finally, you have memorial seemings that, in the past, perceptual experiences under such conditions have been highly accurate. Suppose further that you don't form any beliefs about all of that. Thus your belief that the table is red is based just on seemings, not on any further beliefs. Your red table belief, therefore, is a non-inferential belief – the kind of belief foundationalists consider basic. Nevertheless, your epistemic situation satisfies a principle like KR_D. Therefore, foundationalists are free to endorse KR_D.

13. ~(I-basic) and H-basic

Sosa suggests that “a bit of knowledge could be hierarchically basic without being inferentially basic.” What would be an example of such knowledge? Let's consider a candidate for this status: an inferential belief that is hierarchically basic. Assume you infer and thereby come to know that your car's tank is half-full because the needle of the gas gauge points to ½. Can this bit of knowledge be hierarchically basic? If it is, then its evidential basis does not involve any meta-justification: it involves no reasons for believing that the gas gauge is reliable or trustworthy. The question, then, comes to this: Can you come to know

by inferring it from

without having any evidence for

Gas-gauge dogmatism rejects the KR principle applied to gas gauges:

By itself, gas-gauge dogmatism does not suffice to yield the result that your knowledge of the tank's being half-full is non-hierarchical. To secure the non-hierarchical nature of your knowledge, we also need a kind of inference dogmatism. This view rejects the KR principle applied to the inference in question:

For your knowledge of H to fit the category under examination, a radical form of dogmatism would have to be true: the kind of dogmatism that rejects both KR_gas-gauge and KR_inf. If one thinks that dogmatism of this kind is implausible, one might doubt that there really is inferential knowledge that is hierarchically basic.

14. ~(I-basic) and ~(H-basic)

Suppose I come to know that the tank is half-full because I infer it from (i) the gas gauge's needle position and (ii) from my justified belief that my car's gas gauge is reliable. Given (i) and (ii), my knowledge is neither inferentially basic nor hierarchically basic. It seems unproblematic that knowledge of this type exists. What is controversial is its scope. In The Structure of Empirical Knowledge, Laurence BonJour argued that all empirical knowledge is like that.Footnote ²⁹ The view BonJour defended in this book is extreme: justified perceptual beliefs are always both inferentially and hierarchically justified because they receive their justification from meta-beliefs about perceptual reliability.

Epistemologists who like the KR principle need not agree with BonJour's doxasticism: his view that meta-justification must come from beliefs. If they endorse KR but reject BonJour's doxastic coherentism, they thereby take a foundationalist position according to which perceptual knowledge does not require meta-justificatory beliefs about perceptual reliability. It merely requires evidence of perceptual reliability, giving us justification for believing that perception is reliable were we to reflect on the matter. On this view, all perceptual knowledge is hierarchical, but not all of it is inferential. Dogmatists advocate a type of foundationalism that is more radical than that: dogmatic foundationalism, according to which some perceptual knowledge is neither inferential nor hierarchical.

15. The problem of non-existing knowledge sources

Although Sosa is on balance friendly to KR, he raises two problems for it. First, some instances of knowledge might not come from any source. Second, in many cases, what source a bit of knowledge comes from might be entirely hidden. I consider the first challenge in this section and the second in the next.

As examples of knowledge without a source, Sosa mentions knowledge of the following two propositions: “I exist” and “something exists or not.”

Here, two questions must be asked:

1. 1. Assuming there is source-less knowledge, how much of a problem would that be for advocates of the KR principle?

2. 2. Is there source-less knowledge?

In response to the first question, the following two points seem relevant. First, instances of source-less knowledge do not seem to be counterexamples to the KR principles because they are instances of a type of knowledge that falls outside the scope of KR. If knowledge of p is source-less knowledge, it doesn't come from a source whose reliability, according to KR, needs to be known. Second, if there is source-less knowledge, then there is a type of knowledge that is unaffected by the twin problems of easy knowledge and vicious circularity. However, even if there was such a type of knowledge, a lot of our knowledge would still come from sources. If we wish to explain how such knowledge is possible, we need to find a solution to these problems.

In response to the second question, I believe skepticism is warranted. If we search for the sources of knowledge of our own existence, there are certainly good candidates that suggest themselves. One is perception. I can see and feel my body. Another one is introspection. I'm introspectively aware of my current thoughts. And, as Descartes pointed out, it's difficult to see how I could be having these thoughts if I didn't exist. It would seem, therefore, that knowledge of my own existence is a plausible instance of the KR principle: it has a source, and I can know that this source is reliable (and indeed infallible). Analogous reasoning applies to axiomatic a priori knowledge. I know “Something exists or not” through a priori intuition, and through a priori intuition I can know that this proposition cannot be false. Again, we seem to have a case not only of knowledge coming from a source, but knowledge that fits the KR principle. For, if intuition tells me that p cannot be false, then I have excellent evidence for thinking that my belief that p comes from a reliable source.

Perhaps, when Sosa considers the possibility that some knowledge is without a source, what he has in mind is knowledge without a rational source. The worry might go as follows. My belief in my own existence has its source way back in my distant past. On that approach, the source is identified with the causal process that originally produced my belief. It seems clear that, if we think of my belief's source in that way, I do not have cognitive access to its source. It therefore cannot function as a reason for me on which I base my belief. Yet I know I exist. My belief in my own existence, then, is arguably an instance of knowledge although it does not have a rational source. According to this objection, my belief has a source, but it is hidden from me. I'll discuss this objection next.

16. The problem of hidden sources

Sosa suggests that

The types of beliefs Sosa is talking about are stored beliefs. If Sosa is right that the etiology of stored beliefs is typically beyond our ken, then KR is a recipe for skepticism regarding stored beliefs.

In response, I suggest that we should distinguish between the distant causal basis operative when our beliefs were formed in the past, and the causal basis that sustains our beliefs in the present. The latter can be identified with seemings. Michael Huemer, for example, would say that my stored belief that water is H₂O is causally sustained by a seeming: it seems to me that water is H₂O.Footnote ³¹ I see no reason to think that reliability evidence for a seeming of this type is unavailable. The same reasoning applies to other stored beliefs. Typically, they are causally sustained by seemings we have each time a stored belief becomes an occurrent belief. And we have evidence that such seemings are by and large reliable. For example, many of these seemings are like the following: it seems to me that p because:

• – I learned p at school;

• – I heard it from an expert;

• – I acquired the belief that p from a trustworthy source.Footnote ³²

To sum up, I agree with Sosa that stored beliefs pose a challenge for credentialist views that endorse the KR principle. At the same time, I believe there are resources of which advocates of such views can avail themselves to explain how we can know that stored beliefs have a reliable source.

17. Sosa's response to the circularity problem

The puzzle of reliability knowledge raises the twin problems of easy knowledge and potentially vicious circularity. If we reject dogmatism because we consider the consequence of easy knowledge unacceptable, we must accept some version of KR. Since it is difficult to see how perceptual reliability can be established a priori, we must accept some version of RK as well. And now we face the problem of source circularity: we are bound to use a source to confirm its own reliability. Fumerton thinks that's unacceptable:

In the same vein, Sosa says:

The general worry is clear, whether we are talking about seemings, faculties, competences, or dispositions: using a source to confirm its own reliability seems unacceptable. But why? Sosa suggests that, for a faculty to deliver epistemic goods, it must have proper epistemic status.Footnote ³⁵ Sosa's point, I take it, is that a faculty cannot epistemically certify itself: it cannot confer proper epistemic status upon itself.

Let's apply this point to perception. To establish that perception is reliable, we must use perception. But if perception is to deliver epistemic goods – justified belief in its own reliability – we must already be in a position to justifiably presuppose that perception is reliable. How can we break out of this circle? Here is what I think is Sosa's response. First, he says that “it is far from obvious, and seems in fact false, that our epistemic competences generally operate through such presuppositions.”Footnote ³⁶ And then he moves on to say:

Sosa's solution seems to be that a faculty or competence such as vision can, at a sub-personal level, deliver justified beliefs without a prior justified presupposition of reliability. Prior to having evidence of visual reliability, we can already use vision to generate justified beliefs that permit us to then infer that vision is reliable. So, in effect, Sosa is suggesting that, as far as vision is concerned, we must reject KR. But then it is unclear why vision isn't subject to the problem of easy knowledge in just the way Cohen says it is. And Sosa seems to agree with Cohen that the problem of easy knowledge is pernicious. Sosa, then, would have to explain how vision can be exempted from the KR principle without thereby allowing for the creation of easy knowledge.

18. Diachronic-successive versus synchronic-structural epistemic priority

I would like to suggest a different solution to the circularity problem. I think we should accept both KR and RK and reject

KR and RK both assert priority relations. The following two articulations of KR and RK makes that explicit:

NP is plausible if we interpret the priority relations KR_pri and RK_pri assert as diachronic-successive priority: one thing must happen before the other thing can happen. But there is an alternative. I believe we should think of the priority relations in question as synchronic-structural: one thing cannot happen without the other thing being present at the same time. Understood this way, the two principles may be recast thus:

KR_sim and RK_sim do not say that you must have evidence of perceptual reliability before you can have perceptual knowledge, and that you must already have perceptual knowledge before you can acquire evidence of perceptual reliability. Rather, the principles say merely that you cannot have one without the other. This allows for the possibility that both conditions are satisfied at the same time, with neither one of them being the sort of condition that must be met before the other one can be met. The two conditions need not – and indeed cannot – be satisfied successively. They only need to be satisfied simultaneously.

I suggest that KR_pri and RK_pri should be construed as KR_sim and RK_sim. Next, I will argue that there is good reason to think that the circularity involved in mutual dependence of this kind is benign.

19. Benign mutual dependence: the clown fish and the sea anemone

The world is actually full of examples of benign mutual dependence. Consider the ecosystem in which we find the clown fish and the sea anemone. Both of these creatures depend on each other in various ways.Footnote ³⁸ Here is an argument that, duplicating the structure of the puzzle of reliability knowledge, focuses on the mutual dependence between the clownfish and the sea anenome:

Therefore:

But these creatures do exist. So what's wrong with the argument? Here is the solution: FA and AF must not be understood as asserting conditions that must be satisfied successively. Rather, they assert a relationship of structural mutual dependence: the two conditions must merely be met at the same time. If FA and AF are understood in this way, NP is false.

There are many other examples that illustrate the same point. The organs of an organism depend on each other. It's not the case that an organism's organs must come into play one after the other. Rather, for an organism to be alive, they must all be in place at the same time. The same goes for the parts of an internal combustion engine: cylinders, spark plugs, carburetor, etc. It's not the case that the cylinders must start working before the carburetor does, or that the carburetor must already be doing its job before the pistons in the cylinders start moving. Rather, for the engine to be working, all its parts must be in functioning order at the same time.

I want to suggest that we should think of perceptual justification and evidence of perceptual reliability as necessary parts of the cognitive engine. For the engine to produce perceptual knowledge, they must both be in place at the same time. But neither of them needs to precede the other one. Mutual dependence of this kind involves source circularity: perception confirms the reliability of perception, and evidence of perceptual reliability certifies perception as a justification source. That's circularity all right, but there is nothing vicious about it, just as there is nothing vicious in the way the clown fish and the sea anemone, a heart and a liver, or a cylinders and spark plugs depend on each other.Footnote ³⁹

20. Easy knowledge and Sosa's two-stage model of reflective knowledge

While Sosa is known for advocating that there is such a thing as animal knowledge, in the end, it seems to me, what he really cares about is reflective knowledge. Reliably functioning faculties give us animal knowledge. Reflective knowledge involves more than reliably functioning faculties: it requires knowledge that our faculties are reliable. Thus Sosa faces the puzzle of reliability knowledge and its twin problems: easy knowledge and seemingly vicious circularity. What is his solution? It is, it would seem, a two-stage model that is diachronic-successive. At the first stage, vision delivers epistemic goods although we do not yet have justification for taking it to be reliable. At the second stage, enough epistemic goods are in place to recognize the reliability of vision. Visual knowledge then becomes reflective knowledge.

Unlike Pryor, Sosa does not argue that easy knowledge is benign.Footnote ⁴⁰ But the two-stage model does not steer clear of easy knowledge. If vision can give me epistemic goods although I do not yet have evidence of its reliability, then I can use these epistemic goods for dismissing deceptive alternatives. I share Cohen's intuition that that's epistemically defective reasoning, and it looks like Sosa shares this intuition too. It would seem, therefore, Sosa needs to explain how the two stage model he advocates avoids the problem of easy knowledge.

21. The web

What makes knowledge reflective is having justification for believing in, and ideally knowing, the reliability of our sources. For externalists, reliability knowledge is seemingly unproblematic: if perception produces reliably the true belief that perception is reliable, you know that perception is reliable. For dogmatists, too, reliability knowledge is seemingly unproblematic. Perception gives us knowledge without antecedent evidence of its reliability. Using this knowledge, we can then bootstrap to knowledge of perceptual reliability. According to the objection from easy knowledge, each of these views makes knowledge or perceptual reliability too easy to acquire. The alternative is credentialism: perception can give us knowledge only if we have, at the same time, evidence of its reliability. On this approach, all perceptual knowledge is reflective – or at least quasi-reflective if reflective knowledge requires actual reflection. While credentialism avoids the problem of easy knowledge, it runs into another problem: that of source circularity, or self-support. Sosa endorses the metaphor of the self-supporting web as the right way of understanding how reflective knowledge comes about. He says:

But here is a question about the epistemic role the web plays. Does the web create epistemic justification, or does it merely augment previously created justification? It seems to me that, unless there is a way of defusing the charge of easy knowledge, we must endorse KR. But if KR holds without exception, then there is no perceptual knowledge at all without evidence of perceptual reliability. If we take that to be true, we must conclude that there is no perceptual justification outside of the web. The web creates all the perceptual justification there is.

22. The creation of epistemic justification

Can the web create epistemic justification all by itself? One reason for saying ‘no’ is that the web can't by itself deliver the initial boost. Sosa suggests a second reason. According to him, the answer is ‘no’ because hierarchically basic knowledge is unavoidable. It is unavoidable because, he says,

This argument is sound if we take the ‘know’ in the consequent of the knowledge version of KR, namely

to refer to explicit knowledge requiring occurrent belief. As I mentioned early on in this paper, if we interpret KR in this way, KR is not plausible. But from the implausibility of KR thus understood, it doesn't follow that there aren't less demanding and thus plausible versions of the principle. I believe that, if we interpret the knowledge operator in the consequent to mean “you must have knowledge-grade evidence for the reliability of S,” then KR is plausible. On this approach, KR does not generate an infinite level of reflection. Therefore, I don't agree with Sosa's claim that hierarchically basic knowledge is unavoidable.

Nor do I think that the initial boost objection is compelling. If we adopt the synchronic-structural understanding of epistemic priority, then there is no need for an initial boost. During the cognitive development of a child, perceptual knowledge and knowledge of perceptual reliability can emerge together just as, during the evolution of sea life in shallow lagoons and sheltered reefs, the clown fish and the sea anemone evolved together. On this model, we have gradual emergence instead of an initial boost.

How, then, does perceptual justification gradually emerge? When an infant starts observing the world around her, she notices that what can be seen can be touched, heard, tasted, and smelled. What she saw yesterday she sees again today. A rattle that was red yesterday is still red today. With enough effort, one can fill pages describing similar ways in which perceptual submodules confirm each other. As the infant develops, her cognitive web grows as well and includes introspection, reason, and memory. The latter in particular plays an essential role since, to a large extent, evidence of the reliability of a source is supplied by remembering its performance in the past.

What I have described is of course just an initial and rather rough account of the basic idea: the web creates perceptual justification through a continuous flux of mutual confirmations, which, by virtue of being remembered, constitute evidence of cognitive reliability.Footnote ⁴³ At the beginning of cognitive development, as soon as the infant starts trusting her perceptual powers she remembers perceptual success in the past, and perceptual justification gradually emerges. An essential element of this picture is that belief sources can confirm their own reliability. This is particularly striking in the case of memory, which, through the remembered track record of its past performance, confirms its own reliability. Like Sosa, I see no reason to think that there is anything wrong with epistemic self-support, thus conceived. Sosa, however, seems to think that epistemic self-support merely augments perceptual justification that was previously created through hierarchical basicness. Unlike Sosa, I believe the lesson of the problem of easy knowledge is that there is no hierarchically basic knowledge. The web creates all the perceptual justification there is.Footnote ⁴⁴