Sosa on easy knowledge and the problem of the criterion

It seems incredible that we could not somehow use our faculties to gain access to knowledge of our own reliability.

—from the last page of Sosa’s Reflective Knowledge

Sosa’s Reflective Knowledge (2009) is enlightening, penetrating, and far too rich to be engaged with in its entirety here. I have chosen to focus on the final chapter, in which Sosa draws upon various of the distinctions and doctrines from earlier chapters to address the Problem of Easy Knowledge and the Problem of the Criterion. Although our symposium is entitled “Author Meets Critics,” my own contribution might more aptly be entitled “Author Meets Querist,” as what I offer is an exposition of Sosa along with a sprinkling of questions about whether I have understood things correctly and how things might be developed further.

1 The problem of easy knowledge

The Problem of easy knowledge, as articulated in Cohen (2002), may be set out as a dilemma that arises depending on the stance we take on the following thesis:

• KRel: A potential knowledge source K can yield knowledge for S only if S knows that K is reliable.

If we reject KRel, we make knowledge too easy; if we affirm it, we make knowledge impossible.

The “impossible” horn may be elaborated as follows: On a natural way of understanding KRel, we cannot know anything from a source without first knowing that the source is reliable.¹ But how could we know that a source is reliable without at some point using that very source? To know that sense perception is reliable, for example, we would presumably have to know of many instances in which its deliverances had proven correct. But how could we know that its deliverances were correct unless we knew this through perception itself? Perhaps I could corroborate some deliverances of perception by relying on another source, such as the testimony of friends, but it seems inescapable that at some point I would be using perception itself—for instance, in knowing what my friends were telling me. It thus appears that accepting KRel lands us in a vicious circle, making knowledge through perception (or any other fundamental source) impossible.

If we reject KRel, we confront the first horn of Cohen’s dilemma, the “easy knowledge” horn. Cohen identifies two varieties of easy knowledge, one variety obtained through closure and the other obtained through bootstrapping.² He argues that knowledge of both varieties is made possible by epistemologies that affirm the existence of basic knowledge in the following stipulated sense: knowledge one has without having to know that its source is reliable.³

In what follows, I use the following abbreviations: LR = the wall looks red, R = the wall is red, W = the wall is white, RL = the wall is illuminated by red lights; JR = the subject is justified in believing R; KR = the subject knows R; NFB = there is no funny business going on.

Easy knowledge by closure would come about as follows: In accordance with epistemologies that allow R to be a piece of basic knowledge, one comes to know R simply on the strength of LR, the wall’s looking red to one.⁴ From one’s knowledge that the wall is red, one deduces that it is not white and, a fortiori, that is not white with red lights shining on it, ~(W and RL). By closure (the principle that if S knows P and deduces Q in the knowledge that P implies Q, then S knows Q),⁵ the subject thereby comes to know ~(W and RL), despite not having any independent reason to believe anything about the lighting conditions. That seems too easy.

Easy knowledge by bootstrapping may arise as follows⁶: On occasions when one takes a wall to be red merely on the strength of its look, one also notes introspectively how it looks, thus eventually amassing a body of evidence of the form ‘On occasion one the wall looked red, and it was red; on occasion two the wall looked red, and it was red, etc.; moreover, there were no occasions on which the wall looked red and wasn’t red.’ From this impeccable track record, one infers by induction that one’s color vision is reliable. The resultant knowledge looks too easy, since one’s endorsement of the accuracy of the appearances was based on nothing but the appearances themselves.

Sosa concentrates on the closure variety of easy knowledge. Here is his diagnosis of what is bad about it (pp. 219–221): Knowing R from LR presupposes that there is no funny business going on (NFB), such as would be happening if the wall were white and bathed in red light.⁷ What one presupposes in knowing something else must itself be known or have high some epistemic status. Thus ~(W and RL) cannot come to be known in the indicated easy way, for it would already have to have been known earlier.

I was initially tempted to regiment Sosa’s response to the easy knowledge problem as a definition and two claims:

• Definition: knowing (or having the level of justification requisite for knowing) p on the basis of q presupposes r = df one cannot know p on the strength of q alone, but must base one’s belief in p also on r.

• Claim 1: if knowing p on the basis of q presupposes r, then one knows p on the basis of q only if r has some high epistemic status, “amounting perhaps to knowledge” (p. 20).⁸ Moreover, r’s having this status is prior to (or at any rate cannot be posterior to) one’s knowing p, as happens in the red wall case.

• Claim 2: Knowing R from LR presupposes NFB.

The closure path to easy knowledge may be described in the following sequence, in which the steps aren’t necessarily the subject’s own assumptions and inferences, but ours in describing how things go with him:

• 1. LR

• 2. JR (from LR → J_pfR and assuming no defeaters so J_pfR becomes JR) (‘LR → J_pfR’ abbreviates ‘the wall’s looking makes it prima facie justified for the subject that the wall is red,’ something that typically holds in “basic knowledge” epistemologies.)⁹

• 3. KR (assuming that the subject’s belief is true and unGettiered).

• 4. K(R = >~(W and RL) (assumption)

• 5. K~(W and RL) (from 3, 4, and closure under known entailment)

Some there are who would block the progression to 5 by denying closure. Ernie denies instead the transition from 1 to 2, invoking the ideas about presupposition given above. Even in the absence of defeaters, LR does not suffice for JR, since you need to presuppose (and therefore have high epistemic status for) NFB.

Cohen’s recommended way of blocking easy knowledge of both varieties is to insist on the KRel requirement across the board. Sosa intimates on p. 221 that there is a way to block easy knowledge that does not involve endorsing KRel, since we can block it with the requirement that NFB be known instead. Reading on, however, I have the impression that the picture is somewhat different. Sosa distinguishes between generic and specific reliability (p. 225): a source is generically reliable if it would generally lead us to true beliefs; it is specifically reliable if it would tend to lead us to the truth in the present instance. NFB may be more or less tantamount to specific reliability. Is the distinction between generic and specific reliability meant to line up with the two versions of Cohen’s problem? If so, insisting on KRel for generic reliability would block easy knowledge by bootstrapping, while insisting on KRel for specific reliability would block easy knowledge by closure.

Whether the NFB requirement is a form of KRel or not, it appears to me that insisting on this requirement threatens us with the second horn of Cohen’s dilemma just as much as KRel does. If we must know NFB in order to know R from LR (and more generally X from LX), how could we ever know NFB? If getting beyond the appearances always requires knowledge of a claim that could itself only be known on the basis of the appearances, how can we ever get beyond them? I return to this problem in Sect. 3 below.

Assuming the preceding account of presupposition to be correct, I have several questions to ask Ernie about his solution to the closure version of the easy knowledge problem.

First, what nonepistemic relations does presupposing (as defined above) supervene upon? In our symposium, Ernie replied that he had not wanted to understand presupposing in epistemically normative terms. Presupposing q is being in a mental state with q as its content, and it is therefore simply a matter of psychological fact whether presupposing happens in a given case. This comports well with his saying later in the chapter (pp. 235–236) that subpersonal systems do not do any presupposing. But doesn’t this response threaten to make his solution to the easy knowledge problem inapplicable? We could program an agent or a being to believe that the walls are red whenever they look red, without any thought as to lighting conditions or the possible gap between looking red and being red. If basic knowledge epistemologies are correct, such subjects would get knowledge of R whenever LR and there were no defeaters—they could get as far as step 2 along the path to easy knowledge. Doesn’t Ernie need to be operating with some sort of normative requirement in order to “pin” on subjects a presupposition whose epistemic credentials they have done nothing to establish?¹⁰

Second, do we presuppose something in knowing p from q whenever q does not entail p? Answering this question in the affirmative would be giving a partial answer the preceding question. It would tell us that we presuppose something whenever we aspire to know p from a basis that does not entail it, though it would not tell us what we presuppose or what its logical and other nonepistemic relations to p and q are.

Third, can we give a general characterization of “funny business” (relevant to our knowing h from e) by saying: something is happening that entails e & ~h? If so, NFB would amount to ~(e & ~h) or, equivalently, the material conditional e → h. Answers of yes to this question and the previous one could be combined into an answer to the first question: knowing p on the basis of q presupposes r when (i) q fails to entail p and (ii) r is equivalent to the proposition that q materially implies p. In the example we have been working with, knowing R from LR would presuppose something because LR doesn’t entail R, and what it would presuppose is that conditions are propitious for taking your current color experience at face value, i.e., that LR → R.

If we combine the suggestions about presupposition I have just floated with Sosa’s thesis that one’s presuppositions must have high epistemic status, we arrive at the following (to me) unsettling result: there is no such thing as prima facie justification. In saying this, I rely on the classical characterization of prima facie reasons in Pollock (1974). Prima facie reasons are logical reasons for which defeaters may exist (hence, reasons that do not entail what they are reasons for), and logical reasons are defined as follows:

P is a logical reason for believing Q iff P can be a good reason for someone to believe Q without his having an independent reason for believing P → Q. (1974, pp. 34, 42)

I call the result that there are no prima facie reasons unsettling because it is hard for me to see how it can fail to have skeptical ramifications.

Sosa may not give the general answers I have identified as leading to this result, but would he accept the result itself?¹¹ On the one hand, he may fear as I do that if there is no such thing as a prima facie reason, skepticism threatens. On the other hand, if there is such a thing as a prima facie reason, the problem of easy knowledge threatens. If LR is a prima facie reason for R, we can get from step 1 to 2 along the path to easy knowledge. If we can go that far, what prevents us from going the rest of the way?

2 A skeptical argument

Sosa lays out the following a–b–c–d argument for skepticism (pp. 228–229):

• (a) If there is human knowledge at all, then there is (hierarchically) basic knowledge.

• (b) If one knows that p in a basic way, then one can come to know, based partly on this knowledge, that the source of this knowledge is both specifically and generically reliable.¹²

• (c) In no case where one knows that p can one come to know, based even just partly on this knowledge, that the source of this knowledge is specifically or generically reliable.

• (d) Therefore, there is no human knowledge.

Sosa notes that for most of us, the conclusion constitutes a reductio of the premises (p. 229). He asks (p. 230) whether denying (b) is the best way to avoid the conclusion, but as far as I can tell, he never explicitly answers this question later in the chapter. So which premise would he have us deny?

I take him to deny (c). That is because he affirms (a) and denies (d), so he must deny either (b) or (c). But (b) is hard to deny given the way basic knowledge has been defined: basic knowledge does not depend on knowledge of reliability, so it ought to be usable to acquire knowledge of reliability.

3 The problem of the criterion

The problem of the criterion, as it has come to be understood since Chisholm (1973), is often understood thus: Do we start with (a) knowledge of general epistemic principles or “criteria,” such as principles affirming the reliability of certain sources, or do we start with (b) particular bits of knowledge delivered by these sources, or do we start with neither of these things until the other is in hand?¹³ If we start with (a), we may be able to get (b), and if we start with (b), we may be able to get (a), but if we start with neither, we get nowhere. In stating the second horn of his dilemma, Cohen notes that it lands us squarely in the Problem of the Criterion. If KRel is true, we need to know that our sources are reliable to know anything through them, but it seems equally undeniable that we could only know our sources are reliable by reasoning from things known through them. There appears to be a vicious circle into which we cannot possibly enter.

Sosa lays the groundwork for two kinds of solution to the problem of the criterion. One involves his signature distinction between two levels of knowledge, the animal and the reflective; the other invokes the benefits of coherence that he emphasizes throughout the book. I have the strong impression that his preferred solution is meant to be a blend of the two strategies, but I would like to know more about how it works. In what follows, I shall describe the strategies separately and hope that Ernie will tell us more about how they are to be integrated.

3.1 The two-levels solution

For present purposes, the key difference between animal knowledge or justification and reflective knowledge or justification is this: animal knowledge does not depend on any endorsement of one’s reliability, specific or general, while reflective knowledge does so depend (pp. 238–239). Animal knowledge does not carry any presuppositions.¹⁴

I will designate the lower-level mode of knowing as K₁ and the higher-level reflective mode as K₂. The distinction between these levels gives us the materials for a solution to the second horn of Cohen’s dilemma along the following lines: Since K₁ does not carry any presuppositions, one can have K₁ of R based on LR alone without presupposing NFB. Can we advance beyond this level to K₂ of R? That depends on the answers to two further questions. First, K₂ of R from LR would require knowledge of NFB, but which type of knowledge, K₁ or K₂? If K₂ is required, perhaps we can advance no further. But suppose we require only K₁ of NFB—can we get it? Perhaps we can. I am not sure whether Sosa holds that animal knowledge is closed under (animally?) known entailment, but suppose we do have such closure.¹⁵ If so, our subject could advance, via knowledge of the trivial entailment from R to ~(W & RL), to knowledge (K₁) of the absence of one kind of funny business. He could also advance in the same way to knowledge of ~(~R & LR), which was one of our conjectured formulations of a more general NFB condition.¹⁶ He would thus arrive at K₁ of NFB, elevating his original K₁ of R to the second level. In this way, it would be possible to get K₂ of R even if we insist that such knowledge must go through knowledge of NFB.

What I have just sketched is actually a solution not to the Problem of the Criterion, but to the skeptical horn of Cohen’s dilemma that threatens if we block easy knowledge with an NFB requirement. However, an analogous two-levels solution could be given to the skeptical horn that threatens if we block easy knowledge with a KRel requirement. The idea would be to get K₁ of the reliability of one’s sources or faculties, upon which one’s initial knowledge attained through those faculties would be upgraded from K₁ to K₂.

I wish now to make a general observation about two-levels strategies. The basic idea is that we have K₂ of p when we have K₁ of p and K₁ as well of various other epistemically significant propositions, such as that there is no funny business or that our relevant faculties are reliable or that there are no reasons in play to doubt p. Schematically, K₂p iff K₁p & K₁Hp, where Hp is the epistemically significant proposition; this schema covers Ernie’s theory and several possible others. In such theories, K₂ is compounded out of K₁. That being so, it seems clear to me that the epistemic value of K₂ is directly proportional to that of K₁. And that means there is potentially a dilemma for two-levels strategies. On the one hand, if we set the epistemic value of K₁ too low, the value of K₂ may not be high enough. Gewirth’s solution to the Cartesian Circle in (1941) may provide a case in point.¹⁷ On the other hand, suppose we let K₁ be a fairly robust kind of knowledge. Then K₂ will be impressive, but we may be in danger of letting the easy knowledge problem arise all over again for K₁: there will be a robust kind of knowledge that is all too easily obtainable. Is there a way of setting the epistemic status of K₁ that (like the third bowl of porridge) is just right?

3.2 The coherentist solution

In the final section of Chap. 10 (pp. 239–243), Sosa sketches a coherence model of justification in which each item in one’s web of belief is based partly on other items in the web and also gets some support from the others. He suggests that if a belief in R and a belief in NFB are both elements in such a system, the belief in NFB can get a little bit of extra epistemic status from being based on the belief in R, even though it may be “minuscule” or “vanishingly small” (pp. 240, 242). Likewise, the belief in the general reliability of one’s color vision could get a little bit of extra epistemic status from various episodes of believing LR and R.

The first question I would like to ask about the coherentist solution is how we get the little bit. Easy knowledge by closure was criticized for being circular—for letting A get its justification from B, when B could get its justification only from A. Why is the circularity any less objectionable when the amount of justification involved is slight?

My next question is how we get from a little to a lot. Recall that what was required above in order to know Red from Looks Red was a high degree of epistemic status for NFB. How do we attain that high degree?

A traditional view about the powers of coherence is that it can amplify the amount of justification possessed by a set of beliefs individually (or apart from their coherence) provided each of the beliefs has some appropriate level of justification antecedently (Compare Principle H in Chisholm (1977)). So it seems that coherence might be just the thing to take us from a little to a lot. But Ernie is already invoking coherence to give us the initial small bit for NFB. Is he using coherence twice (or perhaps in an ascending spiral)—once to get us the initial bit and again to turn a little into a lot?

Let me turn now to a point evidently essential to any coherentist solution to the Problem of the Criterion. In elaborating the second horn of Cohen’s dilemma, I took the KRel requirement to imply that there is no knowledge from a source unless one first knows that the source is reliable—that the reliability of K is epistemically prior to any knowledge through K. Assuming that one could know that K is reliable only by using knowledge delivered by K itself, it would follow that knowledge is impossible. Any coherentist solution to the Problem of the Criterion must clearly deny that there is any such mutual priority. Sosa is explicit about this; in his language from the symposium, KRel is false if it requires antecedent clearance of any source, but true if it requires correlative clearance. (Presumably, a requirement of correlative clearance is enough to block easy knowledge, so we are not back on the first horn.)

If knowledge of reliability (specific or general) and knowledge of its deliverances are merely corequisites of each other, not prerequisites, there is no longer a problem of a vicious circle into which we cannot possibly enter. A question remains, however, about the mechanism by which we arrive at knowledge of the two in tandem. If Sam and Sally truthfully pledge to each other that each will go if the other goes, it is not yet assured that either will go; something must happen to make both of them go rather than neither.

In the case at hand, we are trying to arrive at knowledge of both the redness of the wall and a fact about the lighting conditions. The idea is that we could not get either by itself from evidence such as LR, but we can get them as a package deal. In other words, we can advance to a conjunction, but not to either conjunct separately. Is there anything odd (or “probabilistically incorrect”) in the idea that the stronger conclusion would be justified while the weaker is not?¹⁸

3.3 A combined solution?

How are the two-levels and coherentist solutions meant to be combined? What is really key in surmounting the Problem of the Criterion—the distinction between K₁ and K₂, the distinction between prerequisites and corequisites, or both? What are the inputs and outputs in the workings of coherence—do we have animal knowledge going in and reflective knowledge coming out? Is the onset of reflective knowledge sudden, as I gather it would be in a straight two-levels solution, or gradual, as I gather it might be in a coherentist solution? These are matters about which I would love to hear more.

*****

In a passage sometimes quoted to illustrate the alleged absurdity in using our faculties to demonstrate their own reliability, Reid said, “If a man’s honesty were called in question, it would be ridiculous to refer it to the man’s own word, whether he be honest or not” (1785, p. 480). Yes; but it would also be ridiculous to believe a man’s testimony about various matters before the court (the time of the crime and so forth), but to balk for the first time when he said, “Moreover, I am a truthful witness.” If we are not justified in believing what a faculty tells us about itself, we are not justified in believing anything it tells us about anything else, either—in which case, none of our sources can be trusted, and we know nothing. That is one reason for thinking that it must somehow be possible to use our faculties to gain knowledge of their own reliability.