That our attitudes of belief are commonly not self–critical, is a matter upon which we are to be congratulated, in view of the practical consequences of over–much self–criticism and hesitation in action. That we are capable of such self–criticism when it is called for, is the most that could be desired Lewis 1947, 326Lewis, C.I. 1947. An Analysis of Knowledge and Valuation. La Salle: Opencourt..
Consider simple externalism (SE):
A class of cases from BonJour, 1980BonJour, Laurence. 1980. Externalist Theories of Empirical Knowledge. Midwest Studies in Philosophy 5: 53–73. have generally been accepted as demonstrating that SE is inadequate. These are cases in which SE predicts that S knows that p, even though S's overall doxastic state requires that S withhold judgment with respect to p, or actually believe not–p. Many have concluded that what externalists need is a a condition that none of S's mental states undercut or rebut S's belief that p. We call any such condition a no–defeater–condition. By way of contrast with SE, we call any externalist theory that includes a no–defeater condition a No–Defeater Externalism (NDE). Hold that thought.
Briefly shift attention to epistemological level connections. A level connection is a connection between the epistemic status of an object–level belief and a corresponding meta–level belief.†I borrow the term `level connection' from (McGrew and McGrew 1997). Their characterization is different than mine. These level connections are typically stated in conditionals that we will call connection principles. For example, consider a connection principle popular among internalists:
Now bring back that thought about No–Defeater Externalism. I will argue in this paper that NDE, unlike SE, is committed to deep level connections. This is surprising since level connections have traditionally been the hallmark of internalist epistemologies; since Alston's ``Level Confusions in Epistemology,'' most externalists have denounced such connection principles as level confusions.†If one confuses the epistemic status of a belief at one level (e.g. the object–level) with the epistemic status of a belief at another level (e.g. the second–level), one has made a level confusion. It is, of course, an open question whether internalists are making level confusions or are simply picking up on level connections. (McGrew and McGrew 1997) argue that there was never any confusion. But what is most interesting about the connections is what they tell us about a hallmark of externalist epistemology: basic knowledge.†Cohen (2002) coins the term `basic knowledge' to refer to knowledge that can be acquired from a source prior to knowledge of the source's reliability. The level connections to which NDE is committed make basic knowledge easy to lose. I call this the Easy Come, Easy Go phenomenon. Knowledge has always been easy to acquire, according to externalists; but it seems that, according to NDE, it is equally easy to lose. Exploring this phenomenon would be an interesting task. But such an exploration must wait; the first order of business is to argue for the level connections that engender the phenomenon.
Here is the plan. I begin by motivating and explaining the introduction of a no–defeater condition into SE. I then consider a weak connection principle inspired by Chisholm. The externalist intuition is that the principle is false. I argue, however, that commitment to a no–defeater condition is, mutatis mutandis, commitment to this weak connection principle. I then consider two stronger connection principles. I conclude by briefly explaining how these level connections give rise to the Easy Come, Easy Go phenomenon. Externalist explanations of the Easy Come side of the knowledge coin are well–known. But externalist explanations of the Easy Go flip–side haven't been discussed. I offer possible explanations, but do not develop them here.
explaining bonjour's cases
The most troubling cases for SE are due to Laurence BonJour (1980). These cases, unlike Truetemp–type cases, are ones in which a belief is held not merely in the absence of good reason, but in the presence of defeating reasons. The following are the two sorts of cases I have in mind:
Why are these problematic for the SE? First, note that SE counts S (in both cases) as knowing that p. After all, it seems obvious that given the reliability of her clairvoyance, and the fact that her belief is the result of that clairvoyance, it is objectively likely that S's belief that p is true. It also seems intuitive, however, that in these cases S does not know that p.
Let's spell out explicitly why S doesn't know that p in CASE I and CASE II. We'll assume that S can't know that p if, for S, there is some doxastic attitude towards p that is more reasonable than belief. In CASE I, the only evidence S could muster for p would be that it seems to her that p. But this prima facie evidence is overruled by S's strong evidence against p. This suggests that S would be more reasonable to believe not–p, and so, ipso facto, does not know that p.\footnote{I'm using the word 'evidence' here as the sort of thing to which the agent has access; S knows what her evidence is. That's what makes these cases so compelling—S's doxastic system totters near contradiction. In CASE II, S has no evidence against p, but she does have evidence against q. Evidence against q is evidence that her belief is as likely false as it is true, conditional on how it was formed. When S has evidence that her belief that p is as likely false as it is true, it is plausible to say that S would be more reasonable to withhold judgment with respect to p, and so, ipso facto, does not know that p.
The dialectical situation is this:
- Externalism is fundamentally flawed,
- CASES I and II are not really counterexamples, or
- SE is incomplete.
BonJour and many internalists, of course, conclude that externalism is fundamentally flawed. Perhaps there are externalists willing to reject CASES I and II as genuine counterexamples. Most externalists, however, hope to emend their views so as to rule out CASES I and II as instances of knowledge.\footnote{See (Bergmann 1997) for a discussion of the positions of various externalists on this point. Rather than canvass the many conditions externalists have proposed to handle these cases, I will introduce the notion of a defeater to diagnose what goes wrong in CASES I and II. A general account of what goes wrong in these cases will allow me to introduce the idea of a no–defeater condition. This idea is general enough to include any condition externalists might propose to exclude CASES I and II as instances of knowledge.
- Defeater
- d is a defeater at t for S's knowledge that p iff Bergmann 2005, 422Bergmann 2005. Defeaters and Higher–Level Requirements. The Philosophical Quarterly 55: 419–436..
We can now specify two species of mental states that are candidate defeaters:†These definitions are inspired by (Bergmann 20005), but they differ in certain respects. For one, Bergmann defines undercutting and rebutting defeaters. Given the definition of a defeater, it is analytic that if $S$ knows that $p$, then $S$ has no undercutting or rebutting defeaters. I define undercutters and rebutters such that they are not analytically absent in cases of knowledge—that is, it is not analytic that they are defeaters.
- Rebutter
- d is a rebutter for p iff d is (or is an epistemically appropriate basis for) the belief that p is false Bergmann 2005, 422Bergmann 2005. Defeaters and Higher–Level Requirements. The Philosophical Quarterly 55: 419–436..
- Undercutter
- d is an undercutter for p iff d is (or is an epistemically appropriate basis for) the belief that one's actual ground or reason for p is not indicative of p's truth or is (or is an epistemically appropriate basis for) the withholding of judgment with respect to whether one's actual ground or reason for p is indicative of p's truth Bergmann 2005, 424Bergmann 2005. Defeaters and Higher–Level Requirements. The Philosophical Quarterly 55: 419–436..
If we take it that rebutters and undercutters are defeaters, then we can use them to explain CASES I and II. Recall that in CASE I, S has evidence, call it $d$ that supports not–p. Since d is an epistemically appropriate basis for the belief that p is false, d is a rebutter for p. In CASE II, S has evidence, call it $d$, that supports not–q (evidence that S is not a clairvoyant). Since d is an epistemically appropriate basis for the belief that q is not indicative of p, d is an undercutter for S's p. In CASES I and II, it is the presence, respectively, of a rebutter and an undercutter that prevents $S$ from knowing that $p$. What these BonJour cases reveal, then, is that an adequate analysis of knowledge will have the consequence that rebutters and undercutters are defeaters. SE has no such consequence.
For an analysis of knowledge to have this consequence, it is sufficient to have a condition that guarantees S has no rebutters or undercutters. Following Bergmann, I call these 'No–Defeater Conditions':†Again, I define a No–Defeater Condition differently than (Bergmann 1997). Discussing the differences would take a rather lengthy exposition, but I think that my definition will do roughly the same work and is much more straightforward for my approach. His most recent account, for example, in (Bergmann 2006), is very intricate and takes 12 pages of careful definitions and distinctions to lay out.
If an analysis of knowledge includes an NDC, rebutters and undercutters are defeaters (rebutting and undercutting defeaters). We now define No–Defeater Externalism (NDE) as follows:
Any theory of knowledge that is a version of NDE will be safe from counterexample by CASES I and II. Most contemporary externalisms are versions of NDE. It is clause (ii) of NDE, I submit, that commits proponents of NDE to some surprising epistemological level connections. While remaining wary of level confusions, it is an exploration of these possible connections to which I now turn.
a weak connection principle
Recall the connection principle mentioned above:
I would wager that the gut reaction of an externalist is that BK is false.†By 'S considers whether S knows that p' I mean to imply that S understands the proposition that S knows that p—someone who lacks a concept of knowledge does not understand this proposition. So a creature lacking a concept of knowledge always satisfies BK (trivially). The externalist intuition is that S knows that p if and only if S's belief that p is suitably connected to the truth of p. There is no reason to suppose S has access to this truth–connection. Suppose S forms the belief that p in a suitably truth–connected way, so that S knows that p, according to Simple Externalism. We can suppose further that S is not aware of this truth–connection (perhaps it has to do with reliable cognitive mechanisms). To really make the point, the externalist might have us imagine that S comes to accept radical skepticism, while maintaining her object–level beliefs. If S considers whether she knows that p in this situation, the argument goes, she will plainly believe that she does not know. So, the externalist might conclude, S can know that p, consider whether S knows that p, and believe that she doesn't know. If one accepts externalism, BK seems false.
But things aren't as they seem. In the envisioned scenario, S now has an undercutting defeater for her belief that p (her belief that she doesn't know that p). A proponent of NDE must conclude, contrary to appearances, that S does not know that p. She might have known before considering the proposition, but once she acquired a defeater her knowledge vanished.†I would argue to the contrary that, given her acceptance of skepticism, S didn't even know before considering. But we can suppress this detail for the moment. This suggests that NDE is committed to BK. Let's see if we can draw this out a bit.
Assume NDE. Upon considering a proposition, one must take some sort of doxastic attitude towards that proposition. If one considers p, then one comes to either believe p, disbelieve p, or withhold judgment with respect to p. (Why use `disbelieving p' rather than the simpler `believe not–p'? I need to speak of attitudes towards p and believing not–p is an attitude towards not–p rather than p. Aren't they the same? I am hesitant to equate disbelieving p with believing not–p. For suppose p is maximally complex of S, i.e., such that if it were any more complex, S would be computationally unable to understand it. Suppose S disbelieves p. S does not thereby believe not–p since not–p is more complex than p, and p was already maximally complex: S could not believe not–p because could not even understand not–p. Thanks to Nathan Ballantyne for bringing to my attention this method of denying the equivalence.) Let us imagine S considers whether S knows that p. We will survey each of the possible attitudes that S can take towards this proposition. In each case, I will show, BK is satisfied.
- Suppose in the first case that S comes to believe that S knows that p. If S does in fact know that p (there is nothing to prevent it in this case), then S satisfies BK.
- Suppose in the second case that S comes to disbelieve that S knows that p. This disbelief is an epistemically appropriate for withholding judgment with respect to whether S's grounds for p are indicative of p's truth. But then this disbelief is an undercutting defeater for S's belief that p. Ipso facto, S doesn't know that p. So the antecedent of BK is false and BK is true.†In the full version of this paper, I offer a full defense of the claims that disbelief that one knows that p and withholding with respect on whether knows that p constitute undercutting defeaters for the belief that p.
- Suppose in the final case that S comes to withhold judgment with respect to whether S knows that p. As with the disbelief that p, this is an epistemically appropriate basis for withholding judgment with respect to whether one knows that p. So it too is an undercutting defeater for S's belief that p. Ipso facto, S doesn't know that p. So the antecedent of BK is false and BK is true.
I have canvassed all the possible outcomes of S's considering whether S knows that p. In all outcomes, BK is satisfied. I conclude that No–Defeater Externalism is committed to a weak connection principle, and thus to (at least) one level connection. But I anticipate a possible objection. All I have shown, the objection might go, is that considering whether S knows that p can cause S to lose his knowledge that p. BK asserts something stronger: simply by knowing that p, one cannot help but believe that one knows that p, if one considers it. So I have not shown that proponents of NDE are committed to BK.
The first thing I would like to point out in response is that ``all that I have shown'' is, if not BK, still a significant and surprising level connection to which proponents of NDE are committed. And, as I will point out in the conclusion, it is sufficient to generate the Easy Come, Easy Go phenomenon. So even if the objection goes through, the result hardly trivializes my main point. That said, I think this objection trades on an ambiguity in BK. We can disambiguate BK as follows:
The objection accuses me of asserting that BK* is entailed by NDE, though it plainly isn't. As pointed out above, one can know that p at t and (also at t) consider whether one knows that p, and fail to believe at t+ that one knows that p. In this case one wouldn't know that p at t+ (because of the undercutting defeater), but one would still have known at t that p. So, the objection goes, I haven't supported BK*. But I didn't intend to. All I intended was to support BK**, and I have.
An upshot is that basic knowledge is unstable. Upon considering whether one knows, one may not be in a position to judge the reliability of the belief source. Many agents will withhold judgment in these circumstances. When this happens (as it often must when reflecting on basic knowledge) BK** entails that one no longer knows. Knowledge that was easy to get vanishes upon closer examination. This elusive knowledge has long been a feature of contextualist epistemology.†See (Lewis 1996). It is surprising to find something similar in invariantist NDE.
Stronger Connection Principle
BK'' says nothing about whether S's belief that S knows that p is warranted. A stronger connection principle says that it must be. That is, simply by knowing that p, and considering whether one knows that p, one is in a position to be warranted in believing that one knows. We can state this connection principle as follows:
Are proponents of NDE committed to WK? Some initial support comes from CASE II above. This was one of the cases that initially motivated the acceptance of an NDC. S believes she is a (reliable) clairvoyant and that this is the ground for her belief that p. Of course, the belief that p is defeated and so S doesn't know that p. Given her unreasonableness, she might, upon considering whether she knows that p, form the belief that she does know that p.†I say S is unreasonable because she persists in her belief that q ($\ulcorner S$ is a clairvoyant$\urcorner$) in the face of good evidence to the contrary. This belief would have little in the way of warrant, given that her belief that p is defeated.†A defeater for an object–level belief p is presumably a defeater for the meta–belief that one knows that p. So here's a case in which S fails to satisfy the consequent of WK and also fails to satisfy the antecedent, and so satisfies WK. And it looks like Chisholm's diagnosis works here: S's belief that she knows that p is unwarranted for the very reason that S's belief that p is unwarranted. So perhaps No–Defeater Externalists are committed to WK.
But we shouldn't move so quickly. The reason that CASE II satisfies WK is because, even before S considered whether she knew that p, she already had a defeater for her belief that p (her evidence that she is not a clairvoyant). Her subsequent unwarranted belief that she knows that p adds nothing by way of defeaters. We can explain this quite well on the explicit formulation of WK that we have. At t, S does not know that p (she has a defeater) and so the antecedent of WK is not satisfied. Anything else in the case is irrelevant. To show that WK holds, we must show that all cases in which an agent S knows that p at t and t+ and at t considers whether S knows that p, are cases in which S believes at t+ and is warranted in believing at t+ that S knows that p.
It looks like we can't show this. Consider an example. S believes that p in a way sufficient for (No–Defeater Externalist) knowledge. However, S has no access (perhaps she has forgotten) to the grounds of her belief. S has a tendency towards epistemic inertia. S happens to consider whether she knows that p and notices that if she disbelieves or withholds, she'll have to update her attitude towards p. Due to epistemic laziness (updating requires so many computation cycles, after all) she comes to believe that she knows that p.†S might not rehearse or be aware of any of this; I'm just describing the processes that underlie S's belief that she knows that p. So her belief that she knows that p is unwarranted. Neverthelelss, at least before she considered whether she knows that p, she was, ex hypothesi, perfectly warranted in her belief that p.
The question we have to ask is this: after S has formed the unwarranted belief that she knows that p, does she still know that p? I don't see why not (assuming NDE, anyway). If she no longer knows, a defeater must be available for S's belief that p that wasn't available before she considered whether she knows that p. I don't see one. S plainly does not have the belief that p is false or that her grounds for p aren't indicative of p's truth. Does S have anything which is an epistemically appropriate basis for either of those beliefs? She doesn't have an epistemically appropriate basis for the belief that p is false. And it looks like S doesn't have an epistemically appropriate basis for the belief that her grounds for p aren't indicative of p's truth. True, S is unwarranted in her belief that she knows that p, but S isn't aware of this and so it can't serve as a defeater. We can tentatively conclude that this is a counterexample for WK. It seems, then, that NDE are not committed to WK.
To test WK*, we need a case where S believes and is unwarranted in believing that S knows that p, and is aware that this belief is unwarranted. WK* will be established if this state of affairs at t+ entails that S doesn't that p at t+. So consider a case in which S has the following belief:
- \ulcorner$My belief that I know that p is unwarranted.$\urcorner$
This belief is plainly an undercutting defeater for the belief that the agent knows that p. Is it thereby a defeater for the agent's belief that p? It seems so. It is an epistemically appropriate basis for withholding judgment with respect to whether one knows that p. Withholding judgment with respect to whether one knows that p is an epistemically appropriate basis for withholding judgment with respect to p. If '...is an epistemically appropriate basis for...' is transitive, then the agent in this case has an epistemically appropriate basis for withholding judgment with respect to p, and so has an undercutting defeater for his belief that p. I think the '...is an epistemically appropriate basis for...' relation is transitive, so I conclude that this is a case where the has a defeater for his belief that p. But then we should conclude that having an unwarranted belief that one knows, and being aware that this belief is unwarranted, is sufficient to defeat one's knowledge that p. This establishes that NDE is committed to WK*.
WK* teaches us about an important level connection: coming to believe that one has an unwarranted belief that one knows that p will cause one to lose the knowledge that p.
knowledge: easy come, easy go
I have shown that No–Defeater Externalists are committed to the following two connection principles:
B** tells us that if one comes to withhold judgment with respect to whether one knows, or disbelieves that one knows, one will lose one's knowledge; WK* tell us that, even in the event that one does believe that one knows, further reflection that results in a defeating attitude about this belief will eliminate one's knowledge. These results are fully general. Reflection on our object–level belief often leads to second–order beliefs about our epistemic credentials. We can then form third–order beliefs about the quality of those second–order beliefs (perhaps this is where skepticism begins). And so on. How far up this ladder we go depends on our cognitive resources and intellectual and practical concerns. But if, at any level, we form a defeating attitude toward a lower–level, that whole knowledge structure will collapse, object–level belief included. We might call these reflective knowledge failures. They make basic knowledge easy to lose.
I suspect such failures are common. Human justificatory practices are partly responsible for this. Consider a simple case of forgotten evidence or basic perceptual knowledge. If a peer challenges us for justification in such a case, we may have nothing to offer other than a mere seeming–of–truth. If our interlocutor is sufficiently skeptical (and reliable) or has counter–evidence, and we are honest, we will lose our knowledge. The only other option is to make a faulty inference (to evidence or reliability) of the sort that engenders Cohen's problem of easy knowledge. Practical interests are also responsible for this phenomenon. Consider a case where the cost of false positive are high, i.e., a judgment that a rocket is ready to launch. A responsible agent will reflect on the credentials of her belief, and be ready to withhold judgment at the slightest evidence of faulty belief–formation. The result willl be a reflective knowledge failure.
This is an interesting result. One benefit of externalism is its ability to readily explain ordinary knowledge: the sort of knowledge that gets us around the world everyday, and that explains the cognitive behavior of non–human animals; basic knowledge and unreflective knowledge.†The phrase non–human animals is borrowed from Nick Smith. See, for example, his (2002). But No–Defeater Externalism commits us to the view that beings with reflective capabilities can lose that knowledge surprisingly easily. Why is it valuable that our epistemic concepts and rational self–reflection make our knowledge easy to lose? The traditional internalist answers won't help because they depend on a notion of knowledge that is difficult (if not impossible!) to come by. What is needed is an externalist explanation. According to one line of thought, such an ability is a safety valve for beings capable of high–stakes action. According to another, such an ability facilitates the justificatory practices that are so important to cooperation, agreement, and social standards of information transfer. There are, no doubt, other explanations. Whichever is offered, it is clear, I think, that modern No–Defeater–Externalists have a new phenomenon to discuss. And I'm inclined to think that these explanations will involve just the aspects of human epistemic life that internalists have thought externalists unable to account for. But that's a large topic that I leave for another day.