Accidental Truth and Would-be Knowledge
San Francisco State University
Nowadays the traditional quest for certainty seems not only futile but pointless. Resisting skepticism no longer seems to require meeting the Cartesian demand for an unshakable foundation for knowledge. True beliefs can be less than maximally justified and still be justified enough to qualify as knowledge, even though some beliefs that are justified to the same extent are false. Yet a few philosophers suggest that there is a special sort of justification that only true beliefs can have. Call it 'full justification' or simply 'warrant.' One such philosopher is Trenton Merricks. He takes warrant to be "that, whatever precisely it is, which together with truth makes the difference between knowledge and mere true belief," and argues that only true beliefs can have it.1 In his view, then, warrant makes the difference between knowledge and mere belief. Interestingly, Merricks does not concern himself with the nature of this remarkable property. He prefers a "formal characterization" of warrant as the "gap filler" between knowledge and mere true belief. Whatever warrant is exactly, a warranted belief cannot be true accidentally, for then the belief would not qualify as knowledge.
Merricks' strategy is to avoid "assuming any particular analysis of warrant" and argue that infallibilism about warranted belief is "a condition of adequacy of any analysis whatsoever" (p. 842). I believe that this strategy betrays him. He presents a series of subtle arguments for the thesis that warrant entails truth, but he never asks what warrant could be such that necessarily any belief that has it is true. I will argue that there is no nontrivial answer to this question, and that in the interesting sense of 'warrant,' false beliefs can be warranted. Specifically, a false belief can be warranted in the sense that if it were true, it would not be true accidentally. This is the gap-filling property, the property that makes the difference between knowledge and mere true belief, but beliefs don't have to be true in order to have it.
My purpose in showing this is not merely to rebut Merrick's thesis but
to shed some light on the epistemologically important distinction between
accidental and non-accidental truth. It is intuitively obvious that an
accidentally true belief cannot be knowledge, but it is less than obvious
what makes a belief not accidentally true. Merricks views warrant as the
property that does this but, as I point out in the first section, his definition
needs to be understood in a way that avoids making it trivially true that
warrant entails truth. Section 2 rebuts Merricks' argument that if a false
belief could be warranted, it would be possible for the belief to be accidentally
true and yet still be warranted, contrary to the definition of warrant.
Section 3 points out that even on a fallibilist view of justified belief,
according to which justified beliefs can be false, being justified can
still explain why a true belief is true. This suggests why justified false
are generally warranted: in most cases if such a belief were true, it would
not be accidentally true. Section 4 explains how false beliefs can have
the property of being, if true, not accidentally true. Section 5 argues
that the distinction between accidental and non-accidental truth is not
entirely explained by the difference between the presence and absence of
what Merricks calls the "all-important relevant connection" between
a belief's being true and why one holds it. This suggests, as indicated
in the concluding section, that warrant, if understood as the gap-filling
property, is a derivative notion with no independent epistemological interest.
1. Accidental Truth and the Definition of Warrant
To make clear what is and what is not at issue, Merricks is careful to distinguish infallibilism about warrant, which he is defending, from infallibilism about justification, which he does not endorse (p. 843n). Infallibilism about justification is implausibly restrictive, for it entails that beliefs which are less than maximally justified do not qualify as knowledge. It thereby excludes most of those beliefs that common sense regards as knowledge. Most such beliefs are defeasibly justified, although their justifications are not in fact defeated (assuming they are cases of knowledge). They must be more than minimally justified, but they do not have to be maximally justified to qualify as knowledge. Let us say that such beliefs are "adequately justified" (for knowledge).2 However, equally justified belief can be false and therefore not be knowledge. So the difference between warrant and mere justification is not a matter of degree. The existence of Gettier examples, adequately justified true beliefs that do not qualify as knowledge, shows that warrant involves something more than justification to at least a certain degree. For if that degree is high enough to exclude Gettier examples, it is also high enough to exclude some cases of knowledge. That is, there is no point on the scale of justification such that all and only beliefs justified to that degree or more qualify as knowledge. Thus when philosophers speak of "full" or "complete" justification, they mean not maximal but rather undefeated justification.3
What, then, is the difference between warrant and mere justification? Whatever justification is exactly, presumably being justified is the property that conduces to a belief's being true, although, given fallibilism about justification, being justified is not sufficient for being true. Now whereas a justified true belief can be accidentally true and thereby not qualify as knowledge, warrant precludes accidental truth. That is, justified true beliefs can be Gettier examples, but warranted true beliefs cannot be. So warrant is not mere justification. Merricks never says what warrant is-he is officially neutral on that-but presumably it includes something else along with (adequate) justification.
Having defined warrant as the property that makes the difference between knowledge and mere true belief, Merricks makes an important observation which, unfortunately, suggests a trivial interpretation of his thesis that warrant entails truth. When he observes that a belief which is accidentally true cannot qualify as knowledge, he explains that "a belief is accidentally true for one, to a first approximation, if its being true has no relevant connection to the reasons for, or processes involved in, one's holding the belief" (p. 843).4 Since warrant is the property that converts true belief into knowledge, accidentally true beliefs cannot have it-with them the relevant connection cannot hold. Here Merricks needs to be careful to avoid making it trivially true that false beliefs cannot be warranted. For if this property is defined in terms of the relevant connection actually obtaining, then of course only true beliefs can enter into it. Defining it that way would be like defining effort such that only winners can make one.
If Merricks' thesis is not to be trivialized. I suggest we think of
warranted beliefs as being, if true, not accidentally true. This definition
must be understood counterfactually if trivialization is to be avoided.
That is, a belief is warranted iff, if it were true, it would not be true
accidentally. In this way the possibility of a false belief's being warranted
is not ruled out by definition, so that a substantive argument is required
to show that only true beliefs can be warranted. It must be argued that
no false belief can be such that if it were true, it would
not be true accidentally.
2. Falsity and Accidental Truth
Given fallibilism about justification, there is a sense in which almost any true empirical belief, however strongly justified, is lucky to be true.5 After all, something might have obtained which would have defeated its justification. Even so, being lucky in this sense does not prevent the belief from qualifying as knowledge-its justification can be defeasible without actually being defeated. Merricks' examples of justified but accidentally true beliefs are lucky to be true in a stronger sense. They are Gettier examples. They are accidentally true because the "relevant connection" (between their being true and why they are held) does not obtain. It is on this conception of accidental truth that Merricks argues that if false beliefs could be warranted, then so could accidentally true beliefs, contrary to the definition of warrant.
The argument is this: if a false belief could be warranted, it would be possible for this belief to be accidentally true; but if it were accidentally true, it would not be warranted, by the definition of warrant; so it is not in fact warranted. The argument can be put in the jargon of possible worlds. Suppose that in the actual world W a certain belief, e.g., that Jones owns an Escort, is warranted but false. Compare W with a possible world W* in which everything is the same except for some remote happenstance that makes the belief true. Suppose, for example, that thousands of miles away Jones's aunt dies and bequeaths Jones an Escort, shortly before the belief in question is formed. Merricks asks, "how could the far away and unknown death of Jones's aunt make my (otherwise warranted) belief unwarranted?" (p. 850). The only relevant difference between W and W * is that in one the belief is false and in the other it is accidentally true.6 Merricks observes that "there is, presumably, no difference É with respect to how or why I believe that [Jones owns an Escort]," from which he concludes, "how can these worlds differ with respect to the warrant enjoyed by my belief?"
Unfortunately, this pattern of argument can also be deployed to yield the opposite result. Suppose we compare W with a world W! where the belief is non-accidentally true, i.e., is a case of knowledge. Suppose that the only relevant difference is that in W! things are as the holder of the belief supposes (Jones owns the Escort he is seen driving in), but in W, shortly before the belief is formed, a divorce court thousands of miles away awards Jones's Escort to his wife. There is no difference between W! and W with respect to how or why the belief that Jones owns an Escort was formed. So, to paraphrase Merricks, how can these worlds differ with respect to the warrant enjoyed by the belief? How could the far away and unknown event of a judicial decision make any difference to the belief's warrant?
These questions are no less rhetorical than Merricks', but they suffice to show that his rhetorical questions boomerang when the hypothesized warranted false belief is put in a more favorable light. The comparison Merricks makes, between W and W*, is an invidious one, making it seem problematic that a warranted but false belief would have lacked warrant were it to have been made true solely because of some remote event. But if we compare W! with W, it should be equally problematic that a warranted true belief would have lacked warrant were it to have been made false solely because of some remote event. So the real problem seems to be with Merricks' argument. If his argument shows that the possibility of false warranted beliefs implies the possibility of accidentally true warranted beliefs, then the same form of argument shows that the possibility of non-accidentally true warranted beliefs implies the possibility of false warranted beliefs.
This argument commits a simple modal fallacy. The fact that it is possible
for a warranted but false belief to be accidentally true, in which case
it would not be warranted (by the definition of warrant), does not
show that the belief is not in fact warranted. The reason the belief
is warranted in W and in W! but not in W*, where it is accidentally true,
is the same reason the belief is not knowledge in W*. Merricks thinks that
the belief can't go from being warranted in W to being unwarranted in W*
because he implicitly assumes that being warranted depends solely on a
belief's pedigree and not on its circumstances. That is, a belief's "being
warranted is determined by the reasons for, and causes of," one's
holding it (pp. 850-1).7 This is the reason for his rhetorical question
about how a far away and unknown event could make an otherwise warranted
belief unwarranted. But if such external events can affect knowledge itself,
then surely they can affect warrant (besides, as Merricks acknowledges
(p. 841), warrant may be "a messy item," whose "analysis
might be full of disjuncts and conjuncts and conditionals and caveats").
For warrant, by Merricks' own definition, is that which makes the difference
between knowledge and mere true belief. So he should not assume that if
a false warranted belief were accidentally true, it would still be warranted.
It lacks warrant in precisely those circumstances in which, though true,
it is not knowledge.8
3. Would-be True Beliefs
I have suggested that warranted beliefs have the counterfactual property of being, if true, not accidentally true. Now how can false beliefs have this property? Merricks' comparison of warranted false beliefs and Gettier examples makes it seem as though the only hope a warranted false belief can have of being true is to be a Gettier example. He seems to assume that if a belief is false (however well justified it is), it would have been true accidentally if it had been true at all. The more likely course of events, however, is for such a belief to have been a case of knowledge if it were true. So it seems that the difference between a warranted false belief and a Gettier example is greater than that between a case of knowledge and a warranted false belief. In the above example, W! is closer to W than W is to W*. W* involves the occurrence of an improbable event that, by coincidence, just happens to make the belief true, quite apart from how well justified it is.
With this in mind we can ask whether a false belief that is adequately justified would, if true, be accidentally or non-accidentally true. Forget about warrant for the moment and consider (adequately) justified beliefs. Justified beliefs that are accidentally true (Gettier examples) are flukey. Ordinarily, justified beliefs are not lucky to be true (when they are true) but unlucky to be false (when they are false). Fallibilism about justified belief implies that there are ways in which belief-forming processes can go wrong, but it suggests that these sources of error are themselves part of the normal course of events. For example, it is not out of the ordinary to misremember that one left one's glasses on one's desk rather than on the sink but rather extraordinary both to misremember this and to be right about their whereabouts anyway, say because someone moved them from the sink to the desk. For justified beliefs, error is normal as compared to accidental truth. Short of being knowledge, adequately justified beliefs are more likely to be false than to be accidentally true.9 In possible-worlds jargon, worlds in which a justified belief is false are more normal than worlds in which the belief is accidentally true. They are closer to worlds in which the belief is non-accidentally true than to worlds in which the belief is accidentally true. Thus we can say, by applying the standard method of evaluating counterfactuals, that a justified belief has the property of being, if true, not accidentally true.
If it were not relatively normal for a justified belief to be false as compared to being accidentally true (to being a Gettier example), then being justified would not be that property which conduces to beliefs' being true and which (generally) explains their being true when they are true. Whereas being justified explains, in normal cases, why a belief, if true, is true, in Gettier circumstances it fails to explain why the belief is true. It may seem that this notion of explaining a belief's being true when it is true is objectionably weak, but ordinary explanation is weak in the same way. Compare the case of a match that lights when struck. Even though the match wouldn't have lit if it were wet, the fact that it lights is explained by the fact that it was struck. That it was struck explains, if it lights, that it lights. Being justified for a belief is like being struck for a match. Belief-forming processes that generally lead to true beliefs can also lead to error, just as a match that generally lights when struck can fail to light. And, just as an unusual kind of match, say one that lights when punctured, might light when it is struck but not because it is struck (maybe it lights because it is also being touched to a very hot surface), so a belief that is formed via a process into which error has been introduced might still turn out to be true.
Justified false beliefs can have the counterfactual property of being,
if true, non-accidentally true. If that is what being warranted entails,
they are generally warranted (the exception is the case where the belief
would be a Gettier example if it were true). Equivalently, a warranted
false belief is such that, if it were true, it would be true because it
is justified. Now spelling out what it is for a belief to be true because
it is justified would be tantamount to solving the Gettier problem, and
I am not proposing to do that here. I can only repeat that the Gettier
problem presupposes fallibilism about justified belief, since otherwise
a belief could not be both true and (adequately) justified without being
an instance of knowledge. If what makes a justified true belief an instance
of knowledge is the fact that the relevant connection obtains between its
being true and how and why it was formed, and if that's what it is for
a true belief to be warranted, then, in general, there is nothing to prevent
a false belief from having the counterfactual property of being such that
this connection would obtain if the belief were true, i.e., that its being
justified would explain its being true. Presumably the process by which
the belief was formed is sensitive to truth-conducive considerations, so
that the person would have believed differently (formed a different belief
or withheld belief) if his circumstances were different. That is why it
is bad luck when a justified belief is false and extraordinary luck when
a justified belief is true but not because it is justified (whatever exactly
that means). Accordingly, when we consider what if, contrary to fact, a
false but justified belief were true, we may assume, barring special reason
to the contrary, that the conditions obtain under which beliefs that are
justified are true. To suppose otherwise would be to suppose that justified
beliefs are just lucky to be true when they are true, and not reliably
4. Would-be Knowledge and Would-be Gettier Examples
A Gettier example is a belief which, even though it is justified, is lucky to be true, in the same way that an unjustified true belief is lucky to be true. So a "would-be Gettier example" is a justified false belief which, were it true, would be lucky to be true. In this way, it differs from an ordinary justified false belief, which would, if true, not be true accidentally. The difference between the two suggests that it takes special circumstances for justified false beliefs not to be warranted. They are not unwarranted just because they are false.
To appreciate the difference between the two cases, compare the following situations. First, suppose you believe by observation that Jones owns an Escort. Your belief that he does is justified but, as it happens, is false. It is justified because you are very good at identifying makes of cars, but it is false because in this case you have mistaken a rare Eastern European knock-off of an Escort for the real thing (Jones owns the only imitation Escort in the U.S.). This is an ordinary case of justified false belief. Next suppose that you believe that Jones owns an Escort, this time from seeing Jones driving a real Escort, but, unbeknownst to you, just a few minutes ago a bankruptcy court deeded Jones's Escort over to a bank. Meanwhile, Jones's aunt, who has died in Timbuktu, has bequeathed her Escort to Jones but, unfortunately, her executor has just learned that he must sell the Escort to pay off her debts. Your belief that Jones owns an Escort is again both false and justified, but in this case it is not an ordinary justified false belief. It is a would-be Gettier example. But for Jones's aunt's debt, your belief that Jones owns an Escort would be true. Even so, it would be lucky to be true, in just the way that an actual Gettier example is lucky to be true.
The similarity of would-be Gettier examples to full-fledged Gettier examples helps highlight the difference between both of them and ordinary justified false beliefs. An ordinary justified false belief is unlucky not to be true because it has what it normally takes to be true. One is steered wrong on grounds of the same sort on which one is normally steered right. In a Gettier situation the justification, though adequate, is defeated. In the case of a would-be Gettier example, the belief's justification would, under its (abnormal) circumstances, be defeated even if it were true. Whereas a case of would-be knowledge, an ordinary justified false belief, is a belief that is unlucky to be false, a would-be Gettier example is doubly unlucky. Like any justified false belief it is unlucky to be false, but it is also unlucky because, if true, it would be lucky to be true.
The difference between ordinary justified false beliefs, which are would-be
cases of knowledge, and would-be Gettier examples shows that the difference
between being warranted and being merely justified applies to false beliefs
as well as to true ones. Cases of would-be knowledge, unlike would-be Gettier
examples, have the property of being, if true, not accidentally true.
5. Accidental Truth and Misleading Evidence One Does not Possess
We have been assuming all along that for a true belief to be knowledge it must not be accidentally true. But is that sufficient? Perhaps not, on account of misleading evidence one does not possess. This is evidence that would, if one possessed it, defeat the belief's justification and thereby undermine its claim to knowledge. For example, a justified true belief might fail to be knowledge on account of an inaccurate story in a reliable newspaper that one hasn't read.10 One's knowledge is undermined even though one hasn't read the story, for if one had read it (it's only a matter of good luck that one hasn't) and if this story had not yet been discredited, one would have relied on it and believed differently.
Does the phenomenon of misleading, knowledge-undermining evidence one does not possess show that non-accidental truth is not sufficient for knowledge? That would be so if accidental truth always involved a break in the "relevant connection" in terms of which Merricks defines accidental truth. For such evidence does not break the connection between the truth of the belief and the reasons for, or processes involved in, one's holding the belief. In the newspaper example, the belief would have been held whether or not the unread story had been published. Here there is no question of the relevant connection not obtaining, because the belief would have been formed in the same way, on the same grounds, with or without the existence of that evidence. Evidence one does not possess does not impinge on the process whereby the belief is formed.11
It seems to me, however, that the belief in question is accidentally
true, at least in a sense. What is shown by the case of misleading evidence
one does not possess is that the absence of the relevant connection is
not the only way for a belief to be accidentally true . Even though such
evidence does not enter into the reasons for, or processes involved in,
one's holding the belief, one's belief is accidentally true because it
is accidental that one does not possess this evidence. This evidence undermines
one's knowledge not because the relevant connection is severed, but because
it is accidental that this evidence did not get into one's possession and
thereby keep one from forming the belief. What is lucky is not that the
belief is true but that, because it was fortunate enough not to be challenged
by this evidence, it was even formed. What is accidental is not that the
belief one has actually formed is true but that one has formed that belief,
a belief which is true.
6. Conclusion: The Unimportance of Being Warranted
False beliefs can be warranted. They can have the property that makes the difference between knowledge and mere true belief-even though they are false. Understanding how this can be requires distinguishing what goes wrong in ordinary cases of justified false beliefs from what goes wrong in Gettier examples (justified true beliefs that are not knowledge). Both are cases of beliefs which fail to be non-accidentally true, but Gettier examples are further from being cases of knowledge than are (adequately) justified false beliefs. Gettier examples are true not because they are justified but, so to speak, in spite of it. Ordinary justified false beliefs, ones that would not be Gettier examples if they were true, are closer to being cases of knowledge-they are such that, if they were true, then, assuming things are otherwise normal, they would be knowledge. If things are otherwise normal, they would be, if true, non-accidentally true.
This counterfactual property, which a belief has in virtue of being warranted, is really just the property of not being an actual or potential Gettier example. A belief which has it is justified and is or would be true because it is justified. This is consistent with the fact that a false but warranted belief would fail to be warranted if it were accidentally true. Such a belief would lack warrant if an external and perhaps remote event undermined its status as knowledge, but this is to be expected if warrant is that property which makes the difference between knowledge and mere true belief. Given that conception of warrant, a belief ought to lack warrant in precisely those circumstances in which, though true, it fails to qualify as knowledge. This suggests that warrant, the property that fills the gap between knowledge and mere true belief, has no independent epistemological significance. If a belief is warranted, it would be, if true, non-accidentally true. That's a nice property for a belief to have, but it doesn't help us as truth seekers. For what, beyond being justified, makes a belief warranted is that it not be in circumstances that make it liable to being a Gettier example. But these are circumstances which, when we're in them, we have no reason to think we're in. As truth seekers we are on the lookout not for evidence that we are in a Gettier situation but for evidence for or against the proposition under consideration. Actually being in a Gettier situation requires not being aware of such evidence. In other words, you can't know that you're in a Gettier situation; you can only discover evidence ignorance of which would have left you in one.
If warrant is not independently important, what is? What is important
is what beyond justification is involved in filling the gap between knowledge
and mere true belief, and that includes but is not limited to the connection
between being justified and being true. The interesting challenge is not
to give a substantive account of warrant but to explicate what it is for
a belief's being justified to explain its being true. We usually think
of a belief's justification not in explanatory but in evaluative terms,
as giving one reason for holding the belief. The distinction between accidental
and non-accidental truth, to which Merricks calls our attention, invites
us to ask what it is for a belief to be true because it is justified in
this non-causal sense of 'because,' taking into account the fact that false
beliefs can be justified and that justified beliefs can be true but not
because they are justified. To answer this question would be to solve the
Gettier problem and thereby come to know what knowledge is.
1. Trenton Merricks, "Warrant Entails Truth," Philosophy and Phenomenological Research 55 (1995), pp. 841-55. All page references in the text are to this article. Merricks borrows the definition of warrant from Alvin Plantinga, Warrant: The Current Debate (Oxford: Oxford University Press, 1993), p. 3.
2. From here on by 'justified beliefs' I will mean, unless otherwise indicated, adequately justified beliefs. I will insert the word 'adequately' only for clarity or emphasis.
3. Scott Sturgeon, in "The Gettier Problem," Analysis 53 (1993): 156-64, uses the phrase "fully justified," and Merricks cites James Tomberlin as using the phrase "completely justified" (p. 842n). Sturgeon describes the evidence that supports such a belief as "ultimately undefeated," a phrase which Merricks (p. 841n) attributes to John Pollock, who also uses the phrase 'objective justification.' And, as Ernest Sosa explains, objective or "external" justification, whose assessment includes reference to evidence one does not possess, may be contrasted with "subjective" or "internal" justification (Knowledge in Perspective, Cambridge: Cambridge University Press, 1991, p. 10 ). Sosa argues that both are necessary for knowledge. What Merricks calls 'warrant' seems to be a combination of both.
4. From here on I will use the phrase 'relevant connection' as short for 'the connection between a belief's being true and the reasons for, or processes involved in, one's holding the belief.'
Notice that being accidentally true is not a modality, at least not of the usual kind. Whereas, for example, 'it is possibly true that p' is equivalent to 'possibly p,' 'it is accidentally true that p' is not equivalent to 'accidentally p.' Accidental truth is a property of beliefs, not of things believed.
5. In this sense, which is not the sense that figures in the present discussion of accidental truth, there is such a thing as epistemic luck. It comes in various forms: (1) a belief whose motivation is not entirely verific is lucky to be an instance of knowledge anyway, (2) a belief that is an instance of knowledge is lucky not to have been formed in a benign-demon world, (3) a justified true belief is lucky not to be a Gettier example, (4) a justified true belief is lucky that the process leading to it was not adversely affected by a source of error.
6. Merricks observes that in neither case is the belief warranted because in neither case does the truth of the belief have a relevant connection to the reasons for, or processes involved in, one's holding the belief. There is a different explanation inthe two cases for the absence of the relevant connection: in W* the belief is accidentally true, in W it is not true at all. But if this is his reason for denying that the false belief is warranted, his infallibism become trivial in the way described in section 1: obviously the truth of a false belief does not in fact stand in the relevant connection to the belief's justification.
7, In a footnote (n. 21, p. 851), he considers the further requirement that there obtain a "relation that, if [the belief] that p is true, the reasons for, and causes of, [the] belief stand in to the fact that p. one's holding it." But here he falls into the trap of trivialization pointed out in section 1, for he interprets the conditional as material rather than counterfactual. As he goes on to say, "in the example in the text, the relevant relation of S's belief that p to the truth of p is unchanged from W to W*, wince S's belief that p bears no relevant relation to the truth of p in either W or W*."
8. Merricks offers another reductio against the possibility of warranted false beliefs. Suppose that a belief could be warranted yet false. Assume that it is possible for the warrant for a false belief to be "transferred" to a true belief, albeit an accidentally true one. Suppose, for example, that one has the warranted belief that Jones owns an Escort. Then, by this assumption, one's belief is warranted that Jones owns a car (or one's belief that Jones owns an Escort or Brown is in Barcelona). Suppose this belief is true. Then it is accidentally true, hence, by the definition of warrant, not warranted. I will not address this argument because, as Merricks acknowledges, his case does not depend on it. Besides, as he points out, since "warrant may have some odd properties, we therefore have no right to make intuitive judgments about particular cases of warrant divorced from knowledge" (p. 849). This is why I would challenge his undefended assumption that forms of inferences that preserve knowledge also preserve warrant (p. 847).
9. In other words, the prior probability of an adequately justified belief's being a case of knowledge is greater than that of its being false, and the latter is greater than the probability of its being a Gettier example.
10. See Gilbert Harman, Thought , Princeton: Princeton University Press (1973), ch. 9. For an examination and diagnosis of Harman's examples, see William G. Lycan, "Evidence One Does Not Possess," Australasian Journal of Philosophy 55 (1977): 114-26.
11. It could be argued that what does enter into this process is the implicit assumption that there is no undermining evidence one does not possess (Harman's Principle Q, op. cit., p. 151) and that this assumption is false when there is such evidence. In that event one relies on a false proposition (in violation of Harman's Principle P, op. cit., p. 47), so that the relevant connection is broken. The notion of an implicit assumption and how such an assumption can play a role in reasoning are taken up in my "Default Reasoning," Pacific Philosophical Quarterly 65 (1984): 37-58 and in "A Rationale for Reliabilism," The Monist 68 (1985): 246-63.