I want to harass premise 1 of the Argument from Religious Diversity (ARD).
    PK has done both well and praiseworthily to have framed the epistemological issue in terms of Reliabilism, because the Reliabilist
theory of knowledge and justification is his strongest enemy:  The Reliabilist is the epistemologist best equipped to resist 1, so if PK can
defend 1 against the Reliabilist, no more need be said.
    (In this handout I will argue on the Reliabilist's behalf.  In real life I am not a Reliabilist, but an Explanationist; see again "Explanation and Epistemology".  What an Explanationist should say about the ARD is not clear; I will leave that as an exercise or paper topic.)

    The Reliabilist objection to 1:  PK's statement of the Reliabilist view is a good one to begin with.  Let's call it "Basic" Reliabilism:  S knows that P iff (i) P, (ii) S believes that P, and (iii) S's belief was produced by a reliable process, i.e., a process likely to generate truths.  (Actually the Reliabilist distinguishes between knowing and merely being justified in believing, which in epistemology is a gigantic distinction, but I don't think the difference affects the ARD in particular, so I'll ignore it here.)
    The obvious Reliabilist objection to premise 1 is that S's religious beliefs may have been produced by a reliable process and their competitors not.  Even if the other, rival traditions are (in the words of 1) apparently as reliable as S's own, it hardly follows that they are in fact as reliable as S's own.  If S's tradition is highly reliable and significantly more so than any of the competing traditions, the according to the Reliabilist S's religious beliefs do count as knowledge, despite the apparent reliability of the other traditions.  (Notice that religious traditions offer very high degrees of reliability, hyper-reliability.  If your religious tradition has The Big Guy Himself behind it, that's reliability!  The Roman Catholic Church adds the mechanism of Apostolic succession, which--whatever its political drawbacks--I myself think has a lot to be said for it epistemologically.)
    It is important to see that for the Basic Reliabilist, what's criterial for knowledge is that S's belief be produced by a reliable process, not that S can know that it was so produced or has any reason to think it was, or even can produce a plausible argument for the belief.  Basic Reliabilism is in that sense an "externalist" theory of knowledge and justification:  What matters are the nomic relations that actually hold between the subject's beliefs and the relevant parts of the world, not any attitude the subject might take toward those relations or indeed any access the subject might have to them.  No such access is required.

    First defense of 1 against the Reliabilist objection:  Someone surely will protest, on behalf of 1, that even if (say) the Roman church actually is the One True Church and all competing religious doctrines are false, and even if the RCs' religious beliefs are the product of a hyper-reliable mechanism, the RCs have no way of knowing that.  Their beliefs are not "apparently" or visibly any more reliable than are those of the opposing traditions.  The RCs are right only by luck--they were lucky to have been born into and brought up in the correct faith--but, as PK said, if you're right only by luck you do not know.

    Reliabilist rejoinder:  That protest is an ignoratio elenchi: it gets the Reliabilist position wrong.  The ARDist's present complaint is that the RCs have no way of knowing that their beliefs are reliable.  But all that shows is that the RCs do not know that they know.  To know that P, the Reliabilist maintains, is for one's belief that P to be the output of a reliable process.  So, according to our present hypothesis, the RCs do know that their doctrines are true, even though they do not know that they know that.  (For them to know that they know, their belief that they know would have to be itself the product of a reliable process.)<1>
    (Historically, some philosophers have defended the claim that knowing entails knowing that one knows (this is usually called "the KK thesis").  But that claim is at best a minority position.  It was not widely believed in the 20th century, and it certainly is not granted by Reliabilists, who laugh it to scorn.)
    Moreover, it is incorrect to say that in our example the RCs are right "only by luck."  By hypothesis, their tradition has been purposefully guided and sustained by Almighty God Himself!  Given that history, it is hardly a matter of luck that their beliefs are true.  (It may be a matter of luck that a given RC is an RC and was brought up in the true faith, but (a) that that person is an RC is a different fact, and (b) given our theistic hypothesis, it's not obvious that it is a matter of luck either.)

    Second defense against the Reliabilist objection:  A second defense is to go on the offensive and attack Basic Reliabilism itself.  This is what PK did in putting forward his counterexample (the type of example originally devised by Hilary Kornblith and Laurence BonJour):  For whatever reason, A administers to B a hallucinogen that makes people have visual experiences as of rats and snakes.  And B knows this.  A also arranges for actual rats and snakes to be placed in B's path (or perhaps the actual rats and snakes are coincidental--doesn't matter).  But by a fluke, B is immune to the drug; at least, it does not work on him.  B sees the actual rats and snakes.  Despite knowing that he has ingested the hallucinogen, B irrationally accepts the testimony of his senses and believes there are rats and snakes in front of him.  That belief is true.  And it is also the product of a reliable process, because despite the drug, B's visual system is working perfectly.
    So:  B satisfies the Basic Reliabilist conditions for knowing, but intuitively he does not know, because, well aware that he has been drugged, B should not be trusting his vision.  The Basic Reliabilist analysis is not a sufficient condition for knowing.  More work needs to be done by the Reliabilist to make the analysis more demanding and rule out the Kornblith-BonJour type of case.
    (At least two leading Reliabilists, D.M. Armstrong and Fred Dretske, have refused to accept that.  They tough it out, and insist that in the example, B damn well does know.  After all, B is in perfect working order, his visual system is functioning quite normally, and as things are (the drug being ineffective) he would not be believing that there are rats and snakes in front of him unless there were.  But other Reliabilists, such as Alvin Goldman and Fred Schmitt, have agreed that Basic Reliabilism is refuted by the example, and have added epicycles to try to repair the analysis.  Let us go along, for the sake of argument.)
    So far, even if Basic Reliabilism is counterexampled, there is no immediate point to be made in favor of 1.  Even if Reliabilism needs to be tweaked, its externalist feature remains (so far as has been shown), and it is that feature that threatens 1.  So in what way did PK intend his example to defend 1 against Reliabilism in general?
    On that point I wasn't entirely sure.  He wanted to introduce the idea of a relevant range of cases, or range of relevant cases.  That idea comes from a somewhat different school of epistemology, the "Evidentialist" school, according to which you know when on the basis of your evidence you can rule out all the relevant alternate possibilities.  The qualification "relevant" is in there to fend off Cartesian skepticism, as I explained in class.<2>   Now, how is the relevant-cases notion supposed to be grafted into Reliabilism?
    Perhaps the idea is this (I seem to recall PK's saying something like it):  Even the Basic Reliabilist implicitly presumes a range of relevant cases, because a reliable process is conceived as "a process likely to generate truths," and "likely to generate truths" means something like, "usually does generate truths rather than falsehoods across the range of relevant actual and hypothetical cases."   Let's go with that understanding.
    Normally, hypothetical cases of having been fed a hallucinogen do not count as relevant.  (If, while you are reading these words, a Cartesian friend argues that you don't really know what they say because after all it's logically possible that you have earlier been fed a hallucinogen, you would be within your rights to backhand your friend and ignore that skeptical possibility as non-relevant, idle, just made-up, not a "real" possibility.)  But, PK argued, in his example the hallucinogen possibility is obviously and highly relevant:  B knows he actually has been fed the hallucinogen, which makes it overwhelmingly probable that the rats and snakes are not real.
    What should we make of that?  I believe PK's intention was to emphasize the "relevance" condition and thereby save the Reliabilist from his counterexample.  (Let us call our slightly amended version "Relevance Reliabilism.")  Since in the example B's belief-forming process ignored a hugely salient possibility, that process is not reliable.  Thus, B does not satisfy the Relevance Reliabilist's reliability condition, and  Relevance Reliabilism predicts correctly (as we are supposing) that B does not know.
    And what has that to do with 1?  I think PK's idea is that the existence of competing religious traditions is analogous to A's administering of the drug to B.  That is, if you know that there are competing traditions, the possibility that they're right and you're wrong becomes relevant for you; so your tradition is not reliable and you do not know that your beliefs are correct.

    First Reliabilist rejoinder:  I don't think the grafting of the "relevance" qualification into Reliabilism works very smoothly.  Here's why.  Suppose we agree that in the counterexample case, the possibility of hallucination is a relevant possibility.  According to my tentative interpretation of PK's argument, the inclusion of that possibility makes B's belief-forming process unreliable.  But why?  And how are we to determine that?  The proposed test is, does B's process usually generate truths rather than falsehoods across the range of relevant actual and hypothetical cases?  Whether including the hallucination possibility in that range changes the answer from Yes to No is not at all clear, because we don't know the proportion (so far as that notion makes sense) of hallucination cases to normal cases.  If that proportion is small, it may still be true that B's process usually generates truths across the range.  What argument shows that the proportion is large?<3>
    There's also a more fundamental mismatch.  Remember that Reliabilism prides itself on being an externalist view; it focuses, not on the subject's subjective state or on anything of which the subject is normally aware, but on the de facto nomic etc. relations between the subject's belief and the environment that produced it.  And in our example, because the drug did not work on B, those relations do not differ from what they would be in a normal case of seeing real rats and snakes.  The process that produced B's belief is, intuitively, still reliable, even though B himself has overwhelming reason to believe that it is not.  I find it hard to hear any sense in which the process is unreliable just because B has been fed the drug.<4>
    Rather, what I think is going on here is that the Kornblith-BonJour type of case is eliciting internalist intuitions.  The key phrase in the foregoing concession was, "even though B himself has overwhelming reason to believe that it is not."  Two mutually related complaints may be made against B, and both are correct.  First, B is ignoring highly pertinent evidence that epistemically defeats the evidence of his senses.  Second, B is being epistemically irresponsible; he is violating a norm of rationality.  For each of those reasons, B fails to know.  But notice that they are internalist reasons, not really features of B's belief's own nomic etc. connection to the external world.  The import of Kornblith-BonJour is to impeach Reliabilism by attacking its externalism (both Kornblith and BonJour intended their examples in exactly that way).  I believe this is why PK cannot easily graft the "range of relevant cases" feature into Basic Reliabilism.  The example should make us give up Reliabilism, not merely amend it.<5>
    Fine, PK may say; if we give up Reliabilism, we lose the original objection to 1.  But we also lose the nice dialectical feature I mentioned in the beginning: It would be better for PK to confront the ARD's toughest adversaries, who are still externalists and will reject 1 for the same reason as before.

    Second Reliabilist rejoinder:  PK's example is not, or not perfectly, analogous to the alleged undermining of one's religious beliefs by the existence of competing traditions.  In the example, B's visual evidence is rendered utterly worthless by his knowledge that he has been given the drug.  He has no reason at all to believe in the rats and snakes, and plenty of reason to believe they are unreal, because he has positive reason to believe he would be hallucinating rats and snakes even though (probably) there would be no real ones anywhere about.  But regarding 1, the competing religious traditions don't do anything that dramatic.  It's not that their existence gives you strong positive reason to believe that you would hold the religious beliefs you do even though (probably) there is no God of the sort you believe in.  At worst, the competing traditions merely raise the question of why you should think that your tradition is more reliable than theirs.  And that, to the Reliabilist, is a secondary issue.
 
 

1.  PK said something in passing (if I understood it correctly) that, in light of this rejoinder, is wrong:  I believe he said that the Reliabilist offers no strategy against the global skeptic, but can only measure the reliability of a process by presupposing the reliability of the measuring process being used.  It is true that one cannot measure reliability without presupposing the reliability of something else, but it does not follow that the Reliabilist has no strategy against the skeptic.  The Reliabilist makes the following reply to the skeptic:  If my belief is in fact the product of a reliable process, then it is a case of knowledge.  Merely made-up skeptical scenarios are irrelevant.  To convince us of skepticism, the skeptic would have to show that our perceptual, memory etc. faculties are unreliable; the skeptic has not done that, but has only suggested that they may be unreliable.
    Here too it is tempting to protest that the putative de facto reliability of our senses and other faculties is of no use against the skeptic, because de facto reliability is an empirical matter of how things stand in the real world--the actual relation of our beliefs to external states of affairs--and that is just what the skeptic is questioning in the first place.  Even if the beliefs are products of a reliable process, we can't know that from the inside.  But, same reply:  At best, all that shows is that we don't know that we know.  If our beliefs are in fact products of a reliable process, we do know, whether or not we know that we do.

2.  If evil-demon cases and other fanciful skeptical scenarios are allowed in, of course our evidence does not rule them out, since such scenarios are deliberately designed to be compatible with all our evidence.  That was Descartes' point in the Meditations; he felt he had to supplement our empirical evidence by proving the existence of a benign God, in order to rule out the Evil Demon.  Contemporary "Contextualist" Evidentialists instead use the "relevance" qualification to rule out fanciful, made-up, idle etc. skeptical hypotheses.
    Actually Descartes was both an Evidentialist and a Reliabilist.  He worried that in light of the Dream and Evil Demon arguments, our sense organs would not count as reliable.  Proving the existence of God also, he thought, restores reliability.  But the modern Reliabilist holds that sense organs need only be truth-conducive in the real world and in worlds like it that are pretty normal.

3.  Contrast a case originated by Carl Ginet and made famous by Alvin Goldman:  Henry is driving in the countryside.  He stops for a rest and looks around.  In the neighboring field he sees a barn.  It’s a real barn, and the light, his vision et al. are perfectly normal.  So far, we’d say that he knows there is a barn in the field.  But now suppose that, unknown to Henry, the rural district he has just entered is thickly populated by fake barns, papier-mâché facades.  The fakes are so cleverly made that from the roadside they look exactly like barns.  Henry has not encountered any of the fakes; the barn he is looking at is real.  But this is just by luck.  He happens to have come upon the one barn-like object within miles that is a real barn rather than a fake.  Had he been surveying any other field, he would have seen a barn facsimile, believed it to have been a real barn, and been wrong.  Received opinion among epistemologists is that Henry does not know, because he is right only by luck.  The possibility of encountering a fake barn is a (very) real, hence relevant, possibility in his context, not a merely idle or made-up one, and in the district Henry is not a reliable detector of barns.  –But there is a crucial difference between this example and the Kornblith-BonJour cases:  In the barn example, Henry’s environment is weird, and it is in virtue of the unusual environmental features that he is unreliable and fails to know.  There is nothing comparably weird or wrong about B’s environment in PK’s case.

4.  Here is where the variant example may be relevant.  In the variant I mentioned in class, A does not actually administer the drug to B, but merely gives B strong reason to believe she has administered it.  Intuitively, to me, this does not weaken the example at all.  So long as B has such reason (and so long as we are not steel-bellied Reliabilists of the Armstrong-Dretske persuasion), B's belief is quite irrational and he certainly does not know.  If it is therefore inessential to the example that B has actually ingested the drug, it is even more obvious that the drug has done nothing to make B's process less reliable.

5.  For the record, though, both Alvin Goldman and Fred Schmitt have tried to accommodate Kornblith-BonJour cases within the Reliabilist framework.  Goldman appeals to nonactual alternate belief-forming processes "available" in some sense to the subject.