McKinsey One More Time - Oxford Scholarship Jump to ContentJump to Main Navigation

Anthony Hatzimoysis

Print publication date: 2011

Print ISBN-13: 9780199590728

Published to Oxford Scholarship Online: May 2011

DOI: 10.1093/acprof:oso/9780199590728.001.0001

Show Summary Details
Page of

PRINTED FROM OXFORD SCHOLARSHIP ONLINE ( (c) Copyright Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a monograph in OSO for personal use (for details see Subscriber: null; date: 28 August 2016

McKinsey One More Time

McKinsey One More Time

(p.80) 4 McKinsey One More Time

Crispin Wright (Contributor Webpage)

Oxford University Press

Abstract and Keywords

The so‐called McKinsey paradox consists in the apparently valid derivation of apparently non‐apriori knowable propositions about the satisfaction of certain putative external conditions on content from apparently a priori knowable premises. This chapter returns to the suggestion that the correct diagnosis of this paradox discloses a failure broadly comparable to what goes wrong in Moore's notorious “Proof’ of an external world. The chapter offers a further exploration of the relevant notion of warrant transmission failure, then turns to sustain the proposed diagnosis of McKinsey in those terms, albeit with some qualifications.

Keywords:   a priori, belief, externalism, justification, McKinsey paradox, reasoning, self-knowledge, warrant


It is not always true that recognizably valid reasoning from known, or otherwise epistemically warranted premises, can be enlisted to produce knowledge, or other epistemic warrant, for a conclusion. The counterexamples are cases that exhibit what I have elsewhere called warrant transmission-failure. It is nowadays widely accepted that there are indeed such counterexamples, though individual cases remain controversial. One such controversial case is the so-called McKinsey paradox. The paradox presents as a simple collision between three claims that many would find attractive:

  1. 1. That many ordinary types of intentional self knowledge, including in a wide class of cases knowledge of what one thinks or believes, are immediately available to a normal subject, without empirical investigation (in the armchair, or “a priori” in one loose and popular sense of the phrase).

  2. 2. That psychological content is conditioned by various forms of externalist constraints, with the effect that many ordinary concepts are available to a thinker only when she, or her speech community, has had certain kinds of historical interaction with items lying in the extension of the concepts concerned; and moreover, that the fact that such external constraints operate on certain specific contents is something that can be appreciated a priori, by doing pure philosophy.

  3. 3. That it is not possible to know a priori, by armchair reflection alone, anything about the historical interaction of oneself, or others in one's speech community, with particular objects and kinds.

(p.81) Suppose water is a concept of which it may be recognized a priori that point 2 applies. And suppose that ordinary intentional self-knowledge encompasses attitudes to simple contents involving water. That sets up the following kind of reasoning:

(A) I believe that water is wet

(A+) If I believe that water is wet, then I or others in my speech community have had a history of interaction with water.


(B) I or others in my speech community have had a history of interaction with water

Each of the two premises may, on our assumptions, be known without elevation from the armchair, yet the conclusion (B), accessible from them merely by a modus ponens step, may not be so known if claim 3 is accepted.

It is a nice tight paradox.1 How to respond?

  • Should we allow the possibility of some limited degree of armchair knowledge of human history?2

  • Should we concede that ordinary psychological self-knowledge of states involving contents configuring externally constrained concepts cannot indeed after all have the traditional non-empirical immediacy accorded to it in the Cartesian tradition in philosophy of mind?3

  • (p.82) Should we reject content externalism, even for (apparently non-composite) concepts of natural kinds?4

  • Should we try to make out that the intuitive grounds for content externalism, even if convincing, do not support the pure a priori philosophical case that the sense of paradox presumes?5

  • Might there be mileage in the idea that it is never a priori knowable of any particular concept that it is externally constrained in the relevant fashion?6

Although I think there is much to be learned by pursuing them, my proposals here will not depend on whether enquiry in any of these five directions can deliver results sufficient to defuse the paradox.7 The suggestion will be, rather, that whatever the truth about each of those questions, achieving a solution of the paradox does not depend on successful pursuit of any of them. For an acceptance that propositions about socio-environmental history cannot as a class be warranted from the armchair is in tension with the armchair warrantability of (A) and (A+) only if it is assumed that, were (A) and (A+) both so warrantable, that epistemic feature would have also to apply to their consequence, (B). That's effectively a tacit assumption of transmission of warrant. It's the assumption that the diagnosis of transmission-failure repudiates.8

In what follows I propose to refine, and then, to a degree, qualify this diagnosis. Section I of the paper will be given to further reflections on warrant transmission in general. Section II will return to the McKinsey paradox. Although the announced direction of the paper might seem more suitable for a volume on epistemology in general than for an anthology on self-knowledge, we will in the end be in a position to (p.83) draw a conclusion of some interest, I think, for our understanding of the characteristic immediacy of intentional self-knowledge, and of the non-inferential a priori in general.



When does warrant for the (undischarged) premises of an argument, [A1…An], transmit to a validly inferred conclusion, C?9 There are a number of intuitive answers:

  1. (i) When the argument, together with the warrant for its premises,10 provides a potentially first-time reason to believe C–when a rational subject apprised of the warrant for [A1…An], could use the argument to learn C. Transmissive arguments are those that are at the service of the extension of knowledge by reasoning.

That answer is potentially distinct from this:

  1. (ii) When the warrant for the premises, together with the reasoning, provides for a new warrant to believe C.

The potential for the distinctness of answer (ii) from answer (i) resides in the possibility that an argument might somehow be at the service of providing a new reason for believing its conclusion even though not at the service of providing a first-time warrant for that conclusion. There might be some arguments that, in a given evidential context, are essentially tied to warrant enhancement, rather than warrant creation. Possible example: suppose I am trying to count a rather mobile, smallish flock of sheep in a pen. And suppose I am told that there are fewer than twenty and, on counting as carefully as I can, I get the result eighteen. I might reasonably take that result as corroborating what I have been told. But I might also reasonably think that, had I lacked any independent information about the number of the flock, I could not in the circumstances have trusted in the result of the count sufficiently to conclude that their number was fewer than twenty. If that is not an incoherent combination of epistemic attitudes, then the argument

  1. (a) I have counted the sheep as best I could and got the result, eighteen

  2. (b) There are eighteen sheep in the pen,

  3. (p.84) (c) There are fewer than twenty sheep in the pen,11

in combination with the empirical warrant for (a), is not at the service of the acquisition of a first-time warrant for (c), though it may rationally be deployed to enhance independent warrant for (c).

This reflection should not however encourage the general answer that warrant transmits in all cases

  1. (iii) When the argument, together with the warrant for its premises, is at the service of improving one's overall epistemic position with respect to its conclusion, C

—at least, not if “improving one's epistemic position” is identified with raising one's prior credence. It's obvious enough, on reflection, that a valid argument to a certain conclusion, together with the acquisition of a certain ground for its premises, may have the effect of enforcing a lowering of one's prior credence in that conclusion while providing a new—and perhaps still sufficient—reason for accepting it. This will be the effect in general when the availability of the relevant grounds for the premises of the argument itself tells against the obtaining of a significant range of possible cases under which the conclusion would in any case have held good, and thus adversely affects the latter's prior probability. Here is an illustration. Suppose—this is your evidence, (a)—that I deal you twenty-one cards face down from a standard bridge pack, advising you that exactly three aces—two blacks and a red—are among the dealt cards, and that exactly sixteen of the dealt cards are black. I then invite you to pick a card without turning it over. Presumably you can be reasonably fairly confident—in the light of the chances of being wrong: 7 to 1—that (b) you have not selected an ace and very confident—at odds against of 21 to 1—that (c) you have not selected a red ace. However if I then add to your initial evidence the further datum that the card you have selected is in fact red, the odds against (b)—that you have not selected an ace—and hence, what is equivalent to it in the new informational context, the odds against (c), shorten to 5 to 1; yet this is still enough, arguably, to ground something approaching an expectation that the card, when turned over, will not prove to be a red ace. Now, in another context, the grounds for this expectation–the information that you have selected a red card, that just five of the cards overall are red, and that they include just one ace–could have been acquired against a setting of antecedent total ignorance about the suits and denominations of the twenty-one dealt-out cards. It would then have been intuitively perfectly proper to describe you as having acquired a reason to think you had not selected a red ace where before you had none. That reason is exactly the same as that which you finally have in the example as developed above. It is just that acquiring it in the latter context destroys a previous, stronger reason for the same expectation. So warrant can transmit to (p.85) the conclusion of an argument even though its acquisition compromises a previously stronger epistemic position with respect to that conclusion.12


There are at least three other answers to the question, When does warrant transmit?, which are worth considering. Warrant transmits, it may be said,

  1. (iv) When the argument concerned may be rationally deployed to overcome a prior doubt about its conclusion, C.

This answer, in contrast with the previous three, takes the notion of warrant transmission into the territory of rational belief management. In effect, it introduces a new relativity additional to the relativity to grounds noted earlier: it views an argument as transmissive not simpliciter, but relative both to a certain type of grounds for its premises and the background doxastic state of the subject who entertains it. Clearly the strength of any prior doubt about C may be relevant: and here “strength” should be understood to cover both the psychological intensity of the doubt and the strength of the grounds for it. Pursuit of this direction, though, threatens to take us away from the intended subject matter. For no matter what argument we consider, it is always going to be a possibility that a given subject have in advance a sufficiently strong doubt about its conclusion—in either or both of the above senses of “strong”—to make it rational for them to look askance at the argument, and specifically at the grounds for its premises. Since this possibility pervades all arguments, it does not seem that the original intuitive distinction between transmissive and non-transmissive arguments should be compromised by it.

It would not help to finesse this wrinkle if it were suggested instead that warrant transmits

  1. (v) When the argument may be rationally deployed to overcome agnosticism about C

—not if agnosticism too is interpreted as a state of mind that allows of degrees of strength. It is so interpreted when it is viewed as itself a state of belief: the belief that the question, whether or not C, is something that allows of no settlement (at least by the kinds of considerations deployed in the argument in question). Strength, once again, may admit of both rational and psychological interpretation. In either case, though, it may dominate a subject's response to the details of the argument concerned, and hinder what is in fact a perfectly transmissive argument from overcoming a prior agnostic stance concerning C.

(p.86) We should conclude that if warrant transmission is to be characterized in anything like this kind of way, we need to appeal to a prior attitude to the conclusion of the argument concerned that, in the nature of the case, does not allow of rational, or psychological, entrenchment. That motivates

  1. (vi) When the argument may be used to rationally move past an open-minded stance towards C.

Open-mindedness, in its nature, does not allow of strength, or entrenchment: it involves a readiness to respond to whatever can be said for or against C. If an argument, together with a set of grounds for its premises, does indeed generate an all-things-considered sufficient case for C, a rational but open-minded subject will be responsive to it.

It will not be the business of this paper to pursue further these complexities in the notion of what it is for an argument to be transmissive. To be sure, insofar as the notion of warrant-transmissive reasoning is less than perfectly clear, so correlatively is the notion of transmission failure—and this threatens to blur the proposed diagnosis of the McKinsey paradox. I think, though, that we can side-step this threat. The frisson caused by the paradoxical argument resides in its apparent generation of an a priori warrant, in the loose and popular sense of “a priori”, for a proposition about socio-environmental history: we don't want the McKinsey argument to produce a (first-time) warrant to believe the relevant C based purely on the armchair warrants for its premises, and its simple reasoning; nor do we want such armchair considerations to be capable of moving a rational subject who is antecedently open-minded about the socio-historical proposition. It would accordingly suffice for our purposes, I submit, if we can make the claim stick that, at worst, if no other fault is to be found with it, the reasoning of the McKinsey paradox exhibits transmission failure in the senses gestured at by answers (i) and (vi): even if its premises are indeed armchair warranted, and its reasoning valid, it produces no way of learning of the truth of its conclusion and should not move someone antecedently open-minded about that conclusion to conviction that it is true.


There is a second very important issue that must be reviewed before we can approach the question of transmission failure in the McKinsey argument. Here I must ask the reader to indulge me while I engage in a little retrospective self-exegesis. In previous work13 I have proposed two broad ‘templates’ for the characterization and diagnosis of warrant transmission failure. The first of these is

(p.87) The information-dependence template

This is conceived to cover cases where the warrant for the premises is itself grounded in an ampliative inference from a defeasible evidential base:

A body of evidence, e, is an information-dependent warrant for a particular proposition, P, if regarding e as warranting P rationally requires certain kinds of collateral information, I. Some examples of such e, P and I have the feature that elements of the relevant I are themselves entailed by P, (together perhaps with other warranted premises). In that case, any warrant supplied by e for P will not be transmissible to those elements of I.14

Note the play in this formulation with “regarding e as warranting P”.15 The template talks about what it takes for a thinker to be in position to regard a certain body of evidence as a (sufficient) ground for a certain proposition. Unless you want to say—which many will not—that to be in possession of sufficient grounds for accepting a certain proposition is the same thing as being in a position to regard those grounds as sufficient for accepting that proposition, it is a fair observation that the proposed template seems to involve a potential change of subject matter: a change from what it takes for an argument to bestow warrant to what it takes for an argument to bestow being in position to regard as warranted, to put it clumsily. Less clumsily, what the template is seemingly apt to diagnose, or characterize, is a set-up where the relations between a particular e, P, and I are such that a rational thinker cannot lay claim to e as contributing to a warrant for certain of the entailed consequences of P that fall within I, even though it is rational to regard e as providing warrant for P in a context where those elements of I are independently warranted. What, according to the template, is not transmitted across the entailment from P to the relevant elements of I is thus the possibility of rationally laying claim to warrant on the basis of e: a thinker, with the appropriate background information, may rationally claim warrant for P on the basis of e, but she cannot rationally claim warrant for the relevant elements of I on the basis of e (and the deduction from P) above and beyond whatever her background warrant may be for those elements of I. In short, the information-dependence template is a diagnostic for, in one natural sense of the phrase, failure of transmission of second-order warrant.16

In the light of that reflection, now consider

(p.88) The disjunctive template

This was conceived as a heuristic for the detection of warrant transmission failure in cases where the premises of an argument are not based on a defeasible evidential warrant but where their truth is, or so we conceive, apprehended directly by the operation of some suitable cognitive faculty—perception, for instance, or memory if conceived as a form of non-inferential awareness of one's own past, or non-inferential intellection (as would be involved in basic knowledge of the (strictly) a priori, if there is such a thing), or ordinary non-inferential knowledge of one's own psychological states. We can see that such a heuristic is called for by considering Dretske's familiar example—the inference from

That wall is red


That wall is not a white wall so illuminated by concealed red lighting as to appear red.

This presents a very plausible example of warrant transmission failure when its premise is thought of as warranted just by looking. But it does not seem necessary, in order for the example to give that impression, that one thinks of the visual warrant concerned as based on an ampliative inference from defeasible visual evidence (say, that the wall looks red). The impression of ‘overreaching’ still remains even when one thinks of vision as, in the best case, making the colours of surfaces directly available to us.

Here is how I summarized the disjunctive template proposed in Wright (2000):

Suppose I take myself to have a non-inferential warrant—perhaps perceptual, or introspective, or intuitive, or mnemonic—for some proposition A. And let the question be whether this warrant transmits to a certain consequence, B, of A. However suppose A stands opposed to some proposition C which would be true in circumstances subjectively indistinguishable from those in which I actually find myself, and which does not entail B, but would be true if B were false. The set-up is thus

  1. (i) that Aentails B;

  2. (ii) that my warrant for A consists in my being in a state which is subjectively indistinguishable from a state in which the relevant C would be true;

  3. (iii) that C is incompatible with A; and

  4. (iv) that [not-B entails C].17

    The key question is what, in the circumstances, can justify me in accepting A? Why not just reserve judgement and stay with the more tentative disjunction, “A or C”?—for it is all the same (p.89) which disjunct is true as far as what is subjectively apparent to me is concerned. The answer has to be, it would seem: because I am somehow additionally entitled to discount the other disjunct, C. It may be that I have collateral information telling against C. Or it may be that…I am not required to bother about C. But either way, in order for me to be entitled to discount C, and so move past the disjunction to A, I have to be entitled to discount the negation of B, and therefore entitled to accept B; for by hypothesis, if not-B were true, so would C be. So it would seem that I must have an appreciable entitlement to affirm B already, independent of the recognition of its entailment by A, if I am to claim to be warranted in accepting A in the first place. So the inference from A to B is not at the service of a rational first conviction that B.18

Once again, we find conditions proposed as necessary if “I am to claim to be warranted in accepting A in the first place”. The rationale for this template too, so explained, is heavily invested in intuitive conditions on what it takes to be in position to lay claim to a warrant—to defend the claim that one is indeed warranted. It has to be acknowledged, therefore, that even if the two templates—perhaps with further refinement—seem plausible, it is going to be an additional step to argue that they can serve as diagnostics of warrant transmission failure at first-order, so to speak, once a potential distinction is acknowledged between what it takes to have warrant for a certain belief and what it takes to be in position to lay claim to it.

The question is of great importance. There is, in particular, a substantial issue whether the debate between so-called liberals and conservatives over perceptual justification may not turn on an ambiguity of focus between issues concerning first-order warrant and second-order warrant respectively, with the possible result that Moore's proof, for instance, might indeed be serviceable for the provision of warrant to believe in an external world though useless for underwriting claim to such warrant, so useless in the context of epistemological debate, because non-transmissive at second-order. I do not myself anticipate such a finding, but I cannot pursue the issues further here. The crucial point for present proposes is that the McKinsey paradox is in any case, in the first instance, a paradox at second-order. What originally seems intuitively absurd is that someone might rationally put forward the combined loosely a priori grounds for its two premises as co-operating to support the socio-historical conclusion, or rationally explain her acceptance of the conclusion in terms of her reflective warrants for those two propositions. Once the possibility of a contrast is signalled between what it takes to have warrant and what it takes to be in position to claim it, the notion that one might indeed have warrant for a socio-historical proposition on the basis of an immediate piece of self-knowledge and some pure philosophy—although still in no position to claim warrant on such grounds—is less paradoxical than merely philosophically puzzling.

For this reason, I think we are justified in framing the discussion to follow in terms of (an improved) version of the disjunctive template. The most we can thereby directly accomplish will indeed be a diagnosis of second-order warrant transmission failure in (p.90) the McKinsey reasoning. But I think that will be enough to address the intuitive paradox.


The disjunctive template in the above form arguably predicts second-order transmission failure in each of the following three cases:

Let us run through the Zebra case. C is incompatible with A and is the negation of B. So conditions (i) and (ii) are met. I look at the animals in the zoo enclosure and, so it seems to me, perceive that they are zebras. But the subjective state involved in such a



Red Wall


Those animals are zebras

I have a hand

That wall is red


Those animals are not mules cleverly disguised to look just like zebras

There is an external material world (since a hand is a material object, existing in space)

That wall is not a white wall so illuminated by concealed red lighting as to appear red


Those animals are mules cleverly disguised to look just like zebras

There is no external material world

That wall is a white wall so illuminated by concealed red lighting as to appear red

perception is shared with a perception of mules, artfully disguised to look exactly like zebras (condition (i)). So if I am rationally to claim to be perceiving zebras, and to possess warrant for A on that basis, I had better be in a position to discount the mule hypothesis. (Otherwise the suggestion that it only seems to me that I am perceiving zebras is open.) But that is the same as to say that I had better be in a position to affirm that the animals in question are not cleverly disguised mules—precisely B. So that is something that I need to be in position to claim on independent grounds prior to claiming perceptual warrant for A, and on which my claim to warrant for A therefore rests. It is accordingly not something for which I can acquire a claim to warrant by the argument itself.

The template as proposed, however, manifests what I regard as a serious wrinkle.19 Take any case where the warrant for A consists in defeasible evidence of some kind, and let C be the negation of A. Then condition (ii) (I am referring to the quotation from Wright 2000 above) is met: there is a state, namely when I would have exactly the same evidence but the negation of A would hold true (a possibility because, remember, that evidence is defeasible), which is subjectively indistinguishable from my actual state, but in which C would be true. Here C is, trivially, incompatible with A, so condition (iii) is also met. And since C is the negation of A, any consequence, B, of A is such that its negation will entail C (condition (iv)). The template will therefore outlaw transmission (p.91) of any defeasible claim to warrant for A to any conclusion B entailed by A. In particular, it will, for example, outlaw transmission of a testimony-based claim to warrant from

  1. A Those animals are zebras


  • B Those animals are not tigers,

(where C is that [those animals are not zebras].)

Why is this ‘explosiveness’ a problem? After all, it may justly be remarked, the observation is merely that the template (comprehensively) overgenerates when applied to cases for which it was not designed—for it was designed specifically to address the possibility of second-order warrant transmission failures in cases where the premises are not warranted by an accumulation of defeasible evidence but, to the contrary, in some direct cognitive manner. Such a reply might be satisfactory if we had to hand some reason for thinking that the bases for failures of transmission in the two kinds of case are actually quite different and that no single template can be adequate to cover them both. But I know of no such reason and doubt that there is any. So the fact that the disjunctive template as formulated misfires so egregiously when applied to the ampliative-inferential case is disquieting. Rather, it seems more likely that an ultimately satisfactory account of the varieties of transmission failure will disclose commonalities between the cases where the premises of the argument concerned are warranted inferentially, or non-inferentially respectively—that the root epistemic problem in both kinds of case should be the same, and detected by a common heuristic.

How, in pursuit of that direction, can we improve things? Jessica Brown20 has suggested replacing condition (ii):

My having the warrant concerned involves my being in a state which is subjectively indistinguishable from some state in which C would be true,

with (ii)*

If C were true, I would be in a state subjectively indistinguishable from my actual state.

This adjustment does indeed seem potentially to differentiate the two zebra examples in the right kind of way. In the case where I have an apparent perceptual warrant that the animals concerned are zebras, it is true that were they cleverly disguised mules I would be in a state subjectively indistinguishable from my actual state. So Brown's new condition is met and the diagnosis of transmission failure can stand. But in the case where I look unknowingly at the stripy horse-like creatures and am told correctly that (p.92) they are zebras, then were they not to be zebras, my informant would presumably not be saying that they were and my visual state would presumably be different—so my subjective state overall would not be indistinguishable from my actual state. Brown's condition (ii)* is thus not met in the testimonial case, and her modified template therefore presents no obstacle to the intuitively plausible thought that the testimony-based claim to warrant should transmit.

However things are still not in good order. Consider the application of an inductive warrant to an unobserved case, for instance the argument from

A All men are mortal


B The US president in 2090 is mortal,

where C is taken to be: some unobserved man is not mortal. Then Brown's revision is going to diagnose that an inductively grounded claim to warrant for A does not transmit to B; for original conditions (iii) and (iv) are met—C is incompatible with A, and the negation of B entails C—but it is also true, presumably, that Browns condition (ii)* is satisfied: if some unobserved man is indeed immortal, one presumably would still have exactly the same inductive evidence as we actually have for A, and one's subjective state would be indistinguishable from what it actually is.

How can we do better? My suggestion is that the fault with the template as originally proposed lies not with condition (ii) but with condition (iii). When someone lays claim to warrant for some proposition A on the basis of the execution of some cognitive procedure, it is indeed prima facie reasonable to require—pace contextualists and other kinds of friend of “relevant alternatives”—that they be in a position to discount any C inconsistent with A and therefore—since both A and C will generally concern matters independent of one's subjective state—in position to discount any such C which is inconsistent with A and yet compatible with the type of subjective state that one is in. But it is too much to require that one have in general independent warrant to do this. Sometimes the warrant concerned must allowably be the very warrant one putatively claims for A. That is what the inductive example brings out. The question is therefore: which are the cases where it is reasonable to require such independent warrant?

It is germane here to enlist the notion of a presupposition of a cognitive project in a sense I have invoked in earlier work.21 A presupposition of a cognitive project is any condition P such that to doubt P (in advance of executing the project) would rationally commit one to doubting the significance, or competence of the project, irrespective of its outcome.22 Examples of such presuppositions for a wide class of cognitive projects will include: normal and proper functioning of relevant cognitive faculties, the (p.93) reliability of instruments utilized, the amenability of the circumstances to the proposed method of investigation, the soundness of relevant principles of inference utilized in developing and collating one's results, the good standing of relevant concepts used in any aspect of the enquiry, and so on. Such presuppositions are typically things one takes for granted in cognitive projects ranging from the mundane, like glancing at a clock to see if it is time to leave, to the more methodologically self-conscious, like a carefully controlled scientific experiment. My suggestion is that in so far as the original disjunctive template may seem, to a sympathetic reader, to be on to something, what it is on to is this: that you cannot rationally transmit a claim to warrant for certain premises across an entailment to any otherwise uninvestigated presupposition of the very cognitive project which purportedly generates that warrant. There is nothing in general amiss with the idea that merely looking at animals in a zoo enclosure, or looking at a wall, can put one in a position to claim warrant that the animals in question are zebras, or that the wall is red. But one cannot, just by reception of the relevant visual appearances, get in position rationally to claim that conditions are suitable for the identification of animal species, or surface colours, just on the basis of vision. That conditions are so suitable is a presupposition of the two projects concerned. Neither of those general presuppositions, nor the specific forms of them detailed in the relevant B-propositions, are available for merely visual confirmation.


Here, then, is a new template for failure of second-order warrant transmission—transmission of claim to warrant:

The revised template

Where A entails B, a rational claim to warrant for A is not transmissible to B if there is some proposition C such that:

  1. (i) The process/state of accomplishing the relevant putative warrant for A is subjectively compatible with C's holding: things could be with one in all respects exactly as they subjectively are yet C be true

  2. (ii) C is incompatible (not necessarily with A but) with some presupposition of the cognitive project of obtaining a warrant for A in the relevant fashion, and

  3. (iii) Not-B entails C

Notice that, in the case where A is supported by defeasible evidence e, whose supportive relevance to A depends on rational acceptance of some collateral claim I, that I holds good will indeed to be a presupposition in the germane sense: doubt about I will commit one to doubting the material relevance of e and hence the competence of the project of taking a view about A on that basis. So the revised template passes at last (p.94) initial muster as the generalization of the two original templates that I suggested is called for: it is applicable, plausibly, both to arguments whose premises are conceived as warranted non-inferentially by direct cognitive reception, and to arguments with inferentially (defeasibly) warranted premises.

What about the testimonial zebras case and the inductive warrant case? Does the revised template match the performance of the Brown revision in the first case and improve on it in the second? The question is whether the propositions that were selected for C in those two examples and which respectively caused over-generation trouble for the original disjunctive template and the Brown modification are incompatible with anything plausibly taken as a presupposition of either of the two cognitive projects concerned. If not, then the revised template poses no evident obstacle to second-order warrant transmission in those two cases and things are just as they should be. The two respective selections for C were

Those animals are not zebras,


Some unobserved man is not mortal.

The revised template will accordingly diagnose transmission trouble just if either of these is incompatible with some presupposition of the respectively relevant projects: a testimony-based investigation of the species of animal in the zoo enclosure and an inductive investigation of human mortality. Since the presuppositions of any cognitive project comprise a somewhat open-ended and unsurveyable range, it is not to be expected that proof will be forthcoming that this is not so. But on the face of it, neither of these propositions seems to entail the abrogation of anything required by the competence or significance of the two projects concerned. That the animals are, or are not, of any particular species has no bearing on the reliability of one's informant or one's ears, for example; that the generalization in question is in fact false has no bearing on the competence of any particular inductive investigation of it.23

I leave it to the interested reader to take a view on the upshot of applying the revised template for second-order warrant transmission failure to Moore.



Are we, in the light of this revised template, now in position to shed light on the McKinsey paradox:

(p.95) McKinsey

  • A I believe that water is wet

  • A+ If I believe that water is wet, then my speech community has historically had contact with water

  • B My speech community has historically had contact with water?

The first premise, or so we are supposing, expresses a routine piece of non-inferential psychological self-knowledge. And the second (so we are supposing for the purposes of the present discussion) may be grounded a priori upon reflection on the character of the concept water, together with considerations that motivate some germane form of semantic externalism. A solution grounded in the idea of warrant transmission failure is going to propose that, although both premises may indeed be known a priori—in the loose and popular sense—and their entailment of the conclusion is trivially valid, the reasoning is not at the service of supporting a claim to (a priori) warrant for the conclusion. The point is to be compared to the thought that although the reasoning of Zebras or Red Wall is trivially valid, and although their respective premises may be warranted by simple observation, the reasoning is not at the service of supporting a claim to (observational) warrant for their respective conclusions. Can we apply the revised template to corroborate this diagnosis?

If we are to do so, we need to identify a C which is entailed by the negation of B, which could be true in circumstances subjectively indistinguishable from those involved in recognition of the truth of the two premises, and which clashes with a presupposition, in the sense adumbrated above, of one or other of those projects of recognition. Our focus, of course, will be on premise A. So we are looking for some relevant type of presupposition that applies to processes of ordinary non-inferential self-knowledge. It might seem that the very immediacy of this particular knowledge phenomenon renders it virtually presuppositionless. But this is not so. In coming to any putatively self-knowledgeable judgement, “I [attitude] that Q”, it is a presupposition that all relevant concepts utilized, including those configured in the putative content Q, are in good standing: that a genuine content is articulated thereby, meet for the attitude in question. But what content externalism of the kind with which we are concerned opens up is exactly the possibility that, failing satisfaction of the appropriate external conditions, illusions of (truth-evaluable) content are possible—it could be, for example, with a thinker in all subjective respects as if they were thinking the thought (our thought) that water is wet and yet owing to peculiarities of their environmental history, there may be no determinate concept expressed by tokenings of “water” in their thought.24

Let C be the proposition that one or another of the apparent concepts configured in the apparent belief that I express by premise A is deficient, and unfit to contribute towards truth-evaluable thought. We need not here engage the question whether to (p.96) countenance some kind of ersatz—less than truth-evaluable—content in such a case. The important point is that, even if we do, there will be no case for the second premise, A+, unless we take it that it is concerned with thought in the stronger, truth-evaluable sense. There will not be any externalist precondition on mere ersatz thought. The result is that we have a prima facie case that the three conditions required by the revised template are each met:

  1. (i) Proposition C could hold in circumstances subjectively just like those I am in when, suitably attentively, I seem to confirm premise A;

  2. (ii) Proposition C is incompatible with a presupposition of the relevant cognitive project;25

  3. (iii) Not-B entails C—if the external precondition of historic contact with water were not met, the concept of water would be defective, and unfit to contribute to the tokening of determinate, truth-evaluable thoughts.

Here again is the overarching intuitive point. In taking it that I have recognized, in the normal immediate-seeming folk psychological way, that I have a certain specific belief, I take it for granted that the apparent concepts configured in that belief are in good standing, just as in taking it that I have confirmed, in a normal casual-observational way, that there are zebras in a zoo pen, I take it for granted that conditions are suitable for the identification of animal species by casual observation, and hence that no disguise, subtle enough to frustrate that possibility, is operative. My claim to warrant for the premises in each of the arguments thus holds out hostage to a condition which, we may suppose, I have subjected to no independent check. The claim to warrant for the premises rests on the reasonableness of these presuppositions. If someone is open-minded about them, he should be open-minded about whether there he has any warrant for the premises. So neither of the arguments is available to rationally induce an open-minded subject to lay claim to warrant for its conclusion.

The intuitive analogy seems good. And the prima facie applicability of the revised template to both arguments arguably both confirms the analogy, and supports the diagnostic credentials of the revised template.


It is important to remark, however, that even if all this is right, we have done no more than take a step—to be sure, an important step—towards dissolving the McKinsey paradox. We cannot yet claim to have completed the job in all desirable respects. What will have been accomplished, if all is well, will be an explanation of the failure of the types of warrant envisaged for the premises in the McKinsey argument to (p.97) contribute, in conjunction with the derivation, to the construction of a claimable warrant for the conclusion. The warrants concerned do not extend to their presuppositions. Dissolving the paradox needs more than that, however. If no more is said, it will be open to us to suppose that the premises concerned are actually unwarranted unless or until the externalist presupposition is independently warranted, and hence—since accomplishing that will require broadly empirical evidence—that the traditional conception of self-knowledge as fully available in the armchair is, strictly, undermined if the second premise is accepted. To stave off that response, it needs to be shown that nothing has happened to disturb the propriety of the ordinary armchair types of warrant concerned. It needs to be shown that, even after one recognizes external constraints on content, the ordinary self-attribution of attitudinal states can responsibly proceed as before, without additional empirical obligation. Pari passu, it needs to be shown that, even after the possibility of refined disguise is noted, ordinary casual observational identification of animal species, and many other things, can likewise proceed as before, without epistemic irresponsibility.

It might reasonably be observed that the presuppositions in question are things about which we often have considerable accumulated empirical reason for confidence. Experience teaches that people do not often disguise zoo animals, for example, and it would in any case be no easy matter to fool a population as exposed as often as we are to images of the real thing, by film and television of wild life. Natural kind concepts that prove defective in reference are, likewise, a rarity: and the Dry Earth type of case, involving massive sustained hallucination, is surely the merest philosophical fantasy. It is beyond the scope of the present paper to explore the issues that beckon here. Let me say merely that I believe the notion that we normally have empirical reason for confidence in each of the epistemic presuppositions of projects whose results we routinely accept is misconceived—indeed that such empirical reason, if demanded for everything we would admit to be a presupposition in the relevant sense, cannot in general be forthcoming. The view I would seek to defend is that ordinary psychological self-knowledge remains available in the armchair even if we are persuaded of content-externalism; but the reason it does so is not because there is prior empirical reason for confidence in its presuppositions, but because—save in an exceptional context—no empirical reason, or indeed evidential reason of any kind is required. On this view, we are entitled, without specific evidence, to take for granted—to trust in—a wide range of presuppositions of cognitive project; and we are so entitled in particular in every case where there is no evidential reason for misgivings about a presupposition and where any attempt at its corroboration would throw up further presuppositions of no safer an antecedent evidential standing (and so on indefinitely).26

To be sure, whether this proposal, if granted in general terms, can be made to work for presuppositions of the integrity of externally constrained concepts in particular is not immediately evident. It will depend, specifically, on whether an investigation of the (p.98) satisfaction of those presuppositions must inevitably have the regressive character just gestured at. One way in which that might be so would be if, so pervasive are concepts of this character, each successive investigation would unavoidably have recourse to further externally constrained concepts. Here, though, I wish merely to table the following, as it seems to me, very important point. If it is ever the case that (a claim to) warrant for a particular belief may be accomplished in a manner which rests on certain presuppositions without the need for independent corroboration of those presuppositions—if they may indeed rationally be taken for granted—then the possibility is opened that a priori warrants, whether in the loose and popular sense or something stricter, may very well rest on presuppositions whose investigation, were we to choose to investigate them, would have to be empirical and a posteriori. Armchair psychological self-knowledge might consequently depend, without any compromise of the propriety of its provenance in the armchair, on presuppositions whose investigation, if possible at all, would involve rising from the armchair. More generally, defensible claims to a priori knowledge may rest on uninvestigated empirical presuppositions.


Unfortunately, there is a problem even with the partial solution. The case that the McKinsey argument satisfies the three conditions of the revised template has not been satisfactorily made. The problem has to do with condition (iii). It is not straightforwardly true that a failure on the part of anyone in my speech community to have any form of historical contact with water would entail that there would be a deficiency of content in tokens of water is wet in my thought. The entailment fails even if we take it that the only way to have the concept water—as opposed, say, to the concept H 2 O—is indeed by interaction with the relevant stuff. The recalcitrant possibility that blocks the entailment is where there has been relevant contact with samples of some other watery stuff (some other stuff that satisfies the superficial stereotype of water), with the result that tokenings of water is wet in my thought would have been fully contentful but would have concerned that other stuff instead.

A reader who shares my initial reaction will be inclined to suspect that this is merely a difficulty of formulation—that it will likely be finessed by a more judicious selection for the bad possibility, C. It might be suggested, for example, that the presupposition which C is to abrogate should be not that the concepts involved in my putative thought-token, whatever they may be, are in good standing, but specifically that my actual concept water is in good standing. So C should be that my actual concept of water is not in good standing. But this goes nowhere. For if the concept water gets to feature in the formulation of C, then circumstances under which it is divested of content will be circumstances under which C too is divested of content, and it will not be correct to say that C will be true if not B holds. Condition (iii) will therefore fail in any case.

Let's try a different tack. What is true, for the kind of externalism we are concerned with, is that it is a necessary condition for the good standing of the concept water—always presuming that water is indeed a suitably externally constrained concept—that (p.99) I or my speech community have had historic contact with some watery substance, that is, some substance picked out by the reference-fixing stereotype with which water is associated. Had this condition not been met, the concept water—conceived as an archetypal non-composite natural kind concept—could not have got off the ground. So it appears that a diagnosis of transmission failure informed by the revised template could at any rate stand straightforwardly for this variant of the McKinsey argument:


  • A I believe that water is wet

  • A+* If I believe that water is wet, then my speech community has historically had contact with some watery substance

  • B My speech community has historically had contact with some watery substance

In the light of that consideration, consider

Four-step McKinsey

  • A I believe that water is wet

  • A+ If I believe that water is wet, then my speech community has historically had contact with water

  • B My speech community has historically had contact with water

  • B′ My speech community has historically had contact with some watery substance

Here, the first three lines coincide with original McKinsey. But that argument can now be considered in two different cases. First, suppose we explicitly presuppose B′: we take it for granted that our speech community has had and continues to have contact with some unique watery substance. Then in that setting, the inference from A and A+ to B—the McKinsey inference—seems entirely anodyne and unparadoxical. Water, after all, is precisely the concept we have introduced to be properly characteristic of that watery substance, whatever it is. On the other hand—the second case—suppose we do not make an explicit presupposition of B′ but view it rather as something to be justified in turn by inference from B. Then now the McKinsey argument—taken as the first three lines of Four-step McKinsey—presents as a paradox again.

So we may suggest a slightly more refined diagnosis. The McKinsey argument is not a failure of (second-order) warrant transmission tout court, even when we restrict attention to a priori warrants for its premises A and A+. Rather the matter turns on the interpretation of the epistemic significance of its conclusion, B. If B is taken merely to express the identification of the watery substance that, it is presupposed, my speech community has encountered with the reference of the concept that features in my belief, then there seems no good objection to thinking of it as justified on the basis of A and A+. But if B is viewed, instead, as a lemma for a justifying inference to B′, then there will once again be a failure of transmission of warrant in the inference from A and A+ to B, and the original diagnosis of the McKinsey argument can stand. In brief: McKinsey involves warrant transmission failure when its conclusion is considered as a potential justifier of B′. But when by contrast B′ is presupposed, there seems no (p.100) objection to supposing that the warrant for McKinsey A and A+ does indeed transmit to McKinsey B.

This diagnosis is quite attractive. But it is hostage to the plausibility of the claim that there is a good inference from B to B′. Plainly there is not if “good” means: valid, and validity requires necessary truth preservation, since—at least according to the orthodox Kripkean package of claims with which the kind of externalism we are considering is associated—water might not have been a watery substance at all. But maybe valid inference is not required. Is the inference from B to B′ at least a priori truth preserving—is the conditional, if B, then B′, a priori true? If so, an inference concluding in B will license a priori the step to B′, and the four step argument will indeed pose the paradox.

So is it a priori that water is a watery substance—that water meets the reference-fixing stereotype of water? More exactly: is it a priori that if my speech community has had contact with water, that it has had contact with some watery substance? Again, one orthodox thought in the vicinity says it is: the thought, namely, that water is simply the given conceptual name, as it were, of the actual watery stuff of our acquaintance, with the result that the conditional needed for the transition between B and B′ is a priori as an artefact of reference-fixing.

However, although this is not the place to get in to the matter, I do not think it straightforward that there are any claims that hold good a priori just in virtue of what is involved in fixing the reference of certain of the concepts they contain.27 Let me therefore canvass an alternative refinement,28

Four-step McKinsey*

  • A I believe that water is wet

  • (p.101) A++ If I believe that water is wet, then my speech community has had appropriate contact with something uniquely fit to serve as the reference of my concept of water

  • B″ My speech community has had an appropriate contact with something uniquely fit to serve as the reference of my concept of water.

However, it is presumably a priori that water is uniquely fitted to serve as the reference of my concept of water if anything is. So

  1. B My speech community has had contact with water.

Here, as the reader may confirm, the revised template seems to fit relatively cleanly (take C, once again, as: one or another of the apparent concepts configured in the apparent belief that I express by premise A is deficient, and unfit to contribute towards truth-evaluable thought). In particular—condition (iii)—my speech community's having had no appropriate contact with anything suitable to serve as the reference of my concept of water would entail that there is no such determinate concept that I possess, and hence that the conditions for the good standing of the content of the apparent belief reported in A are not met.


Let me clarify what I am trying to accomplish with these variations on McKinsey. An unsympathetic reader may rejoin that it is one thing—assuming everything else is in order—to disclose transmission failures in arguments in some respect akin to McKinsey's, and another to correctly so diagnose the McKinsey argument itself. I agree. This is the respect in which, as I said I would at the start, I am qualifying my original diagnosis. Here in summary is my qualified suggestion. McKinsey's argument is open to two distinct interpretations, not of course in the sense that its premises or conclusion admit of any relevant semantic ambiguity, but in the sense that it may be taken in two quite different epistemic settings. In one we simply explicitly presuppose—with or without entitlement; that's a further issue—that conditions for the efficacy of ordinary processes of self-knowledge, including the good standing of relevant concepts, and further including such externalist conditions as may bear on that question, are all met. With those presuppositions in place, armchair warrant may be claimed for both premises of the McKinsey argument (granting that is available for the second) and transmits to the conclusion, which adds to our presuppositions only that is it water that provides the proper conceptualization of the stuff to which we have been appropriately historically related. Under the second interpretation, by contrast, we do not make the presupposition but proceed as before to what we conceive as warranted acceptance of the two premises. There are then two cases. If we are not entitled to the presupposition of the externalist condition, then immediate reflective knowledge of the psychological premise is not strictly possible and further empirical work is needed to in order for the claim of warrant for it to be sustained; so there is no transmission failure. But if there is such an entitlement, (p.102) then the claim to ordinary armchair warrant for the psychological premise remains in good standing but does not transmit across its entailment, via the externalist premise, to the socio-historical conclusion. A precise diagnosis of why transmission fails in that case, informed by the revised template, is hostage, as we have seen, to what exactly we take the troublesome component of that conclusion to be, once we skim off the harmlessly inferrable claim about the proper conceptualization of the stuff concerned. Four-step McKinsey and Four-step McKinsey* in effect experiment with different versions of this with a view to smoothening the application of the template. The attempt may well repay further effort. But I shall not try to take things further here.


One final issue. At the meeting at which I presented an earlier discussion of these issues,29 Michael McKinsey himself observed that issues concerning transmission of warrant will be beside the point if some relevantly local principle of closure of warrant were to hold. More specifically, if (loose and popular) a priori warrant is at least closed—if it is true that whenever there is a priori warrant for the premises of a valid argument, there is a priori warrant for its conclusion, even if it is not to be achieved by inference from the premises—then there is still a paradox. For it is surely very implausible that the socio-historical proposition is a priori warrantable at all, never mind whether warrant is transmitted from the relevant premises.

My response to this, as the attentive reader will anticipate, is that the relevant closure principle is false. It is, as is familiar,30 no terribly straightforward matter to formulate a general principle of closure of warrant in a form that plausibly passes muster. But let it be that some such principle is good. Still, I regard it as very doubtful whether any plausible form of the principle holds once we additionally require closure with respect to broad kinds of warrant. Maybe undifferentiated warrant is closed, when closure is suitably judiciously formulated. But, for example, visual warrant is not. This in effect is what Dretske noticed. I have a visual warrant for the claim that those animals are zebras; that entails that they are not mules so artfully disguised as to be visually indistinguishable from zebras; but I do not have a visual (or visual-cum-inferential) warrant to suppose that they are not such artfully disguised mules. Likewise I currently have a memory-based warrant for thinking that I cooked organic porridge for breakfast this morning; that claim entails I have not been asleep for the last two months awakening only five minutes ago with a framework of coherent apparent memories about what I had been doing in that period, including one of porridge preparation this morning; but I don't have a memory-based warrant to think that I have not been asleep…etc.

(p.103) There is no general reason to expect matters to be any different with a priori warrant—with reflective, armchair warrant—however specifically it is conceived. Any mode of warrant acquisition may, on any occasion where it successfully delivers warrant, do so courtesy of presuppositions whose obtaining, or not, it is itself impotent to certify. That is consistent with some of the propositions that, in a suitable context, it is capable of certifying—courtesy of those presuppositions—themselves being deductively sufficient, perhaps in conjunction with other certified premises, for the obtaining of some such presuppositions. That, in a nutshell, is the situation with which, in my view, we are dealing in all the examples with which we have been concerned.


Bibliography references:

Blome–Tillmann, Michael (2006), ‘A Closer Look at Closure Scepticism, Proceedings of the Aristotelian Society 106: 383–92.

Boghossian, Paul (2008), Content and Justification: Philosophical Papers (Oxford: Clarendon Press).

—— (1997), ‘What the Externalist Can Know a Priori?’, Proceedings of the Aristotelian Society 97: 161–75, now in Boghossian (2008: 177–88).

Brown, Jessica (2004), ‘Wright on Transmission Failure’, Analysis 64/1: 57–67.

Chandler, Jake (2009), ‘The Transmission of Support: A Bayesian Re-Analysis’, Synthese.

Davies, Martin (2003b), ‘Externalism, Self-Knowledge, and Transmission of Warrant’, in M. Frápolli and E. Romero (eds.), Meaning, Basic Self-Knowledge, and Mind: Essays on Tyler Burge (Stanford: CSLI Publications), 105–30.

——(2003a), ‘The Problem of Armchair Knowledge’, in Nuccetelli (2003: 23–55).

——(2000), ‘Externalism and Armchair Knowledge’, in P. Boghossian and C. Peacocke (eds.), New Essays on the A Priori (Oxford: OUP), 384–414.

—— (1998), ‘Externalism, Architecturalism, and Epistemic Warrant’, in C. Wright, M. Smith, and C. Macdonald (eds.), Knowing Our Own Minds: Essays in Self-Knowledge (Oxford: OUP), 321–61.

Hale, Bob (2000), ‘Transmission and Closure’, Philosophical Issues 10: 173–90.

Lasonen-Aarnio, Maria (2008), ‘Single-Premise Deduction and Risk’, Philosophical Studies 141/2: 157–73.

Loar, B. (1988), ‘Social Content and Psychological Content’, in R. H. Grimm and D. D. Merrill (eds.), Knowing Our Own Minds: Essays in Self-Knowledge (Oxford: Oxford University Press), 271–84.

McBride, Mark (2009), ‘Is Knowledge Closed Under Known Entailment? The Strange Case of Hawthorne's “Heavyweight Conjunct” (and Other Strange Cases)’, Theoria 75/2: 117–28.

McKinsey, Michael (2003), ‘Transmission of Warrant and Closure of Apriority’, in Nuccetelli (2003: 97–115).

—— (1991), ‘Anti-Individualism and Privileged Access’, Analysis 51: 9–16.

McLaughlin, Brian P. (2003), ‘McKinsey's Challenge, Warrant Transmission, and Skepticism’, in Nuccetelli (2003: 79–96).

McLaughlin, Brian, and Tye, Michael (1998), ‘Externalism, Twin Earth, and Self-Knowledge’, in C. Wright, M. Smith, and C. Macdonald (eds.), Knowing Our Own Minds: Essays in Self-Knowledge (Oxford: Oxford University Press), 285–320.

(p.104) Moretti, Luca (2009), ‘Wright, Okasha and Chandler on Transmission Failure’, Philpapers: Online research in philosophy, at 〈〉.

Nuccetelli, Susana (ed.) (2003), New Essays on Semantic Externalism and Self-Knowledge. (Cambridege, Mass.: MIT Press).

Okasha, Samir (2004), ‘Wright on the Transmission of Support: A Bayesian Analysis’, Analysis 64: 139–46.

Patterson, Sarah (1990), ‘The Explanatory Role of Belief Ascriptions’, Philosophical Studies 59/3: 313–32.

Pryor, James (forthcoming), ‘When Warrant Transmits’, in A. Coliva (ed.), Wittgenstein, Epistemology and Mind: Themes from the Philosophy of Crispin Wright (Oxford: OUP).

—— (2007), ‘What's Wrong with McKinsey-Style Reasoning?’, in S. Goldberg (ed.), Internalism and Externalism in Semantics and Epistemology (Oxford: OUP).

Sawyer, Sarah, (1998), ‘Privileged Access to the World’, Australasian Journal of Philosophy 76/4: 523–33.

Warfield, Ted, and David, M. (2008), ‘Knowledge-Closure and Skepticism’, in Q. Smith (ed.), Epistemology: New Essays (Oxford: OUP).

Wikforss, Asa Maria (2006), ‘Content Externalism and Fregean Sense’, in P. Marvan (ed.), What Determines Content? The Internalism/Externalism Dispute (Cambridge: Cambridge Scholars Press).

Wright, Crispin (2004), ‘Warrant for Nothing (And Foundations for Free?)’, Aristotelian Society Supplementary Volume 78/1: 167–212.

—— (2003), ‘Some Reflections on the Acquisition of Warrant by Inference’, in Nuccetelli (2003: 57–77).

—— (2002), ‘(Anti-)Sceptics Simple and Subtle: G. E. Moore and John McDowell’, Philosophy and Phenomenological Research 65/2: 330–48.

—— (2000), ‘Cogency and Question-Begging: Some Reflections on McKinsey's Paradox and Putnam's Proof’, Philosophical Issues 10: 140–63.


Warm thanks to members of the AHRC-funded Basic Knowledge project (now located at the Northern Institute of Philosophy at Aberdeen), to participants at the Self and Self-Knowledge colloquium held at Bigorio, Switzerland, in 2003, and at the Crispin Wright workshop held at the ANU in 2009, for helpful critical feedback. I am grateful to Walter Pedriali for detailed comments on the penultimate draft, deserving of extended discussion which I have not had space to embark on here. Special thanks to Jim Pryor for comments on an earlier version and thought-provoking conversations on the issues.

(1) Michael McKinsey (1991) must take credit for his notably trenchant formulation of the problem, so far as I am aware the earliest in print. An equally forceful but more explicit rendition of the presuppositions of the problem is Boghossian (1997).

(2) Sawyer (1998) advances this conclusion.

(3) I do not know if any philosopher has quite summoned the nerve to endorse this direction of response in any flat-out way. Jim Pryor (2007) has perhaps come closest. Pryor seemingly allows that the truth of premise (i) is not, once a suitably strong form of externalism is accepted, available to ordinary reflective self-knowledge; if that is accepted, it constitutes a retreat from the traditional conception of self-knowledge. However Pryor suggests that a conditionalized version of premise (A), whose antecedent hypothesizes the existence of the content, [water is wet], is verifiable by reflection, even in the setting of a relevant form of externalism, provided that its antecedent is true. This conditional is the strongest relevant claim available (in good circumstances) to reflection and it is of course insufficient to serve as a premise in the paradox, since that its antecedent is true is not verifiable by reflection. So the paradox is defused by the observation that, in this way, one of its premises as stated is indeed not properly knowable ‘a priori’.

The crux of this proposal is that, in contravention of a widespread assumption about internalist conceptions of epistemic warrant, certifiability by reflection—‘a priority’ in the loose and popular sense—doesn't necessarily iterate: that, for certain contents, P, it can be available to unsupplemented reflective thought that P is true but not available to unsupplemented reflective thought that the truth of P is so available. There is a point of contact here with the treatment I shall offer below. I will return to the matter (n. 27 below).

This is not Pryor's last word on the paradox. His final suggestion in his (2007) is that its premises, though individually reflectively warrantable, are not reflectively co-warrantable: that to the extent that one becomes convinced a priori of the correctness of strong externalism, to the same extent one's introspective warrant that the psychological premise reports a bona fide attitude of one's own, with a genuine content, is weakened. That seems to be to be tantamount to the concession that externalism does force a significant qualification of the traditional conception of self-knowledge if the philosophical case for it has been made. It is this concession that the treatment proposed in this paper is designed to avoid.

(4) Again, few have just flat-out rejected the idea of externalist constraint on content. Critics of externalism have usually wanted to concede some form of circumscribed validity to the externalist intuitions. Thus, for instance, Loar (1988) and Patterson (1990) attempt to distinguish the (externally determined) de dicto contents of the that-clauses featuring in mental-state ascriptions from the contents of the psychological states themselves, as determined by their role in the explanation of behaviour, so that subjects in one and the same psychological state can nevertheless, in virtue of differences in their respective environments, be properly described in non-equivalent ways. It is not clear if such a move really helps with the McKinsey paradox though. Even if the epistemic authority of the subject is restricted to the psychological content proper of her states, however that is to be conceived, rather than to their fine-grained description, there are versions of the paradox (see for instance McKinsey* in §IX below) which will still arise unless psychological content is Fregean, that is, referentially undemanding.

A sterner line against externalism, in the light of the paradox, is taken by Wikforss (2006).

(5) Apparently the ‘intuitions’ about Gettier cases that preponderate among discussants in Western analytical philosophy are not shared by native speakers of certain Far Eastern languages. Might the Twin Earth intuitions, e.g., prove to be in similar case? The question whether, and why, such a finding might compromise the philosophical arguments for externalism, rather than, as it were, localize them, is a very interesting one. But I shall not pursue it here.

(6) This response is developed in detail by McLaughlin and Tye (1998).

(7) A sophisticated discussion of some of the issues concerning the handling of premise (ii) by the fourth and fifth directions of attack is sections IIIV of Pryor (2007).

(8) I have previously canvassed this diagnosis in Wright (2000) and (2003). It was independently conceived by Martin Davies, and first published in his (1998). See also Davies (2000, 2003a, and 2003b). There are significant differences between Davies's attempts to detail the mechanism of warrant transmission failure in the present kind of case and mine.

(9) Although the question as formulated explicitly adverts to the multi-premise case, I shall, for ease of exposition, often focus on single premise arguments in the discussion to follow. I hope this will enjoin no loss of generality but the reader should beware there are substantial issues here that lie beyond the scope of the present paper.

(10) There is, of course, a relativity here: arguments are transmissive, or not, relative to a particular type of ground, or combination of grounds, for their premises. There is, for example, no doubt, or so it seems to me, about the transmissiveness of the McKinsey argument if the putative warrant for the second premise is not a priori philosophical but socio-inductive.

(11) For ease of exposition, I will focus on examples illustrating this kind of inferential triptych—evidence (a) defeasibly supporting an intermediate conclusion, (b), which entails a further conclusion, (c)—wherever possible.

(12) There has been some confusion about this point in the literature on Bayesian treatments of transmission failure. See e.g. Okasha (2004). Chandler (2009) attempts to improve on Okasha's account. A useful discussion of the issues is Moretti (2009).

(13) Wright (2000: 143; 2002: §IV; 2003: 59; 2004: 171).

(14) Wright (2002: 335–6).

(15) I did not mark this point at the time of writing.

(16) There is a distinction to be noted here between the issues attending this use of the phrase, “second-order warrant”, and Jim Pryor's discussion of a “higher-order background warrant” model of transmission-failure in his (forthcoming). Pryor is there speaking of a putative model of transmission-failure where higher-order epistemic conditions operate on the transmission of first-order warrant; a theorist might propose, for example, that in order for a certain kind of evidential warrant for its premises to transmit to the conclusion, B, of an argument, it is necessary that someone could rationally regard the evidence concerned as justifying the premises without thereby incurring any obligation to make out a claim to independent warrant for B. I am here concerned, by contrast, with conditions on the transmission of higher-order warrant itself: specifically, with the conditions an argument must meet if it is to enable a thinker who is in position to claim warrant for its premises to advance the argument in the service of a claim to warrant for its conclusion.

Much more needs to be said, naturally, about what it is to be in position to claim warrant. As I intend the notion, it demands a potential for reflective awareness that one is warranted and hence, at least if it is allowed that warrant may be conferred by external circumstances beyond reflective awareness, is not the same thing as having warrant for warrant. But I shall not attempt a more satisfyingly detailed account here.

(17) The original has, “C would be true if B were false”.

(18) Wright (2000).

(19) See Wright (2003: 76–7 n. 4) and McLaughlin (2003: 87–8).

(20) Brown (2004: 59). Brown was of course addressing an issue arising in a context where the distinction between first- and second-order warrant transmission had not been drawn, but I don't think anything of the motivation for her proposal depended on that.

(21) Wright (2004: 189).

(22) The qualification is important if we are to capture the intended notion. If it were waived, a prior opinion about the matter under investigation might rationally lead one to doubt the competence or significance of the project if that opinion happened to conflict with the upshot; and that would then have the effect that the very proposition which the project found in favour of would rank as a presupposition of the project concerned. I'll return to this below.

(23) If we ignored the qualification emphasized in n. 22, we might fear otherwise. For a rejection of the truthfulness of one's informant, or the representativeness of one's sample, e.g., might respectively be forced by prior convictions about the two propositions concerned.

(24) Perhaps they are occupants of Dry Earth—see Boghossian (1997: 185–7).

(25) I am aware of course that talk of a cognitive “project” may begin to seem quaint in such cases. But the oddity is merely terminological. The reader may find it helpful to understand the notion as covering any attempt to answer a question in a cognitively competent way, and the notion of a ‘presupposition’ of such a project as embracing any condition a doubt about which would rationally require doubt about the competence of one's answer, irrespective of what it was. I am indebted here to Jonathan Schaffer.

(26) See Wright (2004: 192) for details.

(27) To get into the matter just a little: how exactly is a priority, as an artefact of reference-fixing, supposed to be generated? Suppose I use a description to fix the reference of a new proper name—perhaps I am an 1890s journalist for the London Evening Chronicle, and coin the name, “Jack the Ripper”, as a moniker for the notorious unidentified serial murderer, which then catches on. If someone so learns the name, do they know a priori that

Jack the Ripper is the killer of all those London prostitutes?

One might think not, since it is not a priori that the dead women were killed by any single person, nor even that they were all killed, rather than died of causes other than their horrific injuries…So: perhaps it is at least a priori that

Jack the Ripper is the killer of all those London prostitutes if anyone is.

But even this is problematic. For if that conditional is knowable a priori, then it needs to be knowable a priori in circumstances where its antecedent fails. But in those circumstances there is no thought expressed by its consequent, it seems, unless “Jack the Ripper” is merely a concealed definite description, and so not a proper name, in the fashion proposed by those Millian and Kripkean theorists who think of names as semantically external expressions par excellence. Admittedly, this point is not sound if Jim Pryor is right—see n. 3 above—that the a priori knowability of such a conditional does not require its a priori knowability in all circumstances but depends precisely on the truth of its antecedent. But whatever one should think of that contention, it is clear at least that if that is to be the model on which the possibility of a priori knowledge of “If my speech community has historically had contact with water, it has historically had contact with a watery substance” is to be defended, then to lay claim to such knowledge is going to demand an independent claim to knowledge of the existence of water. And in that case the inference from B to B′ will not be at the service of a first-time claim to warrant for the existence of water in any case.

(28) Suggested to me by Jim Pryor in conversation. I have tweaked Pryor's suggestion slightly.

(29) At the APA Central Divisional symposium on Externalism, Self-knowledge, and Skepticism with Martin Davies and Brian McLaughlin, which McKinsey chaired, Minneapolis, May 3–6, 2001.

(30) For a useful discussion of some of the wrinkles, see Lasonen-Aarnio (2008) and Blome-Tillmann (2006). For defence of closure against one form of problematic case, see McBride (2009).