On Some Arguments for the Necessity of Necessity BOB HALE Must we believe in logical necessity? I examine an argument for an afrma- tive answer given by Ian McFetridge in his posthumously published paper Logical Necessity: Some Issues, and explain why it fails, as it stands, to establish his conclusion. I contend, however, that McFetridges argument can be effectively buttressed by drawing upon another argument aimed at es- tablishing that we ought to believe that some propositions are logically nec- essary, given by Crispin Wright in his paper Inventing Logical Necessity. My contention is that Wrights argument, whilst it likewise fails, as it stands, to establish the necessity of necessity, establishes enough to close off what appears to me to be the only effective-looking sceptical response to McFetridges original argument. My paper falls into four principal parts. In the rst, I expound McFetridges argument and draw attention to the possi- bility of a certain type of sceptical counter to it. In the second, I begin a re- sponse to this sceptical move, taking it as far as I can without reliance upon argument of the kind given by Wright. Turning, then, to Wrights argument, I explain to what extent I think it is successful and seek to rebut some objec- tions to the argument which, were they well-taken, would show that the ar- gument cannot enjoy even the partial success I wish to claim for it. Finally, I return to my main theme and try to show, with the assistance of what I take to be solidly established by Wrights argument, that the sceptical response collapses. Must we believe in logical necessity? An argument for an afrmative answer was given by Ian McFetridge in his posthumously published paper Logical Necessity: Some Issues (McFetridge 1990). In this paper, I expound and examine the argument I take McFetridge to have been propounding and explain why it fails, as it stands, to establish his conclusion. I believe, however, that it is possible to re-inforce McFetridges argument so that it does work. To explain how this can be done, I need to examine another argument aimed at establishing that we ought to believe that some propositions are logically necessaryhence the plural in my titlegiven by Crispin Wright in his paper Inventing Logical Necessity (Wright 1986). My contention is that Wrights argu- ment, whilst it likewise fails, as it stands, to establish the necessity of (belief in) necessity, establishes enough to close off what appears to me to be the only effective-looking sceptical response to McFetridges origi- nal argument. Since my proposed buttressing of McFetridges argument relies upon my claim that Wrights argument, while not completely suc- cessful, does secure a result very much to the purpose, I need to get clear precisely what that argument establishes. This requires a moderately 24 Bob Hale lengthy digression from my main theme, in consequence of which this paper falls into four principal parts. In the rst, I expound McFetridges argument as I understand it and draw attention to the possibility of a cer- tain type of sceptical counter to it. In the second, I begin what I hope will be an effective response to this sceptical move, taking it as far as I can without reliance upon arguments of the kind given by Wright. I turn, in the third part, to Wrights argument, explain to what extent I think it is successful and, in the course of doing so, seek to rebut some objections to the argument which, were they well-taken, would show that the argu- ment cannot enjoy even the partial success I wish to claim for it. Finally, I return to my main theme and try to show, with the assistance of what I take to be solidly established by Wrights argument, that the sceptical response collapses. 1. McFetridges argument, and a sceptical counter to it The argument with which I shall be concerned comes at the very end of McFetridges paper, and occupies little more than half of one page. As with many short but weighty arguments, however, a degree of preparation is needed to set the stage for it. The conclusion of the argument is that we must believe in logical necessity. But what is logical necessity? And what does belief in it amount to? McFetridge has much to say in the body of his paper about these questions, some of it essential to understanding and appreciating the force of his argument. The rst question may naturally be taken to split into two by shift of emphasis: rstly, what is logical necessity? and second, what is logical necessity? I am quite sure that McFetridge did not see himself as provid- ing a complete characterisation, much less a denition, either of the nar- rower notion of logical necessity or of the broader one of necessity; as we shall soon see, that does not greatly matter. Part of what he says on the rst of these questions is that logical necessity is the strongest kind of necessity, in the sense, approximately, that if it is logically necessary that p, then it is necessary that p in any other use of the notion of neces- sity there may be (physically, practically, etc.). But the converse need not be the case. Something could be for example physically necessary without being logically necessary (McFetridge 1990, p. 137). 1 This 1 McFetridgerightly, I thinkregards the assumption that logical necessity is the strongest kind of necessity as central to a traditional conception of logical necessity. It is worth emphasising that McFetridge does not simply assume that logical necessity has this feature; he gives a very interesting argument for it (1990, p. 138). This is one reason why he should not be seen as attempting to On Some Arguments for the Necessity of Necessity 25 explanation (that is of strongest) will not quite do as it stands, as McFetridge himself notes, because there are epistemic notions of possi- bility on which the contradictory of what is logically necessary may be epistemically possible. Thus on one view, one or other of Goldbachs Conjecture and its negation is logically necessary, but whichever of them it is, the other is epistemically possible (that is roughly, true for all we presently know). 2 The explanation can, I think, readily be adjusted to accommodate this complication. 3 One obvious reason whyeven if there were not other, decisive, reasons against doing soit would be unsatisfactory simply to dene logical necessity as the strongest kind of necessity is that this would leave it quite obscure what logical necessity has to do with logic. McFetridges second main claim speaks to this con- cern. He writes: Deductive validity is the central topic of logic. So if, as Aristotle and others have thought, to think of an argument as valid requires
dene logical necessity as the strongest kind of necessity. I think this argu- ment,which I shall not go into here, is basically sound, but establishes a some- what weaker conclusion than McFetridge claims. What the argument shows is not that logical necessity is the strongest kind of necessitythe highest grade of necessity, as he puts itbut that logical necessity is absolute, in the sense that if it is logically necessary that p, then there is no non-epistemic sense of possibility in which it is possible that not-p. The claim that a kind of necessity is absolute is equivalent to the claim that there is no stronger (non-epistemic) kind of necessity. To establish that logical necessity is the strongest kind of necessity would require showing that there cannot be more than one kind of absolute ne- cessity. It is not easy to see how this might be shown; indeed, it is not clear that it is even true. At any rate there is a signicant question that ought not to be fore- closed by simply dening logical necessity to be the strongest kind. See Hale (1996, pp. 93117) for a discussion of an argument for the absoluteness of logi- cal necessity very closely related McFetridges, and of the philosophical signi- cance of its conclusion. 2 The example is, of course, doubly tendentious. Intuitionists and constructiv- ists generally will not like it. Nor will those who agree that true number-theoretic propositions are necessary, but deny that the necessity can be properly regarded as logical. I leave readers of either persuasion to choose their own examples. Of course, no examples can be given that will commend themselves to those who deny the existence of logical necessity altogether, but that is obviously no cause for legitimate complaint. 3 The matter is not entirely straightforward. It would be easy enough to amend the characterization to: if it is logically necessary that p, then it is neces- sary that p in any other non-epistemic use of the notion of necessity there may be. But such an amendment is in good standing only if we can satisfactorily mark off epistemic notions of necessity. The question is how that should be done. I think that the salient feature of epistemically modal notions is that they involve a relativity to time in one way or another (for example a relativity to our current state of information), and that the best way to frame the needed amend- ment may thus be to exclude all notions of necessity which involve such relativ- ity. 26 Bob Hale us to deploy a notion of necessity, then that notion, if any, will de- serve the label logical necessity. There will be a legitimate no- tion of logical necessity only if there is a notion of necessity which attaches to the claim, concerning a deductively valid argu- ment, that if the premisses are true then so is the conclusion. (McFetridge 1990, p. 136) His thought, clearlyand, in my view, correctlyis that if there is any logical necessity at all, theor at least afundamental case of it lies in the connection between the premisses and conclusion of a valid argument. Accordingly, it will sufce to establish his main thesis to show that we ought to believe that at least some modes of inference are logically necessarily truth-preserving. That is why it is unneces- sary, for his purposes, to provide a fully general characterization, or general denition, either of necessity in general or of logical necessity in particular. What he does need to do is to explain what it is to believe, of a given mode of inference, that it has that character. His explanation (which, in my view, displays something close to philosophical genius in its originality and simplicity) can be seen as the product of two entirely familiar and uncontroversial points, one about valid deductive inference and the other about subjunctive or counterfactual condition- als. The point about deductive inference is, in essence, the very rst thing we all learn when we begin to study logic: that we must distin- guish the question whether an argument is valid from the question whether its premisses are true, the validity, or otherwise, of the argu- ment being independent of the truth-value(s) of the premiss(es). As McFetridge puts it, the acceptability in an argument of some mode of inference is supposed to be quite independent of whether or not the overall premisses of the argument represent beliefs we have or mere sup- positions we are making, as we put it, for the sake of argument. Deductive inferences, then, are supposed to remain valid when they are applied to mere suppositions, and indeed regardless of what suppositions they are applied to, or are made in the course of the argument. (McFetridge 1990, p. 151) The point about counterfactual conditionals is likewise, at least in essence, simple and familiar enough: our assessment of a conditional If it were (had been) the case that p, it would be (have been) the case that q is typ- ically, if not invariably, relative to or dependent upon various background suppositions concerning the truth-values of other statements. When we agree, for example, that if Simpson had not landed on the snow bridge, he would have fallen another 100 feet, probably to his death, we are envisag- ing Simpsons fall occurring in circumstances very much like those in which he actually fell, perhaps differing from the actual circumstances On Some Arguments for the Necessity of Necessity 27 only in that he unluckily misses the snow bridge 20 feet down into the cre- vasse and in other respects consequent upon that difference. We are not envisaging a situation in which the crevasse bottoms out at 30 feet, or in which there is no crevasse there at all. Whilst the point itself is simply stated, an explanation of the signicance McFetridge nds in it requires a few more words. He begins by reminding us of a familiar crux in the theory of counter- factuals. 4 If we try saying that a counterfactual is true if there are suit- able truths from which, in conjunction with its antecedent, its consequent may be inferred by means of a natural law, then we confront (as Good- man rst pointed out) two problems. How is the class of natural laws to be demarcated? And how are we to circumscribe the class of suitable truths from which, together with the antecedent, the consequent is to be inferred? We cannot allow just any true statement to gure as a supple- mentary premiss, since the negation of the antecedent will be true, and from this together with the antecedent itself, any consequent whatever may be drawn. 5 We would like to restrict the admissible supplementary premisses to those co-tenable with the antecedent, where that means: it is not the case that, were the antecedent true, they would be false. But that is hopelessly circular, if our goal is to explain the truth-conditions of counterfactuals. However, as McFetridge observes: this will not be a difculty if we change priorities and try rather to understand reasoning from a supposition (McFetridge 1990, p. 152). His thought, in part at least, is that if we relinquish the aim of explicating the truth-conditions of counterfactuals and raise instead questions about reasoning from sup- positions, there will be no objectionable circularity involved in charac- terising co-tenable suppositions in this way. What specic questions about reasoning from suppositions did he have in mind? Although he does not raise them explicitly, I think it is clear that two of them could be put like this: Given a certain suppositionthat pwhat modes of inference can be employed in reasoning under that supposition? and, for a given mode of inference M: Under what range of suppositions may M be employed? Modes of inference, here, need not be thought of as restricted to patterns of inference to which recognition is accorded in formal logic. We may think of (putative) natural laws, or lawlike general statements, asor as having associated with themmodes of inference. Indeed, we may even view singular statements of various kinds (including, most obviously, 4 What follows is my gloss on McFetridge (1990, pp. 1513). 5 At least if the logic is classical, or even intuitionistic. 28 Bob Hale conditionals) as being employable as such. 6 Of course, precisely because such statements involve expressions which tie them to a particular subject matter, their use as principles of inference will be limited in a way that the use of modes of inference such as modus ponens and modus tollens is not. But this is no objection to viewing them as employable as modes of infer- ence. Limitations of applicability of the kind just noted are limitations of relevance. The reason why I cannot use, say, the generalization that cop- per conducts electricity in guring out what will happen if I dig in lime at the rate of 100 grams per square metre where I plan to plant fritil- laries, is not that the generalization cannot be relied upon under this supposition; it is simply that the generalization has no bearing on the question that interests me. Of greater interest here is a different kind of limitation upon the employability of a mode of inference, of which McFetridge reminds us. The range of suppositions under which a natu- ral law, for example, may be employed as a principle of inference may be very large, but not so large that there is no supposition under which it may not properly be so used. In very many contexts, there is no objec- tion to reasoning in accordance with the inverse square law, say, assum- ing that it can be brought to bear upon the matters under consideration. But if we are trying to work out the observational consequences of a rival physical theory in which gravitational attraction is governed by an inverse cube law, say, then obviously we ought not to make inferences employing the usual square law. Somewhat differently, and equally familiarly, it may be that there are known boundary conditions on the law, satisfaction of which can often be taken for granted; but under sup- positions where their satisfaction is in question, so too is use of the law as a principle of inference. There may, then, be suppositions that can be entertained, in reasoning from or under which a mode of inference is not properly relied upon. We are now ready for McFetridges answer to the question about the content of a belief in logical necessity, or more exactly the content of the belief that a certain mode of inference is logically necessarily truth-pre- serving (so that the corresponding conditional is true as a matter of logical necessity): I suggest that we treat as the manifestation of the belief that a mode of inference is logically necessarily truth-preserving, the preparedness to employ that mode of inference in reasoning from any set of suppositions whatsoever. Such a preparedness evinces 6 Thinking of statements, whether lawlike or not, as usable as principles of in- ference need not, of course, involve any commitment to Ryles notorious attempt to make out that they are merely inference tickets and not genuine true or false statements at all. On Some Arguments for the Necessity of Necessity 29 the belief that, no matter what else was the case, the inferences would preserve truth. (McFetridge 1990, p. 153) 7 To draw out, more explicitly than McFetridge himself does, the contrast he is implicitly making between natural and logical laws, we mightwith a small, but so far as I can see, unobjectionable extension of Goodmans use of the term co-tenablecall the range of suppositions under which the use of a mode of inference M is not questionable for the sorts of reason touched upon its co-tenability range. Part of McFetridges thought is then that natural laws, employed as modes of inference, have typically a large, but not unrestricted, co-tenability range. And his earlier claim that logical necessity is the strongest kind of necessity can now be seen to require that there be no restrictions upon the co-tenability range of logical laws; the corresponding modes of inference would, as he puts it, preserve truth, no matter what else was the case, and so may legitimately be employed in reasoning from any set of suppositions whatsoever. 8 That completes the stage-setting. We are ready for the argument itself. On McFetridges account, to believe that a particular rule of inference R is logically necessarily truth-preserving is to believe that no matter what supposition s we may make, R is truth-preserving in reasoning under s. If we agree with him that the fundamental case of logical necessity, if there is such a thing at all, is that involved in valid inference, belief in logical necessity centrally involvesthough its content may not be exhausted 7 Fairly clearly, McFetridge is not suggesting that the content of a belief in log- ical necessity can be explicated in wholly non-modal terms. He is, rather, proposing that it can be explained in terms of acceptance of a kind of generalized counter- factual. In a little more detail, and going somewhat beyond anything he says in his paper, the idea can be put like this. To believe that an inference p, so q is neces- sarily truth-preserving is to believe the corresponding conditional If p, then q to be logically necessary, i.e. to believe that (p q). If we write for the coun- terfactual conditional, McFetridges thought is that the content of this belief can be given by: s(s (p q)). On any standard semantics for counterfactuals, he is clearly right, at least in the sense that the (logically) necessitated conditional and the corresponding generalized counterfactual are equivalent. If (p q) is true, then p q is true at all possible worlds, and hence at all nearest s-worlds, whatever s may be, so that s(s (p q)) is true. Conversely, if s(s (p q)) is true, then p q is true at all nearest s-worlds, however s is chosen. But since s can be chosen any way we like, this means we must have p q true at all worlds, that is (p q). Obviously the idea can be easily generalized to cover arbitrary beliefs in logical ne- cessity, say by explaining the belief that necessarily p as the belief that s(s p). 8 A Quinean sceptic about necessity may baulk at the characterisation of logi- cal laws as having a completely unrestricted co-tenability range. Such a thinker may agree that what are usually called logical laws have a more extensive co-ten- ability range than what are taken to be laws of nature, however fundamental. So much is merely an alternative expression of the view (to which he, with Quine, subscribes) that logical laws are more deeply entrenched in our overall system of beliefs than any others. But no statements (or principles of inference) are, he may protest, so deeply entrenched as to be unbudgeable (absolutely entrenched). No questions are here begged against this position. It is true enough that McFetridges 30 Bob Hale bybelief that there is at least one such rule. Accordingly, to deny that a particular rule R is logically necessarily truth-preserving is to hold that there is some supposition s such that, with s in play, we may not safely rea- son in accordance with R. And to disbelieve in the existence of logical necessity is to hold that for every rule R, there is some supposition s under which we may not safely employ R. McFetridge does not argue, concerning any particular rule, that it is logically necessarily truth-preserving. The argument he gives is designed to show, rather, that we must believe that some rules have this character. To abandon the belief in logical necessity, he writes, would be to believe that for every acceptable mode of inference M there is at least one proposition r (it might be a very long disjunction) such that it is illegiti- mate to employ M in an argument which makes the supposition that r (McFetridge 1990, p. 53). McFetridge argues that this sceptical position is incoherent. Before we turn to details, we should pause briey to note a point about the scope of his argument which is implicit in the remark last quoted. McFetridge equates abandoning the belief in logical necessity with believing the negation of what, on his proposal, is the content of the belief that at least some rules are logically necessarily truth-preserving, that is that for every rule, R, there is some supposition s such that, if it were the case that s, applications of R would not always preserve truth. The equa- tion might be challenged. In particular, a sceptic about necessity might think he can decline to accept that there are any rules which would pre- serve truth no matter what else was the case, but decline also to endorse the contradictory belief that for any rule, there are or could be circum- stances in which it would not preserve truth. This kind of sceptic might aptly be called an agnostic about necessity. On my construal of his argu- mentwhich seems clearly correct, given his gloss on what it is to aban- don belief in logical necessityMcFetridges target is the kind of sceptic who commits himself to the belief that there are no rules applications of which would be truth-preserving no matter what else was the case. This
characterization pre-empts the use of what a thinker of this persuasion might oth- erwise view as a perfectly serviceable piece of vocabulary: there is, the Quinean may hold, no objection to describing logical laws as (logically) necessary, any more than there is to describing those of physics as (physically) necessary; pro- vided that we do not fall into the error (as he supposes) of thinking that the neces- sity attaching to logical laws is any different in kind from that belonging to the laws of physics, or to any other statements deeply entrenched in our overall theory of nature. But whilst the Quinean may deplore the proposed usage, it begs no question against him: in particular, it is still open to him, for all that has been said thus far, to voice his doubt as the doubt that there are any logically necessarily truth-preserving modes of inference, in the sense proposed. On Some Arguments for the Necessity of Necessity 31 kind of sceptic might, by contrast, be called an atheist about necessity. Thus if agnosticism is a tenable position, it would seem that McFetridges argument, at least as I construe it, suffers from a serious limitation. Per- haps there is an alternative construal of his argument on which it does not suffer from this limitation, but if there is, I have not managed to get it into clear focus. In any case, given my construal of McFetridges argument, the question must obviously be confronted: What should be said to the would-be agnostic about necessity? The answer, I think, depends on what kind of agnostic he would be. The crucial question concerns the source of his unwillingness to be committed either to the belief that some rules are necessarily truth-preserving or to the belief that no rules are. He may be unwilling because, whilst he takes himself fully to understand the alterna- tives, he simply cannot see how to determine which should be accepted. This is straightforward agnosticism. But he may be unwilling because he doubts the coherence or intelligibility of the terms in which the alterna- tives are put, perhaps because, like Quine, he regards all modal talk as somehow defective and philosophically disreputable. The straightforward agnostic may, I think, be answered quite swiftly. This agnostic accepts the disjunction: Either for every rule R, there is some supposition s such that if it were the case that s, applications of R would not always preserve truth or there is at least one rule R which would preserve truth no matter what else were the case. If McFetridges argu- ment is good, it shows that the rst disjunct must be rejected. So unless she can nd fault with that argument, the straightforward agnostic should come off the fence and accept the other disjunct. Evidently a good answer to the more radical agnostic calls for many more words; more, certainly than I can spare here, where I can do no more than indicate the lines along which I think it should run. An answer might begin by pointing out that blanket rejection of the counterfactual condi- tional as simply unintelligible is not only desperate but clearly in tension with the plain fact that nearly everyone seems to understand it well enough, 9 so that this agnostic had better have some very convincing rea- sons for thinking that our apparent understanding is illusory. If the agnos- tic is Quine himself, it would be in point to observe that, ofcial professions to the contrary notwithstanding, he appears in practice to have no difculty in making sense of talk of statements holding true come what may (note may, not does), and that it does not appear possible to formulate the central theses of Two dogmas (Quine 1951) without using some modal word. But a fully satisfying answer would have to go 9 Well enough; I do not claim, and do not need to claim, that we are in posses- sion of unquestionably correct answers to all the hard questions which a philoso- pher might raise about the meaning, analysis or truth-conditions of counterfactual conditionals. 32 Bob Hale deeper than such sniping and challenge the eminently questionable assumptions about the constraints on intelligibility on which this kind of scepticism (about modal and intensional notions in general) feeds. 10 On my reading, then, McFetridges target is the sceptic who commits himself to the belief that there are no rules applications of which would be truth-preserving no matter what else were the case. His argument to the incoherence of this runs as follows: Either (a) for some R, the sceptic holds that we know (and so can fully state) a suitable s or (b) he holds that whilst there is, for every R, such an s, in no case do we know it (and so in no case are we able fully to state it). (a) collapses: let R be a rule for which we can fully state the relevant sup- position, which we shall denote s*. Then, where R is From X infer A, let R* be the rule From X, s* infer A. By hypothesis, s* is a full state- ment of the circumstances in which R (allegedly) fails to preserve truth, so that R* is a counter-example to the sceptics claim. So he must opt for (b). (b), however, is no good either, according to McFetridge, because its effect is to render (all) our rules of inference unusable. Let R be one of our rules, and let p be some supposition. To know whether we can rely upon R under the supposition that p, we have to ask: if it were the case that p, would R be truth-preserving? But answering this question will involve some reasoning, on the supposition that p. What rules shall we use, in car- rying out that piece of reasoning? Not R itself, obviously, since its reliabil- ity in the case in question is sub judice. But no other rule either. For to determine whether we can rely upon another rule, R, we have again to answer the question: were it the case that p, would R be reliable? So we are stuck, or off on an innite regress. As McFetridge has it the same problem will break out again, on the view that no rule is co-tenable with the universal range of suppositions (McFetridge 1990, p. 154). Hence we must accept that there are some rules which are unrestrictedly reliable, in the sense that there are no suppositions under which they may fail to pre- serve truth. To this argument as it stands there is, it seems to me, a fairly obvious counter. Why cant the sceptic retort along these lines: Why do you assume that if I am to use a rule R in reasoning under the supposition that p, I must first be able to ascertain whether R is, under that supposition, reliable? I dont have to do that. It is enough that I have no positive reason to doubt that R will fail un- der the supposition that p. 10 Many of the essential points are made by H. P. Grice and P. F. Strawson (1956), succinctly and forcefully reiterated by Crispin Wright (1986, see espe- cially pp. 18990). On Some Arguments for the Necessity of Necessity 33 This sceptic is claiming that it is not required, if we are to be entitled to employ a rule of inference R under the supposition that p, that we assure ourselves that, were it to be the case that p, rule R would (still) be truth- preserving. His attitude is that in employing R in reasoning under a sup- position, we make a defeasible assumption to that effectso long as that assumption remains undefeated, we have all the justication we need, and all the justication of which the case admits, for employing the rule. He believes in British justice: a rule is innocent until proven guilty. 2. A response to the sceptic begun The sceptics reply is plausible, but is it cogent? I think it is fair to say that the position he seeks to occupy will constitute no genuine alternative (i.e. to the belief that there must be some absolutely or unrestrictedly valid rules) unless we can take his falsicationist methodology seriously. I shall try to show that we cannot do so. If the sceptics professed falsicationist attitude towards rules of infer- ence is not to be empty, it requires us to think that, for any one of our rules R which has thus far survived all attempts to envisage its failure, it is nev- ertheless conceivable that some circumstances p should obtain, in which R would recognisably fail to be reliable. Falsicationism without the pos- sibility of recognisable falsication is not worthy of serious consideration. So let us consider, schematically, how this allegedly conceivable falsi- cation of a rule could come about. For deniteness, and without loss of generality, let us suppose R to be the rule of Conjunction Elimination (&E). What is alleged to be conceivable is that circumstances p might obtain, of which we could recognise that, if they did obtain, this rule would fail to preserve truth. Here p will have to be a description of cir- cumstances in which some particular conjunctive statementwhich we may represent as A and Bwould be true, and yet one of its conjuncts B saywould be false. The claim would be that since, by the rule in question, B can be inferred from A and B, we can see that, were it to be the case that p, this rule would fail. In performing this marvellous feat of recognition, we must accomplish several things which it will be worthwhile to list separately, obvious though they may be: (i) we have to recognise that, were it to be the case that p, A and B would be true (ii) we have to recognise that, were it to be the case that p, B would not be true 34 Bob Hale (iii) we have to recognise that our rule of &E cannot be truth-preserv- ing unless, were A and B to be true, B would be true also. There may be more, 11 but this will do to be going on with. Now, save in the case (which I suppose we may discount) 12 where p is some description of an allegedly possible circumstance which expressly states that A and B is true, discharging subtask (i) will involve some rea- soning. And the same goes, of course, for subtask (ii). Andthough this point needs rather careful handling 13 it goes for subtask (iii) as well. Let us, again without loss of generality, take the rst case. Some rea- soning has to go on hereto the conclusion, in this case, that were it the case that p, A and B would be trueand any reasoning proceeds, either explicitly or implicitly, in accordance with some rule(s). But which rule(s)? Not &E. Of course, we cannot now argue, as McFetridge did, that that rule cant be employed because it is (still) sub judice; for according to the position we are now discussing, all rules are always to be regarded as innocent until proven guilty. They may, if you wish, be said to be, in that sense, sub judice, but they are not so in any way which renders their use objectionable. But we can rule out the use of &E none the less. First, we can see that it could not, in any case, be the only rule employed. For if 11 There surely is more. In particular, we should have to recognise, given the contents of the acts of recognition (i)(iii), that were it to be the case that p, &E would fail to preserve truth. This clearly involves an inference, from the contents of those acts as premisses. So even if it could be maintainedas I do not believe it can bethat no reasoning need be involved in (i)(iii), it remains that the en- visaged recognition of the failure, in the circumstance that p, of &E to preserve truth must involve some reasoning. I conjecture that this point holds quite gener- ally: that whatever rule we are concerned with, and whatever the circumstances in which it allegedly fails to do its stuff, some reasoning will be required for the rec- ognition that the rule would fail in those circumstances. 12 I suppose we may discount this case because the bald supposition that A&B is true but B false is, prima facie, incoherent. What is needed is a spec- ication of some state of affairs which (a) appears to express a genuine possibility and (b) is such that we can be brought to see that, were it to obtain, A & B would be true but B false, or at least that it would be reasonable to accept the former but deny the latter. 13 The need for care derives from the fact that it is crucial to avoid claiming that applying the rule of &E itself involves some further reasoning. What we are concerned with here is not an actual application of the rule but what is required if, in certain envisaged circumstances, the rule is to be truth-preserving. This is rea- soning about the rule, not reasoning by it. Since the rule is general, and the claim that it is truth-preserving equally generalit is the claim that whenever a conjunc- tion is true, so must be each of its conjuncts separatelya denite step is involved in appreciating that if the rule is truth-preserving, then if, in the circumstance that p, A and B would be true, then, in the circumstance that p, B would likewise be true. This seems to involve a step of Universal Quantier Elimination at least, and surely also a step of modus ponens. Acknowledgment of this point does not commit us to the disastrous view that any application of &E has to be mediated by steps of Universal Quantier Elimination, modus ponens, etc. On Some Arguments for the Necessity of Necessity 35 it were to be the only rule employed, p would have to incorporate some statement of the form A and B is true and C. This just takes us back to the case where p explicitly states that A and B is true (or as good as does so), and I think we can discount that at this stage: if we were prepared to swallow this as part of a description of circumstances in which &E would fail, it is quite unclear why we should baulk at a description which just ran Suppose it were the case that A and B is true but B not true . Second, and more importantly, there appears, in any case, to be a fatal instability involved in supposing that we might reason, under the suppo- sition that p, by &E to the conclusion that &E would fail under that sup- position; if we believed that the argument succeeded, we ought not to believe it. If this second claim is correct, then it follows, of coursegiven that some reasoning by some rule is requiredthat some rule other than &E must be used. 14 It might be objected that this second claim just overlooks the possibility that the reasoning in question could proceed by reductio ad absurdum, or something like it; that is, that we could, assuming for the sake of argument that &E is reliable in reasoning under the supposition that p, argue, by applying that rule under that supposition, to the conclusion that it isnt reliable in reasoning under that supposition, and then, appealing to the principle: A A A, conclude that it is not reliable under that suppo- sition. More generally, if we can show, by reasoning employing a pro- posed rule R, that R is not guaranteed to preserve truth, why shouldnt we conclude that R is not guaranteed truth-preserving? On the supposition that R is reliable, we can show that it is not; so it isnt. This objection, it seems to me, misses a crucial difference between reg- ular applications of reductio ad absurdum (or the closely related principle that A A A) and the case with which we are concerned. When we argue by reductio to a conclusion A, what entitles us to our conclusion is that we have a valid argument from the supposition that A to a contradic- tion. Our (continuing) entitlement to the conclusion depends upon our continuing endorsement of that deduction of the contradiction from the initial supposition. Any well-founded doubt about the validity of that deduction would deprive us of our ground for accepting the conclusion. But this condition is precisely not met, in the present case. If our reasoning really did show that rule R is not reliable in reasoning under the supposi- tion that p, it would showsince we reasoned by R under the supposition that pthat that very piece of reasoning cannot be relied upon, thereby 14 There are two separate points here. The rst is that at any rate &E cant be the only rule involved, the second that it had better not be involved at all. Actually, it is the rst pointthat the reasoning must make use of some other rule(s)that is more important for the argument I am giving at the moment. But the second point (or, more accurately, a generalisation of it) will be important later. 36 Bob Hale depriving us of our warrant for the conclusion that R is not reliable in rea- soning under the supposition that p. Likewise with the conclusion A reached by an application of A A A: our entitlement to the conclu- sion depends on our entitlement to the premiss A A. Accepting A deprives us, in the present case (though not, of course, in general) of our ground for accepting A A, and thereby removes our ground for accept- ing A. 15 We may take it, then, that some rule R other than &E is deployed in the reasoning of case (i). There is no need to speculate about what rule this could be; assuming the enterprise is coherent, that will clearly depend upon the exact character of the supposition that p. But it does not matter, for our purposes, what rule R is. Nor can it be properly objected that, before our falsicationist may deploy R, he must rst determine whether its use is good under the supposition that p; for his claim was precisely that we quite justiably treat rules as good until they are revealed as bad. So provided that we have no positive reason to mistrust R, we can use it. But theres the rub. Granting, for arguments sake, that we have so far failed to locate any denite supposition under which R cannot be relied upon to preserve truth, what reason can the falsicationist have, in the present case, for taking it that it is &E, rather than R, that fails under the supposition that p? This looks like a fair question, since someone pre- sented with an appearance that &E fails under the supposition that p might elect to stand by &E, declare that something must have gone wrong, and point the nger of suspicion at R. So long as that is an option, we have no determinate falsication of &E (or of R, of course). Since it always will be an option, the whole idea that all rules of inference have merely default 15 To guard against a possible misunderstanding, I should emphasize that the claim I make here does not mean that a rule of inference cannot be, in a certain sense, self-undermining. Consider, for example, the Crazy Rule: From X, infer A, which allows us to infer any conclusion we choose from any given set of prem- isses, including the null set. Since the Crazy Rule allows us to draw any conclu- sion we wish, we could apply it to obtain, in particular, the conclusion that the Crazy Rule is not truth-preserving. Thus if the Crazy Rule were truth-preserving, it would not be; so it is not. There is nothing wrong with this argument, which is a perfectly good reductio of the supposition that the Crazy Rule is truth-preserv- ing. But this in no way conicts with the claim made in the text about the impos- sibility of showing, by a reductio which uses R under the supposition that p, that R cannot be relied upon if used under the supposition that p. There is no conict, because our reductio of the supposition that the Crazy Rule is sound does not use the Crazy Rule. What it relies upon is not an actual application of the Crazy Rule, but a claimthe correctness of which can be appreciated without reasoning by the Crazy Rule itselfabout what, if the Crazy Rule were accepted, we could in- fer by its means. Someone who rejects the Crazy Rule need have no more dif- culty is seeing that this claim is correct than an intuitionist logician has in seeing the correctness of claims about what inferences can be made by means of Double Negation Elimination. On Some Arguments for the Necessity of Necessity 37 acceptabilitycan be treated as innocent until proven guiltyis a sham, because there is no such thing as proving a rule guilty. To this, it may be retorted: That will indeed be a problem, if you sup- pose that hypotheses in general, and hypotheses about which rules are truth-preserving in particular, ought to be determinately falsiable inde- pendently of any appeal to pragmatic considerations. But why should we not just agree that there will be various options open, and let pragmatic criteria decide which one to go for? So long as such considerations could favour the verdict that &E fails under the supposition that p, that is enough. Must McFetridges bold venture bog down, short of its destination, in the swamps of holistic-pragmatic methodology? I do not believe so. To explain why not, I need to take a small digression, through the other argu- ment for the necessity of necessity mentioned at the outset. If all goes well, we shall see that, whilst that argument (i.e. Wrights) does not, as it stands, establish that any statements must be accepted as necessarily true, it does provide the resources needed to break the present impasse. 3. InterludeWrights anti-Quine argument 3.1 The argument Wrights argument (1986, pp. 1924) is directed against the radical form of empiricism advocated by Quine, according to which all statements whatever are subject to revision in response to recalcitrant experience, with revision in any particular case being a holistic affair essentially involving the application of broadly pragmatic considerations. It runs as follows. Let be some theory we are putting to the test and L our underlying logic. We derive from , using L, various conditional statements whose antecedents describe observationally checkable initial conditions, and whose consequents specify observable predicted outcomes. Let IP be any such. A series of observations E will be recalcitrant (more fully, recal- citrant with respect to + L) if it provides, or appears to provide, grounds to accept I but reject P. Quine accepts that faced with E, we should make some adjustment, but holds that we have many options open. These include: (1) Replacing by some * differing from by omission of one or more of the statements which figure as premisses in the L-deriva- tion of I P 38 Bob Hale (2) Rejecting one or more of the L-axioms or rules required for deri- vation of I P from (3) Denying that E does, after all, warrant acceptance of I and denial of P. Since, however, Es recalcitrance depends upon the conditional I P being derivable, using L, from , Quine mustgiven his claim that all statements have the status of no more than hypotheses, revisable in the face of recalcitrant experiencerecognise a fourth option, viz. that of denying that E is, after all, recalcitrant (with respect to our currently accepted package of theory + logic), even if it warrants denial of I P, by way of denying that this conditional is a consequence of that package, that is, to deny the statement which Wright labels: (W) L I P. In other words, that E is recalcitrant must be regarded by Quine as no more than a hypothesis, no less subject than any other hypotheses involved to assessment and possible rejection via the proper application of his pragmatic methodology. Wrights argument focuses on the question: under what circumstances will it be reasonable, by Quines lights, to retain W. As in the case of other available responses, pragmatic consider- ations should be allowed to determine whether this is a good move. But how, Wright asks, is that to be determined? The decisive consideration ought to be, presumably, the degree of further recalcitrance with which the various alternative courses tend to be beset. But once the recalcitrance of experience be- comes a hypothetical matter, the question is transformed into: how often are the various alternative courses beset by sequences of experience which, according to the best hypothesis, are recal- citrant? And now, in order to decide whether recalcitrance is the best hypothesis, we have to consider how it tends to fare in prag- matic competition with the alternativesand the beckoning re- gress is evident. So the official Quinean answer to the question, when is it reasonable to believe a statement like W, is no answer. (Wright 1986, p. 193) Wrights thought, as I read it, may be elaborated as follows: Provided that it is not a merely hypothetical matter what the consequences (and in par- ticular, the observationally checkable consequences) of a given combina- tion of theory and logic + L are, and provided that it is similarly not hypothetical what the consequences of alternatives + L* are (where and/or L* are variants on and L), we may make comparative assessments of + L and + L* in regard to overall simplicity, explanatory power and whatever else we take to be included among the broadly pragmatic param- eters against which we think competing theories should be evaluated. In particularthough the notion doubtless calls for further elucidation and On Some Arguments for the Necessity of Necessity 39 renementwe can make some comparative assessment of the degrees of recalcitrance attaching to them; that is, roughly, the extent to which each clashes with observationally attested phenomena by yielding conditional predictions whose antecedents appear to be fullled, but whose conse- quents do not. That one of the competitors suffers from a higher degree of recalcitrance than the others will be a strong (though not necessarily deci- sive) consideration against it. But decisive or not, relative degree of recal- citrance will clearly be an essential consideration in pragmatically guided theory choice. Now suppose that we treat W and its kin as mere hypotheses. The imme- diate effect, of course, is that any assessment of the relative merits of com- petitors + L and + L* which we may have made along the foregoing lines now takes on a provisional character, conditional upon certain claims like W which we are now viewing as hypotheses. Under different hypoth- eses about their observational consequences, a comparative assessment might go quite differently. Further, since Es recalcitrance with respect to our currently accepted combination + L is contingent upon acceptance of W, we are confronted with the further option of retaining that combina- tion by rejecting W and thereby dissolving the appearance that E is after all recalcitrant. To make progress, we need to determine whether we should retain or jettison W. To do this, we must assess, inter alia, the rel- ative degrees of recalcitrance involved in combinations respectively retaining and rejecting W. But any comparative assessment we may make will again be merely conditional, upon hypotheses about the conse- quences of the options under consideration. Since all such hypotheses are in the pragmatic melting pot along with all other statements, we have no progressonly regress. 3.2 Some doubts about the argument answered As with any regress argument, doubts about the cogency of Wrights argu- ment will focus on one or other of two questions: Do we really have a regress? and, if so: Is it vicious? Obviously I cannot anticipate and respond to every way in which a doubt of one kind or the other might be raised against Wrights argument. I shall therefore conne myself to what seem to me to be the most promising counter moves. Someone might concede that there is a regress, but deny that it is vicious on something like the following grounds: The regress would indeed be vicious, if the possibility of con- structing it showed that Quines pragmatic methodology [QPM] cannot be applied to determine what choice to make in any par- ticular situation of prima facie recalcitrance. It would show this if it were true that, in order to decide what to do in any given case, Quine would have to resolve each of an infinite sequence of ques- 40 Bob Hale tions of the form: Is such-and-such a hypothesis beset with a higher or lower degree of overall recalcitrance than the other op- tions available? But QPM does not require us, when confronted with recalcitrant experiences, to consider all the possible options which are, in some sense, theoretically open. Given any particular theory , apparently confuted by an E which inclines us to accept I but reject P, where I P is derivable in L from , there will be very many ways of adjusting to some * which does not have this consequence. Most of them, we simply wont have thought of. QPM doesnt require us to think out all these *seven if it were possible for us to do so (as it surely isnt)and assess their relative degrees of recalcitrance. It is, rather, a methodology for deciding among already given options, viz. the ones which have actually occurred to us. If further, previously unthought of, vari- ants on are subsequently suggested, we should evaluate them (in comparison with options already considered), as and when they are proposed for consideration. The same applies to other options, such as adjusting our logic (whilst leaving as it is), so that I P is no longer derivable from , and equallythe crucial point hereto the option of leaving both and our logic L in place, but rejecting W. 16 One point should, I think, be conceded to the above: it would be quite unreasonable to require the Quinean, when faced with an apparently recalcitrant train of experience E, to review all possible adjustments to his currently accepted belief-system which would lead to an accommodation of E. Rather, as the objector says, QPM is a methodology for choosing from among some given (typically small, and necessarily nite) collection of actually proposed options. But this concession does not dispose of the difculty. It is true that, once the concession is made, it cannot simply be assumed that the option of denying recalcitrance by denying some conse- quence hypothesis such as W must be among the range of options actually up for evaluation. But this does not matter. To see why it does not matter, suppose that Quine, who has hitherto subscribed to , but has just had recalcitrant experience E, is actually con- sidering a few options of the type: adjust to *. And suppose that, after assessing the pros and cons, he plumps for one of them. In so doing, he implicitly claims that it is reasonable to move from to this alternative. Since this would not be reasonable, if E were not recalcitrant with respect to , and that is so only upon the assumption W, he is taking it that accep- tance of W is reasonable. That is, his actual pragmatically guided choice of adjustment is really conditional upon Ws acceptance. The point here is not that there may be a further option in addition to those Quine actually considered, such that, if he had included it for con- sideration along with the others, he might have come to a different deci- 16 Something like this response was suggested by my colleague Philip Percival. On Some Arguments for the Necessity of Necessity 41 sion. Nor is it that Quine cannot make his choice of adjustment without rst determining whether or not acceptance of W is reasonable. Of course, he cannot show that that choice is reasonable without doing that, but that is a further issue. Nothing stops him just assuming that acceptance of W is reasonable. The crucial question concerns, rather, what it is for the assumption to be reasonable, by Quines lights. Once claims about the consequences of a given combination of theory+logic (and hence state- ments about its attendant degree of overall recalcitrance) are treated as mere hypotheses, in the pragmatic melting pot along with the rest, there can no longer be any determinate truth about what degree of recalcitrance attaches to any theory+logic combination, with the upshot that there is no target for the pragmatist to aim at. 17 So whilst the argument, if it succeeds, does show that QPM is infeasible, this is not because applying it demands carrying through an innite process (to the bitter non-end, as it were), but because it proves to be completely illusory that there is a determinate con- ception of what a successful adjustment of a belief-system would be, by pragmatic lights. Since the point is crucial, it is worth re-stating more fully. Let O W be the option of keeping + L and accommodating E by denying W, and let O 1 , , O k be the other options up for consideration. Then acceptance of W will be reasonable, ceteris paribus, provided that the degree of recalci- trance attaching to O W exceeds that of at least one of the O i . But what degree of recalcitrance attaches to O W depends upon what its conse- quences are. This is not determined merely by the rejection of W. O W (i.e. + L + W) no more has determinate consequences on its own than does any of the other combinations of theory+logic under consideration. Its consequences have to be settled by appeal to some further statement which, for Quine, can only be a further hypothesis. But which hypothesis? Since, for different choices of H, we shall get different assessments of the degree of recalcitrance attaching to O W + H, we should, presumably, take that H which minimizes recalcitrance. But the claim that some particular H (perhaps one among a small range of alternatives actually considered) minimizes recalcitrance is itself a further hypothesis again, which we may call H. H has its own range of competitors, viz. rival hypotheses about which of the hypotheses co-ordinate with H minimizes the degree of recalcitrance besetting O W . Once again, we should, presumably (and as always, ceteris paribus), select that one among H and its rivals which is 17 This, I take it, is the thought in or underlying Wrights remark: The moral is that it cannot be a correct account of the basis of our confi- dence in statements like W that belief-systems in which they figure enjoy relative success; if that were the right account of the matter, there could be no explaining the requisite notion of success. (Wright 1986, p. 193, emphasis mine) 42 Bob Hale attended by the lowest degree of recalcitrance. But that in turn is a hypo- thetical matter, the relevant hypotheses being H and its competitors . So long as the degree of recalcitrance attaching to hypotheses about con- sequence is itself held to be a further, hypothetical matteras it must be, for Quinethere can be no stepping out of the regress. But unless we can step out, no determinate degree of recalcitrance attaches to any hypothe- sis, and in consequence of that, no denite content attaches to the assump- tion that it is (or isnt) reasonable to retain W. 18 So none attaches, either, to the supposition that some one among the original, limited range of options actually considered (i.e. O 1 , ,O k ) is pragmatically best. So the regress is vicious. It may be conceded that the crucial question concerns what it is for the assumption W to be reasonable, by Quines lights, but denied that he need answer it in a way that generates the threatened regress. The contrary appearance, it may be claimed, results from giving a question-begging, non-pragmatic answer. The Quinean answer should be: any continued acceptance of any statement historically accepted is, by default, reason- able, provided adjustments have been made somewhere so as to defuse what was perceived as recalcitrance. The Quinean background to this answer is that the pragmatist starts from the status quo: a question of what it is reasonable to believe only arises for some specied array of options against the background of an extant web of belief. So judgements of rea- sonableness are always doubly relative: relative to a background of assumptions and relative to the options under consideration. The back- ground of assumptions, being inherited from an earlier stage, is reason- able by default. So it is false that acceptance of W will be reasonable only if the degree of recalcitrance attaching to O W exceeds that of at least one of the O i . This answer would be correct only if rejecting W were among the options being explicitly considered. But if W belongs to the web of beliefs not considered for revision on this occasion, then its continued acceptance is reasonable by default. 19 This defence rests upon three claims: (A) The question of what is reasonably believed only arises in regard to an explicitly mooted set of options concerning what adjustment (if any) we should make to our total belief-system, faced with some ostensibly recalcitrant experience, and then arises only against a background of historically accepted assumptions (B) The background assumptions are themselves reasonable by de- fault 18 Hence Wrights remark (1986, p. 194) that the Quinean injunction (roughly: adjust so as to minimize recalcitrance) is hopelessly impredicative. 19 This line of defence was proposed by my colleague Jim Edwards. On Some Arguments for the Necessity of Necessity 43 (C) There is no reason why W should not be one of these background assumptions. It would sufce, to undercut the defence, to reject any one of them. I shall, however, argue against all three. First, a preliminary point.There is an obvious prima facie inconsistency between (A) and (B). (A) seems to imply that no question can arise as to the reasonableness of background beliefs, while (B) declares those beliefs to be reasonable, which they can hardly be if the question of their reasonableness cannot even arise. I take it that this apparent inconsistency is to be resolved by drawing a distinc- tion, in effect, between the way or sense in which the background beliefs are reasonable, on the one hand, and the reasonableness that is in question in relation to the options actually up for assessment in any given case. The former are default reasonable, and default reasonableness is not a matter of being what is best believed, where that is to be assessed by pragmatic criteria. But the preferred option, among those actually up for evaluation, is precisely the one which, among those options, comes out tops by prag- matic criteria. 20 The apparent inconsistency is to be avoided by reading (A) as claiming The question of what is reasonably believed by prag- matic criteria only arises in regard to an explicitly mooted set of options and only against a background of historically accepted assump- tions. So, the proponent of this defence is not refusing my question: What is it for the assumption of W to be reasonable, by Quines lights?. Rather, he is answering it: It is for it to be default reasonable (as glossed above), with the aim of blocking the regress by denying that the correct answer to the question turns on what would be the result of applying prag- matic criteria. I begin with (B). Quines defender sees that it will not do just to say, without qualication, that any continued acceptance of just any statement historically accepted is, by default, reasonable. Hence the qualication: provided adjustments have been made somewhere so as to defuse what was perceived as recalcitrance. But I do not see how even with the qual- ication, this can possibly be right, as it stands. Can it really be the case that retention of a previously accepted hypothesis is default reasonable, provided some adjustmentsno matter whichhave been made to defuse what was perceived as recalcitrance? Suppose, in some previous situation of our being confronted with recalcitrance, we made a bad choicebad, by Quinean pragmatic guidelines, I mean. (Nothing guarantees that we never misapply QPM.) We considered options involving rejecting some 20 This assumes, but only for convenience, that pragmatic criteria will deter- mine a unique best option. If there isnt (i.e. if there are two or more options be- tween which pragmatic criteria do not decide), then we are free to choose from among the pragmatically best options, and whichever we choose will be reason- able. 44 Bob Hale statement S, but plumped for some other option and retained S. In fact, we miscalculated; perhaps we over-estimated the overall degree of recalci- trance attaching to the former options, or under-estimated that attaching to the option we went for, or perhaps both. We should have dumped S, but we didnt. But we did make an adjustment somewhere so as to defuse what was perceived as recalcitrance, so the proviso is met. Does that make our retention of historically accepted S reasonable by default? How can the mere fact of historical acceptance make it default reasonable to continue to accept a statement which we should have previously rejected, and would have rejected, had we not misapplied Quines pragmatic directives? In real life, we have to live with our mistakes, no doubt (at least to the extent that we cannot undo them), but this does not make them any less mistaken. In one way, indeed, the comparison with practical affairs is unduly favourable, for there we are often justied in saying: Weve gone down that road now, its too late to turn back and we must just stick to it and make the best of a bad jobbut we are never, so far as I can see, justied in saying anything really analogous to this when it comes to theory. The incorrectness of (A) is a straightforward corollary of the objection to (B). If the argument against (B) is sound, then there can be cases in which the continued acceptance of at least some of our historically accepted back- ground beliefs is reasonable, if at all, not by default but only because, when the option of rejecting them was previously entertained, they wereby the proper application of pragmatic criteriareasonably retained. So the argu- ment against (B) shows that there can be cases in which at least some of our historically accepted background beliefs are subject to assessment as reasonable or not by pragmatic criteria. So, even when (A) is qualied as I have said it should be, it is false, because it implies that the question of reasonableness, by pragmatic criteria, can arise only for options explicitly up for assessment, and not for any background beliefs. It is true, I think, that this argument only gets us the limited conclusion that the question of reasonableness-by-pragmatic-criteria (briey, p-rea- sonableness) can arise for certain of our background beliefs, those for which the option of rejection has been previously considered. That is, it does not follow that the question of p-reasonableness must arise for all background beliefs. So someone might, for all that I have argued thus far, insist on the minimum concession, and say: Very well, I admit that the question of p-reasonableness is properly raised in relation to those of our background beliefs for which the option of rejection has, as a matter of historical fact, been actually entertained. But as for the rest of our back- ground beliefs, the question of p-reasonableness (as distinct from reason- ableness simplicter) is out of order and they are simply default reasonable. This cannot but appear a pretty desperate move, if only On Some Arguments for the Necessity of Necessity 45 because it seems just plain wrong to make the admissibility of a question turn on whether that question has, as it happens, been previously asked. The argument against (B) shows that a question as to the p-reasonableness of some statement can be in good order even when none of the options explicitly proposed explicitly raises that question. What justies putting so much weight on the sheer contingency whether the question of a state- ments p-reasonableness has previously been raised? Even this desperate move will, in any case, be futile unless we accept claim (C), that W may be counted among the background assumptions. As against this, I think that in any reasonable sense of historically accepted background assumption, (C) should be rejected. W is, recall, a quite spe- cic statement about one particular (putative) consequence of + Lthe statement I P. Prior to the occurrence of the putatively recalcitrant E, we need never have considered W. So if the force of the claim that W is one of our historically accepted assumptions is that we have previously explic- itly endorsed that very statement, it will be an entirely contingent matter whether it holds true. I do not need to deny that cases are possible in which the claim, so understood, does in fact hold, since I am free to stipulate that I am concerned with cases in which it does not. If the claim is rather that, while W may not be among those statements we have previously explicitly considered and endorsed, it is nevertheless to be reckoned among our background assumptions because it is implied by others which have been explicitly endorsed, then it is regressive. Its effect is to shift attention to the further consequence claimW is a consequence of such-and-such explicitly endorsed background assumptionswhich gives rise, at one remove, to difculties precisely analogous to those aficting W. 3.3 What does Wrights argument establish? The immediate, negative, conclusion Wright draws from his argument is: the reasonableness, or otherwise, of judgements of recalcitrance must be exempted from appraisal via the Quinean methodology. And that must go for the ingredients in such judgements, includ- ing statements like W . If we are supremely certain of the truth of at least some such statements, the source of this certainty sim- ply cannot be accounted for by Quines generalized holistic mod- el. The very coherence of the model requires an account of a different sort. (Wright 1986, p. 194) I think that Wrights argument entitles him to this conclusion, but I shall not, here anyway, attempt to furnish further support for that claim, beyond 46 Bob Hale what I have tried to provide in the preceding sub-section. 21 Wright himself makes three further claims. First, he goes on, more positively, to claim: The right account is the obvious one: such statements, or at least an important subclass of them, admit of totally convincing proof. We must, I suggest, take seriously the idea of proof, as a theoretically uncontaminated source of rational belief. (Wright 1986, p. 194) By this last phrase, Wright pretty clearly means that the capacity of a suit- able proof to warrant belief is independent of any claim that is properly subjected to empirical appraisal. Thus his rst further claim is tantamount to the thesis that suitable proofs can yield a priori knowledge, or at least a priori warranted belief. 22 A little later, Wright makes it plain that he thinks the argument entitles him to claim: First, that we do possess some sort of concept of logical neces- sity. Second, the correct account of the basis for the majority of judgements of logical necessity which we are prepared to make must make reference to the utterly convincing, self-contained character of suitable proofs. (Wright 1986, p. 195) However, as McFetridge observed (1990, p. 149) and as Wright now accepts, there is a step herefrom the immediate negative conclusion and the rst further claim Wright makes to the claims just quotedwhich appears quite unwarranted. It is one thing to accept that we have to regard a proof as establishing a statement (such as a W) in a way that is not sub- ject to holistic appraisal, and quite another to hold that a statement thus established is necessary. What the argument given shows, at best, is that we have to accept such statements as established, by suitable proofs, as true; nothing in the argument demands their acceptance as necessarily true. 23
There is also, it should be observed, a gapthough, in my view, a much smaller onebetween Wrights immediate conclusion and his rst further claim. The immediate conclusion, as rst enunciated, is that the reason- ableness, or otherwise, of judgements of recalcitrance must be exempted 21 McFetridge, though dissatised with Wrights argument as an argument for necessity, agrees that it establishes at least this much (cf. McFetridge 1990, pp. 1489). Indeed, he thinks it establishes Wrights rst further claim. 22 That this is a fair gloss on Wrights actual words seems clear from, inter alia, his critical remarks about the incoherence of Quines attempt to Duhemise the traditional realm of the a priori (cf. Wright 1986, p.194). 23 I think it is possible to bridge the gapin effect, between a priority and ne- cessitywhich McFetridge identies in Wrights argument. But to go into that here would make too great a distraction from present concerns, so I shall leave the matter for another occasion. McFetridge diagnosed the gap as resulting from an inadequacy in Wrights account of what it is to regard a statement as logically nec- essary, and that his own argument for the necessity of necessity takes its start from what he regards as a better account of it (cf. McFetridge 1990, p. 150). On Some Arguments for the Necessity of Necessity 47 from appraisal via the Quinean methodology. And that must go for the ingredients in such judgements, including statements like W (Wright 1986, p. 194). If we read the antecedent of the conditional which comes soon after (viz. If we are supremely certain of the truth of at least some such statements, the source of this certainty simply cannot be accounted for by Quines generalised holistic model) as hypothesising rational (as distinct from merely unshakeable, but perhaps dogmatic) conviction, then the conditional itself, even if not quite a mere restatement in other words of the immediate conclusion, is at least a corollary of it. The rst further claim, viz. The right account is the obvious one: such statements, or at least an important subclass of them, admit of totally convincing proof (Wright 1986, p. 194), evidently presupposes that we are rationally war- ranted in accepting judgements of recalcitrance (along with such state- ments as W, on which they depend). I think there can be little doubt that Wright intended his intermediate claimThe very coherence of the model requires an account of a different sortto imply that that presup- position is fullled. But that begs the question whether the coherence of the model requires such an account at all. What if some recalcitrant Qui- nean agrees that statements of the kind in question must be exempted from appraisal via the Quinean methodology but denies that their acceptance can be rationally justied in any other way? What should we say to him? The rst thing that should be said is that this would represent a very sig- nicant concession, no hint of which is to be found in Two dogmas . Not only does it involve abandoning any aspiration that the Quinean meth- odology might be quite generally applicable; it is directly inconsistent with one of the most prominent, and most controversial, claims of that article: viz. that every statement whatever is subject to revision in face of recalcitrant experience. Further, once it is conceded that certain kinds of statement must be exempted from the scope of that claim, and from assessment in the only terms Quine allows, an explanation is called for. Clearly it would be insufcient explanation for the Quinean to retort that such statements must be exempted from holistic empirical appraisal, because the attempt so to appraise them causes that methodology to crash. So much is obvious enough, but not to the purpose. What is wanted is an explanation of the principle on which exceptions are to be recognised. It is quite unclear that, much less how, the Quinean can provide one without giving the game away, and unclear, too, how he could ght his corner against an opponent whose less straight-jacketed conception of the vari- ous means by which statements of different kinds may be rationally sup- ported enables her to explain both why holistic empirical appraisal is appropriate to a large but limited range of statements and why it is not 48 Bob Hale properly applied to others (including many whose acceptance can be war- ranted a priori by deductive reasoning). It would be rash to hope that these brief remarks might do more than indicate one direction in which opposition to Quinean epistemological holism might be developed. Important as the issue is, I shall not pursue it further here. Instead, I want to return to McFetridges argument and, with some assistance from the argument of this section, complete my response to the sceptical challenge we left it facing. As will soon appear, the use to which I mean to put Wrights argument does not depend upon making good Wrights positive thesis that statements of logical consequence (and, in particular, statements like W) can be warranted a priori. 4. The response to the sceptic completed The sceptic with whom we were grappling in 2 sought to brush aside the objection that his professed falsicationism about rules of inference was fraudulentbecause he could give no content to the idea of determinate falsication of any such ruleby recourse to the idea that he need not be committed to holding that there can be any such thing as determinate fal- sication independently of broadly pragmatic considerations. Once such considerations are allowed a role in the decision whether to regard a rule as falsied, he claims, his position can be seen to be a perfectly workable one. It should be clear that this move takes us into what is by now familiar territory. To continue with the example of a supposed falsication of &E, the suggestion is that in the choice between the options before usof holding that &E fails under the supposition that p, and of holding that it is not &E but R which then failswe should be guided by consideration of which option does best across a range of pragmatic desiderata such as simplicity, economy, overall theoretical unity, etc. Whatever other desid- erata are to be considered here, one of them will certainly, and crucially, be minimization of avoidable recalcitrance. Thus whatever other desider- ata are to be taken into account, and however they are to be weighted together, one question we shall certainly face concerns the relative degrees of recalcitrance besetting the options before us. Answering this question will require us to carry out some reasoning, to ascertain the con- sequences, respectively, of taking &E to be unreliable under the supposi- tion that p, and of taking R to be unreliable. There arises, accordingly, a question about what rules of inference are to be employed in this reason- ing. And now, it seems clear, the sceptic faces a fresh dilemma. Either this On Some Arguments for the Necessity of Necessity 49 reasoning will involve the application of rules other than &E and R, or it wont. Either way, the sceptic is in trouble. Suppose, rst, that it does, and let R 1 , , R m be the other rules involved. Then there will be open to us, in addition to the options of concluding that &E isnt invariably truth-preserving, and of concluding that R isnt, the further options of drawing the conclusion that R 1 is not invariably truth- preserving (and in particular, fails under the supposition that p), or that R 2 isnt, or or that R m isnt. If we ask: Why should we stand by these other rules, and see the shortcoming as lying with one of &E and R?, the answer, in general terms, has to be that pragmatic considerations favour this course. But that answer will once again rest upon an assessment of the degrees of recalcitrance besetting these various options, including the options of modifying or scrapping one or more of the R i . This assessment will involve some reasoning . We are at the start of a familiar regress. Suppose instead that only &E and R are involved in the deduction of the consequences of rejecting &E. There are three cases to consider, accord- ing as (a) both &E and R are employed, (b) only &E is used, or (c) only R is used. Cases (a) and (b) can be dismissed right away; for the reason already given, the case for rejecting &E will be fatally unstable if it relies on reasoning in which &E itself is essentially involved. This leaves just case (c). But this is no good; if R is used, it will always remain an option to deny that the alleged consequences of rejecting &E really are so, by way of rejecting R. I conclude that the proposed falsicationist methodology, if applied to rules of inference quite generally (including those we might otherwise view as necessarily truth-preserving), cannot be sustained. It collapses because it fails to give real content to the idea that rules can be selectively falsied. The attempt to mitigate that failure by appealing to the idea that pragmatic considerations might guide selective retention and rejection of rules breaks down because it leads into a vicious innite regress of essen- tially the same character as that disclosed by Wrights argument. That completes the main argument of this paper. To close, I would like to return to a limitation on the argument noted earlier and venture a sug- gestion about how it might be lifted. The limitation is that the argument at best establishes a general conclusion: we must believe that some rules of inference are logically necessarily truth-preserving. In and of itself, the argument does not tell us, concerning any particular rules, that we must believe them to be necessarily truth-preserving. Is there a way in which we might advance to some conclusions of that stronger form? I think there may be, a way that makes another connection with Wrights argument, or more precisely, with the rst of the further claims Wright makes, to the effect that some statements must be reckoned not only to lie beyond the 50 Bob Hale scope of holistic empirical appraisal but to be capable of being warrant- edly believed a priori. If, as I think we should, we accept Wrights second claim, then there is clearly a good deal more to be said about what the range of a priori war- rantable statements comprises. Wrights argument singles out a particular kind of statement as thus assertible, namely: statements about logical con- sequence. As Wright is careful to point out, our entitlement to assert such statements of consequence does not depend upon acceptance of the spe- cic underlying logic L to which they refer. The point may be simply illus- trated by reference to a classical derivation of p from p: 1 (1) p assn 1 (2) p 1, DNE An intuitionist, or other logician who rejects DNE, is not thereby pre- vented from recognising that this is a classically correct derivation, estab- lishing that p C p. One might be tempted to infer from this that a priori recognition of the correctness of consequence statements is entirely inde- pendent of acceptance of any logical principles whatever. But this temp- tation should be resisted, as can quickly be seen by considering how, even if we are intuitionists, we might convince ourselves of the correctness of the foregoing consequence claim on the basis of the short derivation above. In recognizing its sole ingredient step as classically correct, we reason as follows: In classical logic, if a statement is of the form A, the corre- sponding statement A may be inferred from it as sole premiss, with the conclusion depending upon whatever assumptions the premiss depends upon. The statement p is of the form A, and depends upon assumption 1. So, classically, the state- ment p may be inferred from that premiss, depending upon the same assumption. This inference is, of course, entirely neutral with respect to acceptance/ rejection of the principle of double negation elimination. But it is an infer- ence, by modus ponens, and it would seem to follow that to the extent that we can be warranted, a priori, in asserting that p C p, we must be warranted a priori in our acceptance of that principle of inference. This raises an obvious, and obviously interesting question, namely: just what logical principles are included in the minimal kit, as it were, needed for (a priori) appreciation of the correctness of claims about logical conse- quence (i.e. claims about what follows from what, relative to a pre- assigned underlying logic)? I think it is clear that this minimal kit will include at least (some form of) modus ponens (or some principle equiva- lent to it, at least in the presence of other minimal principles), and that it will also include some principle(s) of quantificational inference (such as On Some Arguments for the Necessity of Necessity 51 E). What else it may include is a question I must leave for another occa- sion. 24 Meanwhile, I offer a tempting, albeit tentative, conclusion: whatever the exact contents of the collection of rules of inference we should believe to be logically necessarily truth-preserving, it will include, at least, modus ponens and universal quantifier elimination (or their equivalents), together, of course, with any other rules indirectly justifiable by their use. 25 Department of Philosophy BOB HALE The University of Glasgow Glasgow G12 8QQ Scotland email: R.Hale@philosophy.arts.gla.ac.uk REFERENCES Grice, H. P. and Strawson, P. F. 1956: In Defense of a Dogma. The Philosophical Review, 65, pp. 14158. Hale, Bob 1996: Absolute Necessities, in James Tomberlin (ed.) Philo- sophical Perspectives, 10, Metaphysics, pp. 93117. McFetridge, I. G. 1990: Logical Necessity: Some Issues, in John Hal- dane and Roger Scruton (eds.) Logical Necessity & Other Essays Aristotelian Society Series, vol. 11, pp. 13554/ Quine, W. V. O. 1951: Two Dogmas of Empiricism in his From a Log- ical Point of View, Harvard: Harvard University Press, 1953. Originally published in The Philosophical Review 60, pp. 2043. 24 An interesting question is whether the minimal kit will include some princi- ple(s) governing negation, and if so, which. It seems to me that, even if no nega- tion principles are needed for the ratication of consequence claims of the kind were concerned with, acceptance of the principle of non-contradiction may be in- tegral to operation with the idea that recalcitrance obliges us to make some revi- sion in our overall corpus of accepted statements. This will carry with it acceptance of some form of the principle of reductio ad absurdumat least a weak, non-classical version, i.e. from X, A B, togethre with Y, A B, infer X, Y Aand therewith, in the presence of modus ponens, acceptance of the principle of modus tollens, i.e. from X AB, together with Y B, infer X,Y A. 25 Earlier versions of this paper, or parts of it, were presented at a conference on Modality held in Bled, Slovenia, at a seminar in Cambridge and at the XXth World Congress of Philosophy in Boston. I am grateful to my audiences on these occasions, as I am to my colleagues Jim Edwards, Philip Percival and Adam Rieger. I am particularly indebted to Crispin Wright for many hours of helpful dis- cussion and much constructive criticism. I should also like to give special thanks to an anonymous referee for Mind, whose patient, careful and perceptive com- ments have helped me to write a paper which, though appreciably longer than the one originally submitted, is at least in other respects greatly improved. Much of the work was carried out during my tenure of a British Academy Research Read- ership; I am grateful to the Academy for its generous support. 52 Bob Hale Wright, Crispin 1986: Inventing Logical Necessity in Jeremy Butter- field (ed.) Language, Mind and Logic. Cambridge: Cambridge University Press 1986, pp. 187209.