30

The argument for this is quite straightforward and recapitulates the considerations we have just been looking at. You don’t ever want the possession conditions for a concept to foreclose on the possible falsity of some particular set of claims about the world, if you can possibly avoid it. You want the possessor of the concept to be able coherently to ask whether there is anything that falls under it, and you want people to be able to disagree about whether there is. If in a certain range of cases, however, it is logically impossible to hold the governing theory at arm’s length then, in those cases, obviously, it can hardly be a requirement that one do so. But in all those cases where that is possible, it ought to be done. What about Sober? It should be clear, given the kind of conditionalization that is in view here, that not every meaningful term in a language can be thought of as expressing a concept that conditionalizes on the existence of an appropriate semantic value for it. Take the case of flurg'. The stipulation that would correspond to a conditional version forflurg’ would amount roughly to this: If there is a property which is such that, any elliptical equation has it, and if something has it, then it can be correlated with a modular form, then if x has that property, x is flurg. The corresponding introduction and elimination rules that would specify the possession condition for it would therefore be: and As this makes clear, the only thinkers who could follow such rules-and, hence, the only thinkers who may be seen as implicitly conditionalizing on the existence of an appropriate semantic value for flurg'-are those who (a) possess a basic set of logical constants and (b) are able to refer to and quantify over properties in particular, and semantic values, more generally. It follows, therefore, that conditional counterparts for one's primitive logical constants will not be available and hence that one could hardly be blamed for employing their unconditionalized versions. In particular, if the conditional is one of your primitive logical constants, you couldn't conditionalize on the existence of an appropriate truth function for it, for you would need it in order to conditionalize on anything. In such a case, there is no alternative but to acceptconditional theory’-modus ponens in effect-if you are so much as to have the conditional concept. It thus couldn’t be epistemically irresponsible of you just to go ahead and infer according to modus ponens without conditionalizing on the existence of an appropriate truth function for it-that is simply not a coherent option in this case. What is the full range of those concepts for which conditional counterparts would not be available? To answer this question, one would need to have a clear view of what the minimal logical resources are that are needed to conditionalize in the envisaged answer on the truth of an arbitrary theory, and I don’t have a systematic theory of that to present at the moment. What does seem clear, though, is that some set of basic logical constants would have to be presupposed and that is enough to get me the result that inference in accord with their constitutive rules can be entitling even though blind. If we go back to the MEC, then, it seems clear what we should say: Any rules that are written into the possession conditions for a non-defective concept are a fortiori entitling. With that in hand, we have the answer to our question: how could MPP premises warrant MPP conclusions while being blind? Answer: they do, because they are written into the possession conditions for the conditional, and the conditional is a non-defective concept. X. CONCLUSION If we are to make sense of the justified employment of our basic logical methods of inference, we must make sense of what I have called blind but blameless reasoning-a way of moving between thoughts that is justified even in the absence of any reflectively appreciable support for it. In this paper, I have attempted to sketch the outlines of an account of this phenomenon, one that avoids the pitfalls both of an overly austere Reliabilism and an overly intellectualized Internalism. The account seeks to revive and exploit two traditionally influential thoughts: first, that following certain inferential rules is constitutive of our grasp of the primitive logical constants; and, second, that if certain inferential rules are constitutive of our grasp of certain concepts, then we are eo ipso entitled to them, even in the absence of any reflectively appreciable support. 28 APPENDIX29 In his response to this paper, Timothy Williamson objected to this line of reasoning as follows: Although 3 and occur in the Carnap sentence 3FT(F) T(Neutrino), in place of that sentence Boghossian could have used the rule that allows one to infer T(Neutrino) directly from any premise of the form T(A). That rule is formulated without reference to the logical operators in the objectlanguage, but is interderivable with the Carnap sentence once one has the standard rules for 3 and -…. Logical operators may of course occur in the theory T itself, although Boghossian does not appeal to that point. In any case, it seems insufficiently general for his argument, since for some less highly theoretical concepts than neutrino, the analogue of the theory T for conditionalization may consist of some simple sentences free of logical operators. (Williamson 2003, p. 287) Of course, I did not mean to suggest that one could simply read off the Carnap sentence that existential quantification and conditional would be presupposed by any conditionalization, though no doubt my presentation was overly elliptical. In the cases of most central interest, the affirmation of the Carnap sentence would be implicit in the thinker’s behavior and could not be supposed to amount to an explicit belief from which one could simply read off the ingredient conceptual materials. To see whether we could have nothing but conditionalized concepts, we have to ask whether it is possible for someone to implicitly affirm the Carnap sentence for, e.g., boche, without possessing any of the logical concepts with which we would explicitly conditionalized our concepts. We have agreed that for someone to affirm T(boche) implicitly is for them to be willing to infer according to the following introduction and elimination rules: Gx/Bx Bx/Cx. Now, the question is: What would it be for a thinker to implicitly conditionalize his affirmation of T(boche) on the existence of an appropriate semantic value for these rules? Williamson says that this could be adequately captured by picturing the conditionalizing thinker as operating according to the following rule: But what this seems to me to say is something very different from what is needed. A thinker operating according to Williamson’s rule is like someone who already has the concept boche but is now simply relabeling it with the word `boche.’ Whereas what I want to capture is the idea of someone who is only prepared to infer according to the boche rules because they antecedently believe that There is a property F, such that Gx Fx and Fx – Cx And I don’t see how their reasoning could depend on that without their having, at a minimum, the conceptual materials that make up the antecedent of the Carnap sentence, including the quantificational apparatus and the conditionals that make up the statement of the theory. If all of this is right, it follows that conditional counterparts for one’s primitive logical constants will not be available and hence that one could hardly be blamed for employing their unconditionalized versions. In particular, you couldn’t conditionalize on the existence of an appropriate truth function for the conditional, for you would need it in order to conditionalize on anything. In such a case, there is no alternative but to accept “conditional theory”-modus ponens and Conditional Proof, in effect-if you are to so much as have the conditional concept. It thus couldn’t be epistemically irresponsible of you to just go ahead and infer according to MPP without conditionalizing on the existence of an appropriate truth function for it-that is simply not a coherent option in this case.