1 Concepts, Experience and Modal Knowledge 1 C.S. Jenkins Draft only – please check with the author before quoting from or citing this draft

Abstract This paper focuses on two questions concerning modal epistemology: 1. How can the senses ground our knowledge of how things must or might be, as opposed to merely knowledge of how things are? 2. How can conceivability be a guide to what is possible or necessary? A single proposal will be made, with the aim of making progress towards answering both questions. I shall suggest in answer to question 1 that, in at least some cases, the senses may ground modal knowledge by providing what I call epistemic grounding for our concepts, which concepts (help to) determine what we can and cannot conceive of, which in turn guides our modal beliefs. And in answer to question 2 I shall suggest that the fact that our concepts are grounded in this way is what accounts for the fact that conceivability – which I take to be a matter of what our concepts will allow us to conceive of – can be a guide to independent modal truth.

1 Experience and Modal Knowledge 1

I am grateful to members of the Arché Modality Project Team at the University of St Andrews for many

conversations about this material: Ross Cameron, John Divers, Bob Hale, Aviv Hoffmann, Daniel Nolan, Sonia Roca, Anna Sherratt and Crispin Wright. Thanks also to Katherine Hawley and an audience at University College, Cork.

2 In this paper I shall be discussing our knowledge of modal facts – facts, that is, of metaphysical (as opposed to nomic, epistemic, or any other kind of) modality. In particular, I shall be discussing the roles of experience and conceivability in modal epistemology.

Empirical

knowledge for the purposes of this paper is knowledge which either is, or is ultimately derived (through deduction, inference or other such rational transitions) from, knowledge obtained in a way that involves some essential use of the senses.2 Empiricists for current purposes are those who hold that all knowledge of the mind-independent world is empirical. (For details of my preferred notion of mind-independence, please see Jenkins 2005a.) It has been widely accepted, since (at least) the time of Plato, that the senses can only supply us with knowledge of how things in fact happen to be, not of how things must be. Philosophers as diverse as Plato, Hume and Kant are all agreed on this. Whewell (1840, pp. 5961) offers a particularly vivid expression of this line of thought, saying that experience:

can observe and record what has happened; but she cannot find, in any case, or in any accumulation of cases, any reason for what must happen … To learn a proposition by experience, and to see it to be necessarily true, are two altogether distinct processes of thought … If anyone does not clearly comprehend this distinction of necessary and contingent truths, he will not be able to go along with us in our researches into the foundations of human knowledge; nor indeed, to pursue with success any speculation on the subject.

2

„Essential‟ uses of the senses are to be contrasted with uses of the senses which merely „prepare‟ or „enable‟ the

mind so that it can begin the process of securing knowledge. The latter are not supposed to be epistemologically significant, the former are. (See the Introduction to Kant‟s Critique of Pure Reason for the classic exposition of such a distinction.)

3 What explains this unusual degree of consensus among philosophers, and such strength of conviction? One would imagine there must be some really solid motivations for the view that experience cannot underwrite our knowledge of facts about necessity. I want to begin by investigating whether that is so. There does indeed seem to be a common strand of motivating thought (when, that is, it is thought necessary to mention one at all). Here is a statement of it, due to Craig (1985, 104):

[W]hat affects my senses is the fact of the tree‟s being there; it wouldn‟t affect them any differently if its being there were necessary. „That‟, it can be said, „is just to repeat what is well known, that sight, touch and so on are not sensitive to necessity‟.

I shall discuss this comment in some detail in a moment. First let me note that not all knowledge of necessities is as divorced from empirical input as Craig‟s comment suggests. Some necessities – the famous a posteriori necessities – clearly require experiential grounding in order to be known. However, it is not clear that we can secure knowledge of these necessities without already knowing some other necessity facts which we rely on to secure the new knowledge. (For instance, it might be that we need to know that water is necessarily water, as well as that water is H2O, before we can know a posteriori that water is necessarily H2O.) So even if it is obvious how experience plays a role in grounding our knowledge of these a posteriori necessities, it is not obvious how experience could underwrite our knowledge of facts about necessity in general, including any that we need to know already in order to be able to secure this kind of a posteriori modal knowledge.

4 Similarly, although some knowledge of facts about possibility is fairly clearly empirical, not all of it is so easily handled. I see that it is raining, and I deduce that it is possible for it to rain. This way of securing knowledge of a possibility fact requires me to use the rule form of the modal principle T, which allows ◊p to be deduced from p, but it is not for that reason any less an empirical piece of knowledge. (Analogously, when I see that the apple in my fridge is red and deduce that something in my fridge is red, my knowledge of the latter is clearly a piece of empirical knowledge although I have employed the rule of existential introduction to get to it.) But not all our knowledge of possibilities is of this straightforwardly empirical kind. It certainly seems that we have knowledge of a great many facts of the form ◊p despite not knowing the corresponding propositions p. I know, for instance, that it is possible for there to be a red apple in my fridge, although I also know that there isn‟t one. I also know many facts of the form p. It is an interesting question whether these kinds of knowledge might be grounded in experience. Again, although it is fairly clear that testimony is an empirical source of a posteriori knowledge of various kinds of facts about necessity and possibility, it is also evident that there are other ways of finding out such facts. There are a lot of modal facts that we don‟t need to wait to be told (including many which are not obviously deduced from empirically-known non-modal facts). And in any case, somebody must at some stage have worked them out without being told, or else the chain of testimonial knowledge could never have got started. Knowledge of modality – at least, such modal knowledge as is not of the straightforwardly empirical kinds considered above – may seem to present a dilemma for the empiricist. If the senses cannot supply knowledge of all these modal facts that we seem to know,

5 then it seems the empiricist must either deny that there are mind-independent facts about what is possible and necessary (or at least that there are anything like as many such facts as we ordinarily think there are), or else deny that we can know the modal facts we think we can know. Some familiar options open up at this point in the debate. These include going rationalist and claiming that we have a faculty of „rational intuition‟ which we can employ to gain nonempirical modal knowledge,3 or arguing that modal facts belong to some special class of facts: facts which can be known without relying on sense experience because they are not, despite appearances, facts about the independent world. This latter is the option preferred by Craig (and by other contemporaries – see e.g. Wright 1980). Various forms of this view are available, including versions corresponding to Ayer‟s conventionalist account of knowledge of logic and mathematics (see Ayer 1936, chapter 4) and Carnap‟s framework account of knowledge of numbers (see Carnap 1950). Less familiarly, one might try to question the original claim, that the senses cannot supply us with all the requisite knowledge of mind-independent modal facts. Without more detailed supporting argument than we have seen so far, this claim cannot be said to amount to a decisive objection to empiricism concerning modal knowledge, but is merely a gesture towards a certain kind of worry which needs to be more fully spelled out and argued for. Of course, we do not see or hear or bump into or otherwise causally interact with modal states of affairs. But neither do we see or hear or bump into or otherwise causally interact with events inside the event horizons 3

An option dismissed by O‟Leary-Hawthorne (1996, p. 187) as „at best a little embarrassing, at worst downright

silly‟. At the less silly end of this spectrum are the moderate rationalist views of (e.g.) Peacocke (2000) and Bealer (2000), who at least do not posit a special faculty of rational intuition, but rather try to assimilate the required ability to other abilities we possess.

6 of black holes. We (or rather, the experts to whom we leave such things) infer the existence of such events because our best empirically supported theories predict them. It is not immediately clear what exactly is wrong with a simplistic empiricist response to the challenge, which asserts that the relevant modal truths are empirically known qua parts of our best empirically supported theories.4 Much hangs here on what sorts of roles modal truths can be taken to play in our theories, but the point remains that insofar as an objection to empiricism is to be made by arguing that the senses cannot supply all the modal knowledge we think we have, the burden of proof would appear to be with the objector to show that they cannot play the sort of role they would need to play in order for the simplistic modal empiricism described above to work. It is far from being so evident that all – or even any – of our modal knowledge is non-experiential that we do not need to give any arguments for that claim, and that (as Whewell supposes) anyone who is suspicious of it should be disbarred from researches into the foundations of human knowledge. One way to set about trying to show that modal truths cannot play the required role in our best empirically supported theories might be to generalize Craig‟s point, quoted on p. 3 above. That is to say, one could try arguing that, for any modal proposition P, 5 even if matters stood differently with regard to P, our sensory input would still be the same. I shall call this an insensitivity counterfactual, since it is supposed to establish that sensory input is not sensitive to the modal facts. I shall allow for the sake of argument that something in the vicinity of this insensitivity

4

5

Versions of this type of response include forms of Quinean holism. I.e., for current purposes, any proposition governed by a modal operator.

7 counterfactual could present a problem for the simplistic empiricist view. I note, however, it is far from clear exactly how this would be established, for it is not at all obvious what the relevance of this kind of counterfactual is to whether or not appearance in our best theory could support a modal proposition. (Much of the discussion surrounding the indispensability argument in mathematics would seem to be relevant to this point; for a good overview of this field, see Colyvan 2001.) An initial point to note about the insensitivity counterfactual as stated is that (provided our modal logic is as strong as S5) it has an impossible antecedent, which, given the classic Lewisian semantics for the counterfactual, will render it trivially true. 6 However, I don‟t want to make much of this.

Even a proponent of the traditional semantics can appreciate that some

counterpossible conditionals are non-trivially interesting as well as trivially true. For instance, certain counterpossible conditionals can appear at the beginning of persuasive reductio proofs (e.g. „If √2 were rational it could be written as n/m with n, m integers‟), because there are nontrivial reasons for taking these counterpossibles to be true – reasons which can be accepted by someone who does not already accept that their antecedents are impossible. I am not claiming that this epistemic feature adequately, or most perspicuously, characterizes the non-triviality of the counterpossible conditionals in question – there is much more to be said on this topic – but I do take it to be indicative of non-triviality of the kind I‟m interested in if a counterpossible conditional has this feature. There is a more interesting point to note about the claim that, for any modal proposition P, even if things stood differently with regard to P, our sensory input would still be the same. This 6

Lewis 1973, §1.6.

8 is that, absent reasons having to do with the impossibility of the antecedent, it is far from clear what reasons we might have to think that it is true. To take a simple case, suppose that (it is nontrivially true that) if it were no longer contingent that there‟s a church outside my window, then that proposition would be necessarily false. If so, given that there is in fact a church outside my window, then if things stood differently with regard to P‟s contingency, my sensory input would be very different. In particular, it would no longer look to me as if there were a church outside my window. We need a more careful expression of the insensitivity counterfactual to avoid interference from this quarter. We could try: for any proposition P, even if P‟s modal status were different, provided that all the non-modal facts were the same, our sensory input would also be the same. But this is trivial (for reasons having nothing to do with the impossibility of its antecedent). The facts about how our sensory input goes are non-modal facts, so obviously if the non-modal facts did not change they would not change. Here‟s a better effort: C: For any proposition P, even if the modal status of P were different, provided that P‟s truth-value was the same, our sensory input would also be the same. To claim that C is non-trivially true is, I think, to raise an interesting challenge, and one which the empiricist can productively engage with. From here on, I shall refer to C as „the insensitivity counterfactual‟. Assuming, as I am for the sake of argument, that the non-trivial truth of C would be a problem for the simplistic empiricist, in this paper I shall be exploring the plausibility of a kind of empiricism which doubts that there are non-trivial reasons for taking C to be true, and proposes instead that (insofar as we can make non-trivial evaluations of any of these kinds of

9 counterfactuals), there are non-trivial reasons for thinking that, for propositions P whose modal status we know, if the modal status of P were different, then even if P‟s truth-value was the same, our sensory input would be different. I shall call this „the sensitivity counterfactual‟.

2 Conceivability and Modal Knowledge In counterpoint to the prima facie problems with looking to experience as the source of our modal knowledge, there is a traditional, and prima facie rather compelling, thought that conceivability has some role to play in explaining how we know many of the modal truths we do know. This tradition goes back at least to Descartes (1641) and Hume (1739). In fact, however, once we scratch the surface we find that a number of troubling issues surround this proposal too. Take, for instance, the claim that a situation is possible iff we can conceive of its obtaining. There is a common worry about the insufficiency of conceivability for possibility: the worry that some conceivable situations are not really possible. A posteriori impossibilities of the sort described by Kripke (1980) may be a source of examples here, and some have even flirted with the idea that obvious impossibilities, not of this special kind, are conceivable (see e.g. Nolan 1997, footnote 11). Then there is another well-aired worry about the non-necessity of conceivability for possibility: the worry that some possible situations are inconceivable. We might look to quantum mechanics, or to non-Euclidean geometry as a theory of physical space, as possible sources of examples of inconceivable yet actual – never mind merely possible – situations. Then there is the issue of how we should characterize the notion of conceiving, and in particular the question of whether to understand conceiving in an internalist way (that is, as being

10 such that you can tell from the inside whether or not you‟re doing it) or in an externalist way (that is, as something you might not always be able to tell from the inside whether you‟re doing). Corresponding to this choice is a dilemma. It seems that in order to be a guide to modal truth, conceivability should be both reliable and transparent.7 If our way of knowing about possibility is through conceivability, there should be a sufficiently good match between being able to conceive of a proposition P being true and P‟s being possible (reliability), and we should be able to tell whether or not we can conceive of P‟s being true (transparency).

If we

characterize conceiving in the internalist way it is transparent, but what guarantees that it will be reliable? The insufficiency and non-necessity worries described above might be answerable if we can say that, in the problem cases, the conceivability or inconceivability in question is only apparent, not genuine. But the internalist approach to conceivability seems to collapse the required apparent/genuine distinction. On the other hand, if we characterize conceivability in an externalist way we seem to have more scope for making it reliable (indeed, some externalist proposals might simply build reliability into the conditions on genuine conceiving), but at a cost to transparency. How now can we tell whether we‟re really conceiving of P‟s being true, or whether it just seems as if we are? All of these are interesting issues which, despite their importance, will not concern me here. I mention them only in order to set them aside for the time being, and to note that efforts are already underway to address them, particularly the insufficiency and non-necessity worries.8 My second focal point in this paper will instead be one of the simplest, but I think also one of the

7 8

I am grateful to Sonia Roca for a helpful discussion here. See e.g. Chalmers 2002.

11 deepest, worries about the link between conceivability and possibility. That is the question of why we should think conceivability has, or even could have, anything to do with metaphysical modality. One way of raising this question is to note that, in order for conceivability to be a guide to modal truth, our powers of conceiving would have to be attuned to modal fact. The concern, then, is that, as Yablo puts it (1993, pp. 3-4, with nods to similar concerns in Blackburn 1986 and Wright 1986):

Not only are we aware of no bodily mechanism attuned to reality‟s modal aspects, it is unclear how such a mechanism could work even in principle.

The problem as Wright (1980, pp. 439-40) sees it is that if we are unable to conceive of something, then „that is how things are with us; it is a further, tendentious step to inflate our imaginative limitations into a metaphysical discovery‟. A rationale for taking such a further step (or at least, a way of reducing the amount of ground such a step needs to cover) might be forthcoming if we were prepared to adopt some form of mind-dependence anti-realism about modality, whereby modal truth was taken to be dependent upon ourselves in such a way as to make plain how conceivability could be an epistemic guide to it. For instance, if we thought that modal truth was generated by the way we use our concepts to divide up the world, then it wouldn‟t be wholly surprising that what we can conceive of – assuming this is somehow tied to the way our concepts are – is a guide to what is possible.

12 On such a view, for instance, we might argue that we cannot conceive of a something which is red and green all over because our concepts of red and green won‟t allow us coherently to represent a thing as being red and green all over. And then we can say that the fact that they will not do so is an excellent sign of the impossibility of such a thing, since the fact that they will not do so is constitutive of (or at least metaphysically underwrites) this modal fact.9 It is no coincidence that Hume, who believed in a strong connection between what he calls imaginability and possibility (1739, I.ii.2) also believed in the mind-dependence of necessity (1739, I.iii.14). It is tempting to think that something so apparently introspective and non-world-directed as conceivability could only be a guide to mind-independent modal truth if we possessed some faculty of rational intuition. But many of us would prefer to preserve our realist intuitions about metaphysical modality, and will not be prepared to compromise them for the sake of an easy life on the epistemological front unless there is no alternative. So I think we should explore the possibility of respecting both modal realism and the thought that conceivability is an epistemic guide to possibility, while still eschewing appeal to a faculty of rational intuition. Philosophical work on the relationship between conceivability and possibility has tended to focus on the non-necessity and insufficiency worries alluded to above. 9

(Gendler and

Modal epistemology is not by any means straightforward for such an anti-realist; for one thing it is far from

obvious that modal truths should be epistemically accessible to us through an examination of our concepts just because modal reality depends metaphysically on how our concepts are. (It could be, for instance, that our concepts have many features of which we are incapable of becoming aware, but which go to determine how things stand with modal reality.) But at least she has some response to the challenge to say what, in broad terms, what the relevance of conceivability to modal truth might be.

13 Hawthorne 2002 offer a nice summary of the debate which makes this clear.)

However,

questions in the vicinity of my target question, that of why we might think that conceivability can be any sort of guide to modal truth in the first place, have recently exercised Peacocke (e.g. 2000), and Bealer (e.g. 2000). Both have been lead to the conclusion that the question can be answered by some form of „moderate rationalism‟ (though they mean somewhat different things by that phrase). I have some sympathies with this „rationalist renaissance‟ (as Bealer 2002 calls it), and indeed my own answer to the question shares some significant features with Peacocke‟s in particular (which I describe in Jenkins 2008, §2.5). But I believe the central insights of these moderate rationalist proposals can be preserved within an empiricist framework; taking conceivability as a guide to modal knowledge is not the preserve of the rationalist. Traditional forms of empiricism cannot accommodate them, it is true; but by paying careful attention to the relationship between experience and concepts, I think we can develop a form of empiricism which can. In arguing for this claim I will be relating the question of how conceivability can be a guide to modal truth to my other guiding question, that of how the senses can ground modal knowledge.

3 Concept Grounding I propose that, for present purposes at least, we construe conceivability in terms of what follows epistemically from our concepts (a notion I will explicate in a moment), such that a proposition p is conceivable iff p‟s falsity does not follow epistemically from our concepts. I do not mean by this to imply that it is possible for us with our actual cognitive abilities to conceive of every proposition whose falsity does not follow from our concepts; of course, many such propositions

14 are too complex for us even to entertain. What I mean is that all such propositions are conceivable in the sense that our concepts permit of their being conceived, so that there is no bar to their being conceived except such familiar cognitive limitations as our being unable to entertain propositions beyond a certain level of complexity. Concepts, for my purposes, are sub-propositional mental representations. I take it that they are available to introspection, in the sense that we can learn by introspection (some of the facts about) what they are like and how they relate to certain other concepts. Some of these relationships are what generate the conceivability and inconceivability results that we use as a guide to modal truth.

For instance, the relationship between our concepts bachelor and

unmarried (plausibly something akin to inclusion) is what is responsible for the inconceivability of there being a married bachelor. Gendler and Hawthorne (2002, p. 1, footnote 1) briefly discuss the etymology of the words „conceive‟ and „concept‟, which have a common root in the Latin verb concipere. They claim that modern usage allows „a broad sense for the term „conceive‟ that permits as instances certain uses of …. “envisage, envision, fancy, fantasize, image, imagine, picture, see think, vision, visualize” – that is, a use of the term that is non-committal on the relation between conceiving and concept-deployment.‟ I am not sure what notion of „concept-deployment‟ Gendler and Hawthorne have in mind here, but the activities they list, at least insofar as those activities can be taken as guides to modal fact, surely do involve concepts in some sense. Take visualization, for instance. When I attempt to visualize a blue cat I am guided in that attempt by (among other things) my concepts of cat and blue. My understanding of what blueness is, my notion of blueness, is instrumental in my

15 attempts to form a mental image of a blue cat; that is to say, my concept of blueness is instrumental. Most importantly for my purposes in this paper, concepts seem to play a role in constraining what I can visualize. The nature of my concept of cat seems to be implicated in the fact that I cannot visualize a cat which is a plant, but can visualize a cat which is blue. The reason I cannot visualize a cat which is a plant is that my concepts will not allow it – which is to say that it is a conceptual falsehood. Whereas my concepts will allow me to visualize a blue cat, since it is not does not follow from my concepts that it is false that some cat is blue. Therefore I do not think the construal of conceivability I am proposing is too narrow to be the one in everyday philosophical circulation, as one might initially suppose on reading Gendler and Hawthorne‟s discussion.10, 11 Here I come to some important preliminaries to the provision of such an explanation. I have suggested that a proposition p is conceivable iff p‟s falsity does not follow epistemically from

10

I also think that caution is needed if we are to mention everything on Gendler and Hawthorne‟s list in connection

with conceiving, insofar as we want to pin down the notion of conceiving which has some apparent tie to modal knowledge. It is easy, for instance, to form a mental image of certain impossible things taking place: an image, say, of someone proving a very complex logical formula which is in fact logically false. We know what that eventuality looks like: she frowns, concentrates, scribbles away on her notepad, and finishes with a „QED‟ and a satisfied smile. If we weaken the notion of conceiving so far as to allow entertaining this image to count as conceiving of a logical impossibility being proved, there will be little prospect of explaining the epistemic link between conceivability and modal fact which we have set out to explain. 11

Nor, of course, is the thought original with me that there some sort of link between conceivability/imaginability

and conceptual truth. One recent proponent of such a view is Chalmers (see his 1996, §2.4).

16 our concepts. But what does this „following epistemically‟ amount to? The basic idea is that of a belief‟s being deliverable, in the right sort of way, as a result of conceptual examination. I shall say that a proposition p‟s truth follows epistemically from the nature of our concepts iff a correctly-conducted examination of certain of our concepts12 can suffice to ground a belief in p which either is knowledgeable13 or else would be knowledgeable if it weren‟t for the fact that there is something wrong with the concepts involved.

A true proposition which follows

epistemically from our concepts I call a conceptual truth. I shall also say that a proposition p‟s falsity follows epistemically from the nature of our concepts iff a correctly-conducted examination of certain of our concepts can suffice to ground a belief in p‟s falsity which either is knowledgeable or else would be knowledgeable if it weren‟t for the fact that there is something wrong with the concepts employed. A false proposition whose falsity follows epistemically from the nature of our concepts I call a conceptual falsehood. What does this talk of there being „something wrong‟ with a concept amount to? Well, one possibility is that the concept is not what I call fitting. A fitting concept either: A: succeeds in picking out – referring to – some real feature of the world (be it an object, a property,14 or any other kind of feature that the world has), or:

12

Typically these will be concepts appearing in p itself, though I want to allow that in some cases it might be

necessary to invoke further conceptual resources. 13

A belief that p is knowledgeable if it amounts to a piece of knowledge that p, or (for those who do not like to

assimilate pieces of knowledge with beliefs) is a belief that p for which the subject has sufficient warrant, so that she counts as knowing that p in virtue of having that belief with that warrant. 14

NB: I allow that a property may be a feature of the world even if it is uninstantiated, or even necessarily

uninstantiated.

17 B: is composed entirely of concepts which are fitting (in the sense in which one might say that the concept of bachelor is composed of the concepts man and unmarried). Another possibility is that the concept is fitting but misrepresents in some way. However, I wish to allow that a certain amount of misrepresentation may be harmless for epistemological purposes when it does not affect the items of knowledge we are interested in. I shall therefore introduce a notion of relevant accuracy (or sometimes, for convenience, accuracy) for concepts, saying that a concept is relevantly accurate iff it is fitting and does not misrepresent the world in any way that is relevant to our use of it in securing the conceivability result we are interested in. (Note that which concepts count as relevantly accurate may depend on which conceivability results we are interested in.) A third way for there to be something wrong with a concept, even when it is relevantly accurate, is that there may be something lucky or accidental about its being accurate. The kind of accident I have in mind here is analogous to the kind of accident which can sometimes prevent a true belief from being knowledgeable. (There will be two sub-cases here, corresponding to justified but non-knowledgeable true belief, such as occurs in Gettier cases, and unjustified true belief. More details may be found in Jenkins 2008, §2.5.) I shall therefore introduce another piece of terminology: I shall talk of epistemic groundedness for concepts. This is non-accidental15 relevant accuracy.16 I take an externalist 15

Note that different accounts of what sort of non-accidentalness is required for concept grounding can be

constructed to cohere with different accounts of what sort of non-accidentalness is required for knowledge. See Jenkins 2005b, p. 739. 16

I think this is an improvement on the definition I offered in Jenkins 2005b. One of the differences is that only

fittingness, not accuracy, is discussed in my 2005b definition.

18 view of what is required for non-accidentalness here; that is, I allow that a concept can be nonaccidentally accurate even when we are unaware of, and cannot reconstruct the kinds of factors which make it so. Finally, I shall say that concepts are empirically grounded just in case the process which makes them non-accidentally accurate involves some essential (i.e., epistemologically significant, rather than merely enabling) use of the senses. Grounded concepts can be thought of as a trustworthy map of the world: they represent the world as it really is, and it is no accident that they do so. Therefore, I think, we can recover information about the world by examining this map. Those pieces of information that we can so recover are the ones which I want to describe as „following epistemically‟ from the concepts in question. To help clarify my notion of a proposition‟s following epistemically from some concepts it may be helpful to run through an example of a proposition‟s falsity following from certain of our concepts, where one of those concepts is unfitting. What I‟m saying is that a proposition p‟s truth (falsity) follows epistemically from our concepts iff a correctly-conducted examination of those concepts can suffice to ground a belief in p‟s truth (falsity) which either is knowledgeable or else would be knowledgeable if the concepts involved were grounded. Suppose we‟re forced onto the second disjunct; suppose, for instance, that we are users of the concept tonk and are considering the proposition T: A tonk B and ~B. (See Prior 1960 for the introduction and elimination rules for the connective „tonk‟, and assume for the sake of exposition that „tonk‟ so defined expresses an unfitting concept. See Jenkins 2008, §5.6 for more on tonk.) T‟s falsity follows epistemically from our concepts17 because a correctly-conducted investigation of our

17

Although I suspect that T is not in fact false – see footnote 22 below. To say that its falsity follows from our

19 concepts can suffice to ground a belief in the falsity of conditional T which would be a knowledgeable belief if only tonk were a grounded concept. This counterfactual claim is a little tricky to evaluate, at least on the standard semantics for counterfactuals. Are there any possible worlds where its antecedent is true? Maybe, if there are possible worlds where the concept tonk has quite different features (a different elimination rule, for instance). If so, then at the closest such worlds where an examination of this concept tonk and the other concepts involved leads us to a belief in the falsity of A tonk B and ~B, that belief is knowledgeable. For our method of extracting information from our concepts is a trustworthy method: it delivers beliefs which are as secure epistemically as the concepts it operates upon. And we are to hold this fact fixed when assessing the counterfactual. (Note that it may be that the proposition we believe at these worlds is different from the proposition expressed at this world by „A tonk B and ~B‟, owing to the difference in the concept tonk. This does not matter; what the counterfactual is getting at is that any belief arrived at via a correctlyconducted examination of grounded concepts would be knowledgeable – whatever belief it was.) On the other hand, it might be held that any concept with different introduction and/or elimination rules would not be tonk, and that therefore the antecedent of the counterfactual is not possible. In this case, even if we are inclined to say that the counterfactual is trivially or vacuously true, we can still also hold that it is non-trivally interesting. For instance, we can argue that one reason why it is true is that our concept-examining process is a trustworthy way of recovering information from our concepts: it delivers beliefs which are as secure epistemically as the concepts it operates upon. That is to say, either it delivers knowledge or the concepts it‟s

concepts does not imply that it is false – rather, the concepts may be leading us astray.

20 operating on are ungrounded. The counterfactual we are assessing is simply mean to convey that this disjunction is reasonably stable across a range of other (possible or impossible) situations.

A claim I have previously defended (Jenkins 2005b; see also Jenkins 2008) is that if all the concepts involved in a conceptually true proposition p are grounded, then we can secure knowledge of p just by examining those concepts. That is what I mean by saying that grounded concepts are like a trustworthy map from which information can be read off. I argue that the reason examining our concepts can secure us knowledge of the mind-independent world is that grounded concepts are not merely things we happen to have inside our heads, but are nonaccidentally accurate representations of the mind-independent world.18 I have also suggested that if our concepts are empirically grounded, knowledge of a proposition p which is secured through conceptual examination is empirical knowledge: that is, knowledge which relies on an essential use of the senses. The role played by the senses is epistemologically significant: it is through experience that information gets encoded into our concepts, waiting to be recovered through conceptual examination. Hence experience is doing crucial work in putting us in touch with how things are in the mind-independent world, and not 18

One point which I do not bring out in Jenkins 2005b is that the way in which fitting concepts correspond to, or

represent, features of the world is crucial to the claim that we can use them as a guide to how the world is. Words in the English language represent features of the world, but that doesn‟t mean we can learn anything substantial about the world just by thinking about the words ‘vixen‟ and „female‟. By contrast, the fact that points on a map correspond to towns enables us to learn substantial facts about the towns in question. Concepts must represent in a way which is, in the relevant respects, map-like in order for my approach to work. Thanks to Katherine Hawley for pressing me to clarify this point.

21 merely enabling or preparing the mind so that it is ready to be put in touch with how things are with that world. Yet such knowledge it retains many characteristic features of the a priori: it is knowledge of conceptual truths through conceptual examination alone; no tests need to be conducted beyond the examination of one‟s concepts, so it can be secured in the armchair (by people whose concepts have previously been suitably grounded) rather than the laboratory; this sort of justification could only be removed through conceptual revision … and so on. It is therefore somewhat tempting to describe such knowledge as empirical and a priori, at least on many standard definitions of each of those terms (see Jenkins 2005b, pp. 742-3 and Jenkins 2008 §9.5). But not much hangs on how we decide to use the „a priori‟/‟a posteriori‟ terminology. It may be helpful to say a few words here about how concepts can be grounded in experience. In Jenkins 2005b I discuss the case of arithmetical knowledge; however, I think that the points made are generalizable. I shall give a sketch here of the argument of Jenkins 2005b, using arithmetical concepts as my example, but readers should note that the possible accounts of empirical grounding for arithmetical concepts that I discuss there are, equally, possible accounts of the empirical grounding of other concepts. In my 2005b, then, I argue that arithmetical knowledge may be secured through the examination of empirically grounded arithmetical concepts.

My suggestion there is that

arithmetical concepts may be sensitive to the way the world is in virtue of their being sensitive to sensory input – that is, the effect of the world on the brain through the normal functioning of our sensory apparatus. I discuss two possibilities: the camera-brain hypothesis and the filter-brain hypothesis. On the camera-brain hypothesis, we acquire arithmetical concepts in response to sensory input, and end up with concepts that accurately represent the world. Accurate concepts

22 are selected because they are the most useful to us in our attempts to make sense of the world around us. On this hypothesis, the process of concept formation is a bit like the process of taking a photograph. The incoming sensory input is like light hitting a piece of photosensitive film and shaping what appears there. The concepts we end up with are like a photograph of the relevant aspects of the world. On the filter-brain hypothesis, we are born with arithmetical concepts but they are nonetheless sensitive to sensory input (and hence the world), because as sensory input comes in our brains filter out any concepts that are not earning their keep by being useful to us. Again, it is hypothesized that the concepts we keep are useful because they are accurate representations of the world. So again, the concepts we end up with are sensitive to the way the world is, and experience plays a key role in making them so. I claim that if our arithmetical concepts are sensitive to the way the world is in this way then they are grounded in the sense defined above, and moreover empirically grounded, because I think that if something like one of the envisaged stories is true, then the role played by sensory input in mediating between the world and our concepts is epistemologically significant: experience is what makes our concepts non-accidentally accurate guides to the way the world is. However, even if I am right that concept examination can be an epistemic guide to certain non-modal truths about the world, it is quite another matter to say how it can be a guide to what is possibly or necessarily true about the world. What I shall do in the next section is investigate the prospects for arguing that modal knowledge can be secured through the trusting a certain kind of guidance from our empirically grounded concepts.

23 4 Concept Grounding and Modal Knowledge First, let me make a suggestion: that the formation of modal beliefs can be guided, in a knowledge-conducive way, by our grasp of the structure of the actual world. That is to say, a proper grasp of actual-world structure can epistemically underwrite (at least some of) our modal knowledge. The suggestion has some plausibility. For instance, there is clearly an intimate relationship between our knowledgeable belief that it is impossible for there to be a married bachelor and our understanding of the way the (actual-world) properties of unmarriedness and bachelorhood are structurally related (the relationship which we express – perhaps metaphorically – by saying that the property of bachelorhood includes that of unmarriedness). Clearly, if anything like this is right, the truth-value and the epistemic status of the modal beliefs we end up with when we allow ourselves to be guided by our grasp of actual-world structure in the envisaged manner is highly dependent upon whether the structural understanding we started out with was accurate.

If we are right in taking bachelorhood to include

unmarriedness, the resulting belief that bachelors are necessarily unmarried will be true. If we were mistaken in that understanding, it would be pure luck if we somehow ended up with a true modal belief nonetheless. Knowledge in such cases would seem to be out of the question. I will not argue further here for the suggestion that modal beliefs can be guided in this way by our understanding of actual-world structure, hoping that it will at least be found prima facie plausible enough to make its development and scrutiny worthwhile. Rather, my concern will be to investigate the use to which this suggestion can be put. Note, though, that the suggestion is meant to be metaphysically modest. I am not proposing that facts of metaphysical modality are, or obtain in virtue of, facts about the structure of the actual world (though for all I

24 say this might be true ), merely that one way of coming to believe – and to know – modal 19

propositions is by relying on one‟s understanding of structural relations between actual things (where the actual things include individuals, properties, relations, and whatever else there is in the world). How can this suggestion help us in the search for an account of the epistemic link between conceivability and modal truth? Well, building on the thought that conceivability is bound up with what our concepts are like (as suggested above), the basic idea will be that structural relations between relevantly accurate concepts are correlated with structural relations between the features of the world to which those concepts correspond. This means that being guided epistemically by the structural relations between our concepts – which I think is intimately bound up with the activity we describe as „attempting to conceive‟ of something – can lead to a correct understanding of the structural relations between those features of the world to which the concepts correspond. And if I am right in thinking that such understanding guides the formation of modal beliefs, then the correctness of that understanding – the accuracy of our concepts – can explain why the modal beliefs we end up with are true ones. But we are not just interested in true modal beliefs here, but also in modal knowledge. That is why we need to specify (among other things, such as the correctness of all the cognitive 19

One way of cashing out this idea would be reminiscent of the idea (see e.g. Armstrong 1983) that nomological

necessity is to be accounted for in terms of relations between universals. It could be suggested that metaphysical necessity is to be accounted for in terms of relations – perhaps relations of a different kind – between universals, or indeed, between other things: individuals, for instance. Another – perhaps compatible – way, suitable for combinatorial theorists of metaphysical modality, would be to suggest that structural relations between things in the actual world determine which recombinations there are.

25 processing involved) that the concepts we are working with are grounded, not merely accurate. I take it that relying on what a set of accidentally fitting concepts will allow us to conceive of is a process that might in some cases lead us to true modal beliefs, but that even when it does, these beliefs will not amount to knowledge. Whereas I think that relying (in the right way) on a set of grounded concepts will lead to modal knowledge. Because the concepts relied upon are nonaccidentally accurate representations of the world, the modal beliefs we end up with are nonaccidentally true.

So far things have been rather abstract; I shall therefore now work through an example in the hope of making clearer what I have in mind. I propose the following epistemology for the proposition Necessarily, all vixens are female. (Note that this is not an argument, appreciation of which enables us to know the target proposition, but a description of the process through which we know it.) 1: Our concepts of vixen and female (together with any other concepts that are relied upon in obtaining the inconceivability result in this case – e.g., perhaps, our concept of universal quantification) are empirically grounded. 2 (a consequence of 1): There are real features of the world corresponding to each of these concepts.20 20

Actually, this is a simplification. These concepts – or some of them – could in fact be mere compounds of

referring concepts, or compounds of such compounds, etc.. The simplification employed here just makes the argument easier to follow, and does not affect it substantially, for reasons which are explained Jenkins 2005b at pp. 733-4. Just as a compound of referring concepts can serve as an epistemic guide to truths about the bits of the world which correspond to its referring constituents (or constituents of constituents …), so when a concept which is a mere

26 3 (another consequence of 1): Any relations between these concepts (that are relevant to our assessment of whether to accept Necessarily, all vixens are female) accurately reflect relations obtaining between the parts of the world that the concepts represent. (For instance, the inclusion-like relation between our concepts vixen and female is an accurate reflection of the inclusion-like relation between the properties of vixenhood and femaleness. 4 (also from 1): And it is no accident that 3 is true. 5: When we try to conceive of All vixens are female being false, we find we cannot do it. 6: This is because the concepts mentioned in 1 are so structured as to prevent our conceiving of such a scenario. 7: We take the inconceivability result as an indication of the truth of Necessarily, all vixens are female. 8: Therefore what guides our modal belief is a non-accidentally accurate understanding of the structure of the independent world. 9: Therefore the modal belief we end up with is non-accidentally true.

Now, in order for this kind of epistemological story to be complete, we would need to go into the question of exactly why a non-accidentally accurate understanding of actual-world structure can serve a good epistemic guide to modal truth. Different metaphysical theories will differ on how

compound of referring concepts guides our conceptualizing that activity is based on an accurate understanding of the relations between the bits of the world to which those referring constituents (or …) correspond.

27 to continue the epistemological story beyond this point. I want to remain metaphysically neutral, for now, but to express optimism that for many (at least) of the metaphysical options currently available to mind-independence realists about modality, we will be able to find a matching epistemic story to plug in.21 Things are relatively straightforward for those who think that modal facts are, or metaphysically depend upon, facts about actual-world structure.

For such people, it is no

surprise that an accurate grasp of actual-world structure is an epistemologically adequate guide to the modal facts. Various (actual or possible) views about the metaphysics of modality might be construed as being of this type.

One that sits particularly well with my proposed

epistemology is that of Roy (1993, especially §2), who argues that the modal facts are determined by the actual structures of nonmodal properties. Alternatively, we might consider the view that the modal facts depend metaphysically upon facts about essences (as Fine 1994 believes), supposing also that essences can be cashed out non-modally. Or the view that modal facts are facts about recombination, where what recombinations there are depends upon actualworld structural relations between actual things. It is not necessary, however, to hold that modal facts metaphysically depend upon facts about actual-world structure in order to appeal to the proposed explanation of the relevance of conceivability to modal epistemology. All that is needed is some sort of epistemically significant connection between, on the one hand, certain elements of actual-world structure and the modal

21

I might also less modestly suggest that, in the absence of any other plausible realist but non-rationalist

explanation of why conceivability is a guide to possibility, the lack of a matching epistemic story is a worry for the metaphysical account which lacks it, rather than a problem for my epistemological proposal.

28 facts. One might, for instance, believe that the modal facts are metaphysically grounded in the existence of concrete possible worlds (like those of Lewis 1986), and that the existence of these worlds does not depend metaphysically on actual-world structure, yet believe that actual-world structure can give epistemic guidance as to which possible worlds exist. It might be, to give a simple example, that if the actual world is structured in such a way that the property of vixenhood includes that of femaleness, then there is no possible world at which there is a nonfemale vixen, although this fact about worlds is not metaphysically reducible, to or otherwise dependent upon, the corresponding structural fact about our own world. Why should this sort of epistemically favourable correlation obtain if there were no such metaphysical dependence of the possible upon actual-world structure?

Well, it could be

maintained that facts about actual-world structure are metaphysically dependent upon facts about what is possible, though this does not have much prima facie plausibility. But even if we prefer to believe in a two-way metaphysical independence, we might think that a correlation obtains because both facts are dependent upon some third sort of fact. Or it might be that there is a satisfying explanation of the correlation which does not mention metaphysical dependence at all. At any rate, these are options which should not be ruled out without further discussion, although it is not my aim to pursue them here. One issue that might be bothering the reader is that I have only accounted for knowledge of modal propositions involving fitting concepts. What about modal knowledge of propositions involving unfitting concepts? My answer is that I do not think there is any such thing, since I do not think unfitting concepts (concepts, that is, which do not represent anything and are not composed of concepts which represent things) can be used to express truths. They simply aren‟t

29 connected up to the world in the way that would be required in order for them to be so usable. Unfitting concepts are analogous to non-representing word-like constructions such as „squirk‟, which cannot be used to express truths of any kind. (There are true sentences which mention them, e.g. "„Squirk‟ does not mean anything in English", but none which use them.) Unfitting concepts may appear in a conceivable proposition p, and if that happens then p may seem to us to be possible. But p will not be true, and nor will the claim that p is possible be true.22 So our belief that p is possible will not amount to knowledge. Readers should not be concerned that this position forces any undesirable modal skepticism. For (I think) it is actually very hard to find examples of concepts that are unfitting, at least among the concepts we are actually inclined to use. Recall from footnote 14 above that a property concept may be fitting even though the property it represents is uninstantiated (or even necessarily uninstantiated). So concepts like perfect circle may well be fitting although there are no perfect circles. And recall that fittingness is preserved under concept composition. So concepts like mermaid may be fitting even if there is no single property of mermaidhood, provided that there are properties of womanhood, fishhood, tailhood, and so on corresponding to the constituent concepts out of which the concept mermaid is composed. Unfitting concepts will have to be much more drastically odd than perfect circle or mermaid. Something like tonk has a better shot of being unfitting – but we do not use the concept tonk or rely on it to guide our 22

I do not think these propositions will be false either (since I do not think their negations, which also involve

unfitting concepts, will be true). They may still be meaningful, I think (see Jenkins 2008, §2,2,5 for more on this issue). Notice that a consequence of this is that no conceptual truth or conceptual falsehood involves unfitting concepts (though of course, propositions involving such concepts can still be such that their truth or falsity follows epistemically from our concepts, in the sense defined above).

30 beliefs, modal or otherwise. For instance, although it follows epistemically from the concepts involved that If A then A tonk B and hence B, but we do not therefore believe that proposition. We don‟t trust the concept tonk. And rightly so – tonk is not connected up in the right sort of way to the world to be relied upon. It is not grounded, or, plausibly, even fitting.23 Readers should also not be concerned that admitting the possibility of conceivable propositions involving unfitting concepts places pressure on my claim that conceivability is „a guide‟ to possibility.24 What this reveals is one way in which it is not an infallible guide. For there might be conceivable propositions which are not possible. We run the risk, when we rely on conceivability as a guide to possibility, that the concepts we are relying on are unfitting or inaccurate and hence that we are being lead astray. However, if I am right in thinking that most or all of our concepts are fitting and largely accurate, and indeed non-accidentally so, then it is fine for us to rely on conceivability as a guide to possibility despite this risk. Most of the time, this reliance will yield knowledge. For those with internalist intuitions, it might seem that this possibility of error means that we cannot rely on any of our concepts until we have some assurance that they are fitting. I do not share the intuitions which suggest this. I think it is important that our concepts be nonaccidentally accurate, but not necessary that we should be (or be capable or becoming) aware of 23

The only way it might count as fitting is if it is a very inaccurate representation of some other logical feature of

the world. For instance, someone might say that tonk is really just a very inaccurate representation of what we normally use the concept or for, and that the inaccuracy shows up in the inappropriate elimination rule. I don‟t find this view tempting myself. For one thing, there seems no way to choose between it and the view that tonk is a very inaccurate representation of what we normally use the concept and for, and that the inaccuracy shows up in the inappropriate introduction rule. 24

I am grateful to an anonymous referee for pressing me on this point.

31 their having this status. For similar reasons, I do not think the fact that it is possible that I am a brain in a vat and have (and can have) no assurance that this is not so means that I cannot know by looking that there is a desk in front of me.25 Relatedly, let me note that the account I propose does not account for any knowledge we might have of modal relationships between alien fundamental properties (fundamental properties that do not exist in our world). I do not find this troubling, as I do not suppose that we have any such knowledge. Similarly, the account does not deal with knowledge secured by relying on concepts which are not empirically grounded. Concepts which are not grounded at all, I suggest, are not such that we can get knowledge by relying on them. But there might be non-empirical ways for concepts to be grounded. I have no suggestions as to what these might be, since I do not believe that we have epistemologically significant links with the world except through experience. But others who are more sympathetic to such links could in principle develop alternative, non-empiricist, versions of the proposal offered here by arguing that some concepts are grounded in a non-empirical way.

(Perhaps, for instance, they might hold that innate

concepts selected for by evolution are reliable in the right kind of way to count as grounded and enable us to secure knowledge by relying on them.)

25

More accurately, I am inclined to be a contextualist about these issues; that is, to say that what is conveyed saying

that someone „knows‟ that p depends on the context of utterance. A quick way of introducing context-neutral vocabulary is to say that in some contexts „knows‟ expresses „knows-high‟ and that in other contexts it expresses „knows-low‟, the idea being that failing to know-high is consistent with knowing-low. The point I am pressing here is that the fact that I can‟t rule out my being a brain in a vat doesn‟t mean it‟s impossible for me to know-low that there is a desk in front of me. I think, similarly, that our lacking assurance that our concepts are fitting doesn‟t mean that we can‟t secure knowledge-low of modal facts by examining them.

32

5 Concluding Remarks Notice that my proposal is in certain ways quite different from what is offered by a moderate rationalist such as Peacocke. Peacocke (e.g. 2000) proposes that a priori knowledge is possible because the possession conditions for certain concepts26 are tied to, or correlated with, the individuation conditions of the semantic values of those concepts. This claim is in some ways analogous to my claim that our concepts are accurate representations of the world. But Peacocke does not say anything why the needed correlation obtains. In Jenkins 2008 (§2.5), I argue that he does not rule out the possibility that it is accidental in a knowledge-destroying way. My aim is to explain why the accuracy of our concepts is in the relevant sense non-accidental by appealing to the impact of sensory input. That is why I, unlike a rationalist such as Peacocke, feel the need give sensory input a key role to play in the explanation of how we can get knowledge through conceptual examination. It may be that many parties to this debate would agree with me that sensory input is at least causally involved in the selection and/or maintenance of our concepts. What I am saying, in addition to this fairly uncontroversial thing, is that its involvement is epistemologically important. If the involvement of sensory input in the selection or maintenance of our concepts is what makes those concepts non-accidentally accurate guides to the structure of the world, then that sensory input is what enables us to know modal truths by examining those concepts.

I should like to finish up by returning to the insensitivity counterfactual I discussed at the end of §1: 26

Peacocke uses „concept‟ slightly differently from how I use it, but the differences do not matter here.

33 C: For any proposition P, even if the modal status of P were different, provided that P‟s truth-value was the same, our sensory input would also be the same. To claim that C is non-trivially true is, I suggested, to raise an interesting challenge for empiricists who believe in knowledge of mind-independent modal truths. What I want to do here is argue that if the foregoing proposal is adopted, then there is no reason to accept the insensitivity counterfactual as true except that it has an impossible antecedent. Thus I see no reason to allow that it is true in the non-trivial way required for the objection to have force. As suggested in §1, I think there are in fact non-trivial reasons to accept that for propositions P whose modal status we know (in the way described in this paper), if the modal status of P were different, then even if P‟s truth-value was the same, our sensory input would be different. And it would be different in such a way as to ground concepts with different properties – i.e. concepts which placed different constraints on what we could and couldn‟t conceive of, such that attempts at conceiving would lead to different modal conclusions. For example, if the contingent truth that there is a church outside my window were in fact necessary, then we would have different concepts of church, window etc.. And these concepts would be different in such a way as to constrain our powers of conceiving so as to make it inconceivable that there not be a church outside my window. Thus we would cease to believe that it is contingent whether there is a church outside my window.27

27

There is a tricky issue in this vicinity which I am skating over. I am assuming here that our concepts can be

slightly different with respect to what follows epistemically from them without being (numerically) different concepts. If this is not so, then the concepts involved in any proposition Q with different modal status from P would

34 Why suppose that our sensory input would go differently, in such a way as to have these knock-on effects on belief? Because to say that the relevant concepts are grounded, which is one of the assumptions I am working with, is to say that they are sensitive to the structure of the be different concepts from those involved in P, and hence Q would be a different proposition from P. On this view, it is very hard to evaluate counterfactuals which begin „If the modal status of P were different ...‟; these are comparable to counterfactuals which begin „If P were a different proposition from P ...‟. This need not be too worrying for me dialectically; the difficulty of making sense of these counterfactuals on this view puts pressure on the original counterfactual objection to empiricism too, and, although I have been happy to consider the insensitivity counterfactual seriously as a threat in this paper, I would be equally happy if it were urged by a proponent of the current objection that the best thing to do is simply to say that talking about these kinds of counterfactuals just isn‟t helpful for the purposes of understanding what sensitivity to experience amounts to, and get on with expounding modal empiricism in other ways (as I have done in the preceding sections). However, for current purposes, I am allowing that these counterfactuals are useful ways of conveying sensitivity and insensitivity claims. And I do think we might be able to get a grip on the tricky issue currently under consideration, by thinking about counterparts of propositions. Perhaps in order to evaluate „If the modal status of P were different ...‟ we should think about situations (or worlds) where there is a counterpart of P which has a different modal status from that of P. Suppose that the relevant counterpart Q involves the concepts Q-church and Q-window which are in significant ways similar to of our concepts of church and window – e.g. play similar roles in our representations of the world – but are such that, according to the (perhaps impossible) situation of evaluation, it is necessary that there is a Q-church outside my Q-window. Now consider the target counterfactual, „If the modal status of P were different, then even if P‟s truth-value was the same, our sensory input would have been different‟. This requires us to evaluate a situation in which Q, the counterpart of P, has the same truth-value as P actually does, and see whether, in that situation, our experience is different from how it actually is. If my account of modal knowledge is right, I think it will be. I think our concepts are sensitive, through experience, to how the world is. So I think that, in the relevant situation, our experience will be different in such a way as to ground different concepts (namely, those of Q-church and Q-window) instead of our actual concepts (church and window).

35 world. This sensitivity, I suggest, has two stages: our concepts‟ sensitivity to the structure of our sensory input, and our sensory input‟s sensitivity to the structure of the world. It is this twostage sensitivity that gives us a non-trivial reason to think that, were the structure of the world different in the way it would have to be were the modal truths different, we would respond by possessing concepts with different properties (in particular, concepts from which different propositions followed epistemically). Remember that I am also assuming that the structure of the actual world can be a guide to modal fact. This, I take it, means that (it is non-trivially true that) if the modal facts were different, the actual world would be differently structured. And because our concepts are sensitive to the structure of the actual world, our concepts would be different if we were presented with such a differently-structured world. This is so even if, as in the above example, P‟s truth-value remains unchanged, and even if we continue to know P by (for instance) seeing that it is true. The differences in our sensory input would not be of an everyday kind, comparable to the difference between its looking like there‟s a church outside the window and its not looking like there‟s a church outside the window. Rather, the differences might manifest in the structure of all sorts of daily sensory interactions with the world. It is very hard to say in much detail what our sensory input would be like if it grounded different concepts. One reason why this is so hard is that (I think) all the concepts I use are grounded ones, so I just do not have the right kinds of concepts to describe the envisaged scenario in any detail. The kinds of concepts that would describe this scenario well, and would provide us with good ways of describing the sensory input we would get, are ones I do not have available to express here and ones which you probably wouldn‟t be able to make sense of even if

36 I did manage to express them. So what becomes of the challenge to empiricism as developed in §1? Well, it becomes obvious that it is question-begging. It amounts to an assertion that our experience is not sensitive to the modal facts (because the insensitivity counterfactual is non-trivially true).

I say

experience is sensitive to the modal facts, and have made suggestions as to what that sensitivity is like and how it impacts upon our modal epistemology. I conclude that if the insensitivity counterfactual is true it is only trivially so. If the objector disagrees, he‟ll have to tell me what is wrong with the account I have proposed. It won‟t do to assert again that experience is not sensitive to the modal facts.

Finally, let me make three more remarks. Firstly, one might wonder whether the same accusation of question-begging could be used as a reply on behalf of the simplistic empiricist of §1 above to the objection that experience is not sensitive to the modal facts. The simplistic empiricist asserts that modal truths are empirically known qua parts of our best empirically supported theories. My impression is that the accusation of question-begging is also a fair one for the simplistic empiricist to make, and indeed that some of our modal knowledge may well be of the kind the simplistic empiricist describes. However, defending that claim would be a substantial task, one I shall not undertake here. For one thing, we would need to get clearer on the question of what kind of sensitivity to modal fact is envisaged on the simplistic empiricist‟s picture. For another, our project would be strongly analogous to that of defending some kind of indispensability argument about mathematical knowledge, and therefore we should need to engage with analogues of many relevant – and subtle – issues that have been discussed in connection with

37 mathematical indispensability arguments. For instance, one way of developing the objection that experience is not sensitive to modal fact, as against the simplistic empiricist, would be to run analogues of Sober‟s (1993) arguments that empirical confirmation accrues to the mathematical parts of our best theories only if there are other theories available that do not involve mathematics, and that mathematics cannot enjoy the benefits of appearing in well-confirmed theories because it is never blamed when it appears in disconfirmed theories. A defence of the simplistic empiricist strategy for modal knowledge would need to involve a response to the modal analogue of these concerns. My second remark is that, like any plausible epistemology, the one I‟ve been sketching does not attempt conclusively to disprove sceptical hypotheses.

A sound grasp of how

perception gives us knowledge of the physical world does not give us the resources to prove, to the satisfaction of the sceptic, that we inhabit a physical world which is roughly the way we think it is, as opposed to being wildly different or even non-existent. Similarly, my epistemology for modal facts does not give us the resources to prove to the sceptic that the modal facts are as we think they are, as opposed to being wildly different or even non-existent. Our modality-detecting mechanisms are far from infallible, on the envisaged account. There is room for error at various points, and hence there is room for sceptical arguments which trade on the worry that we cannot rule such error out in any non-question-begging way. Among the possible sources of error three are worth noting. Firstly, our sensory input could be misleading as to the structure of the world, resulting in our possessing unfitting or otherwise inaccurate concepts (which are nonetheless fully justified). Secondly, we might have responded to our sensory input incorrectly when forming or selecting concepts, resulting in our

38 possessing unjustified concepts (which, unless we are lucky, may also be inaccurate). Thirdly, we might have perfectly healthy, grounded concepts, but be mistaken when it comes to working out how these concepts relate to each other.28 However, as far as I can see, sceptical worries seem no more devastating for my proposal than their analogues in the perceptual case are devastating for accounts of perceptual knowledge which leave room for the possibility of error.

If my epistemological story places modal

knowledge on as secure an epistemological footing as perceptual knowledge of the physical world, that will be more than enough for me. Thirdly, to avert any potential misunderstanding, let me specify that I do not think that every truth we can learn through reflection on our concepts is such that we can also learn through reflection on our concepts that it is necessary. Propositions of the form Actually p iff p look like good counterexamples to that claim.29

28

This possibility corresponds to the „externalist‟ possibility that we might be mistaken as to what our concepts will

allow us to conceive of; see pp. 9-10 above. 29

Thanks to Daniel Nolan for encouraging me to clarify this.

39 References

Armstrong, D. 1983. What Is a Law of Nature? Cambridge University Press. Ayer, A. 1936. Language, Truth and Logic. London: Penguin. Bealer, G. 2000. „A Theory of the A Priori‟, in Pacific Philosophical Quarterly 81, pp. 1-30. ------------ 2002. „Modal Epistemology and the Rationality Renaissance‟, in Gendler and Hawthorne 2002, pp. 71-125. Blackburn, S. 1986. „Morals and Modals‟, in G. MacDonald and C Wright (ed.s) Fact, Science and Morality., Oxford University Press, pp. 119-41. Carnap, R.

1950.

„Empiricism, Semantics and Ontology‟, in Revue Internationale de

Philosophie 4, pp. 20-40, repr. in H. Feigl, W. Sellars and K. Lehrer (ed.s) New Readings in Philosophical Analysis, 1972, New York: Appleton-Century-Crofts, pp. 585-96. Chalmers, D. 1996. The Conscious Mind. Oxford University Press. ------------ 2002. „Does Conceivability Entail Possibility?‟, in Gendler and Hawthorne 2002, pp. 145-200. Chisholm, R. 1977. „The Truths of Reason‟, in his Theory of Knowledge (second edn.), repr. in Moser (ed.) A Priori Knowledge, 1987, Oxford University Press, pp.112-44. Colyvan, M. 2001. The Indispensability of Mathematics. Oxford University Press. Craig, E. 1985. „Arithmetic and Fact‟, in I. Hacking (ed.) Exercises in Analysis: Essays by Students of Casimir Lewy, Cambridge University Press, pp. 89-112. Descartes, R. 1641. Meditations on First Philosophy, ed. J. Cottingham, 1986, Cambridge University Press.

40 Fine, K. 1994. „Essence and Modality‟, in J. Tomberlin (ed.) Philosophical perspectives 8: Logic and language, Atascadero: Ridgeview, pp. 1-16. Gendler, T. and Hawthorne, J. 2002 (ed.s) Conceivability and Possibility. Oxford: Clarendon Press. O‟Leary-Hawthorne, J. 1996.

„The Epistemology of Possible Worlds: A Guided Tour‟, in

Philosophical Studies 84, pp. 183-202. Hume, D. 1739. A Treatise of Human Nature, ed. L. Selby-Bigge and P. Nidditch, Second edition, 1978, Oxford: Clarendon Press. Jenkins, C. 2005a. „Realism and Independence‟, in American Philosophical Quarterly 42, pp. 199-211. ------------ 2005b. „Knowledge of Arithmetic‟, in British Journal for the Philosophy of Science 56, pp. 727-47. ------------ 2008. Grounding Concepts: An Empirical Basis for Arithmetical Knowledge. Oxford University Press. Kant, I. 1781. Critique of Pure Reason. Edn. of 1929, trans. N. Kemp Smith, Basingstoke: Palgrave. Kripke, S. 1980. Naming and Necessity. Cambridge, Mass.: Harvard University Press. Lewis, D. 1973. Counterfactuals. Oxford: Blackwell. ------------ 1986. On The Plurality of Worlds. Oxford: Blackwell. Nolan, D. 1997. „Impossible Worlds: A Modest Approach‟, in Notre Dame Journal for Formal Logic 38, pp. 535-72. Peacocke, C. 2000. „Explaining the A Priori: The Programme of Moderate Rationalism‟,

41 in P. Boghossian and C. Peacocke (ed.s) New Essays on the A Priori, Oxford: Clarendon Press, pp. 255-85. Prior, A. 1960. „The Runabout Inference Ticket‟, in Analysis 21, pp. 38-9. Roy, T. 1993. „Worlds and Modality‟, in The Philosophical Review 102, pp. 335-361. Schofield, M. 1998. „Plato‟, in E. Craig (ed.) Routledge Encyclopaedia of Philosophy. London: Routledge. Retrieved July 30, 2006, from http://www.rep.routledge.com/article/A088. Sober, E. 1993. „Mathematics aand Indispensability‟, in The Philosophical Review 102, pp. 357. Whewell, W. 1840. Philosophy of the Inductive Sciences Founded Upon Their History. London: J.W. Parker and Son. Wright, C. 1980. Wittgenstein on the Foundations of Mathematics. London: Duckworth. ------------ 1986. „Inventing Logical Necessity‟, in J. Butterfield (ed.) Language, Mind and Logic. Cambridge University Press. Yablo, S. 1993. „Is conceivability a guide to possibility?‟, in Philosophy and Phenomenological Research 53, pp. 1-42.