By Anil Gomes
David Landy’s Kant’s Inferentialism: The Case Against Hume sets out to provide an inferentialist reading of Kant’s theory of mental representation and to show how Kant’s theory responds to and improves upon the theory of mental representation set out by Hume. We can think of an inferentialist theory of mental representation as one which takes the content of a mental representation to be connected in some essential way to the role that representation plays in a certain system of inferences. In Kant’s case, this reading takes him to hold that “concepts are rules for relating intuitions to one another by subsuming them under conditions that prepare them to be used in various syllogisms” (p. 2). It is the revolutionary nature of Kant’s inferentialism which Landy sees as grounding Kant’s case against Hume.
The idea that there is an inferential aspect to Kant’s account of mental representation can be traced back to the work of Wilfrid Sellars (1968) and it has become influential in contemporary debates through the work of Robert Brandom (1998, 2015). In many ways it is Sellars whose spirit animates and guides Kant’s Inferentialism, and Landy self-consciously situates himself in a tradition of scholarship which begins with Sellars, and to which we might add, along with Brandom, such Kant interpreters as Jay Rosenberg (2005) and James O’Shea (2012). But the aim of Landy’s book isn’t just to find a kernel of insight in Kant which can be developed into an inferentialist account of mental representation: it is rather to argue that Kant has a fully worked out inferentialist account of mental representation, to ground that reading in the relevant texts, and to show how that reading gives Kant the materials for his case against Hume.
This detailed sourcing of an inferentialist programme in Kant’s texts is one of the highlights of Landy’s thoughtful book, and I enjoyed looking anew at specific passages in Kant with Landy’s reading alongside them. In these comments, however, I want to abstract from some of those details to focus on the overall structure of Landy’s account of Kant’s case against Hume. This will involve some loss since I won’t get to discuss all of the careful, textual work which Landy uses to support his reading. But since that detailed textual work is used in support of a particular account of the overall structure of the First Critique and, in particular, its relation to a set of challenges raised by Hume, I hope that the slightly wider-scope format will bring out some of the more general, structural issues which Landy raises.
Here’s the structure of what follows. I’ll start by setting out Landy’s account of the case against Hume (Section 1). Then I’ll consider how we should understand the nature of the problem that Hume poses for Kant (Section 2). I’ll suggest that we need to distinguish two problems that Hume bequeaths to us: one concerning the possession of certain concepts, and one concerning our entitlement to use certain concepts. Landy takes Kant’s case against Hume to concern our possession of certain concepts, and in Section 3 I’ll raise some questions about this reading, focusing on the question of whether it can account for role that genealogy plays in Kant’s response to Hume. Finally, in Sections 4 and 5, I’ll turn to the issue of justification and consider whether the inferentialist reading can explain our entitlement to use the pure concepts of the understanding. I’ll suggest that the inferentialist ends up grounding the necessity of our using the categories in aspects of our own subjective disposition in much the same way as preformation systems of pure reason do, and that this falls short of establishing the objective necessity which would secure our entitlement to use the categories. In each case, these comments are intended not as criticism but as a spur for Landy to say more about the issues which frame his development of an inferentialist reading of Kant.
1. The Case Against Hume
What is the structure of the case against Hume, as Landy sees it? Landy takes Hume to endorse the following claims:
(NC) We have no idea that is an idea of a necessary connection.
(EW) We have no idea that is an a idea of the external world.
(SSE) We have no idea that is an idea of a single subject of experience persisting through time. (p. 19)
Understanding these claims would require us to understand the ideology of ideas, so I hope it doesn’t do too much violence to Landy’s understanding of the claims if we restate them as follows:
HR1. We cannot represent necessary connection.
HR2. We cannot represent the external world.
HR3. We cannot represent a single subject of experience persisting through times.
These are the claims which Hume endorses and which Kant will reject.
One immediate problem with assessing these claims is that it looks like it is a condition on understanding them that they are false. (What is it we cannot represent? This?) Landy suggests that this air of paradox arises only when Hume speaks in the voice of his philosophical predecessors, making claims which he recognises as “strictly speaking nonsensical” (p. 20). There’s a lot one might say about this, but for now let us accept this explanation and put the problem to one side. Instead, I want to consider how Landy understands Kant’s argument against these claims.
Landy suggests that Hume’s reasons for endorsing HR1-3 arise out of a particular theory of mental representation. Call it Hume’s Theory of Mental Representation (HTMR). In broad outlines, Landy presents it as a conjunction of claims. First, a copy principle holding that simple ideas are copies of simple impressions. Second, a representational copy principle holding that our simple ideas are of that of which they are copies (pp. 21–2). Finally, an extension of the representational copy principle to complex ideas, those ideas which are aggregates of simple ideas, which says that “complex ideas represent that which their simple parts represent as being arranged in the way that those simple parts are arranged” (p. 48). Hume uses this theory to show that we don’t have ideas of the items in question by arguing that we don’t have simple impressions of the items in question, and that we couldn’t arrange simple ideas in a way which would allow them to represent the items in question. Add in some assumptions about the connections between having an idea and a representation, and this gives us an argument for HR1–3 above.
How does Kant respond to Hume? Landy takes Kant to argue against HR1–3 by arguing that HTMR is false. More specifically, Landy takes Kant to hold that HTMR fails for two reasons. First, it cannot account for our capacity to represent complex states of affairs as complex. Second, it cannot account for the ‘unity of the proposition’, by which Landy means “the problem of how to combine independently meaningful expressions in a way such that the resulting combination is not merely an aggregate of expressions but is a judgment that makes a claim” (p. 7). Since Hume’s Theory of Mental Representation cannot account for these two features of representation, it must be rejected. In its place, Kant offers his theory of mental representation (KTMR), one which takes all representations of a complex state of affairs to be conceptually structured through the involvement of concepts, where concepts are to be understood as inferential rules that represent the necessary connections between the objects represented by intuitions. Spelling out the details of KTMR, and sourcing those details in Kant’s texts, is the main task of Kant’s Inferentialism.
One important aspect of Landy’s reading is that he takes Kant’s arguments against HTMR to have been completed by the time we reach the Transcendental Deduction of the Categories. The argument that HTMR cannot account for our capacity to represent complex states of affairs as complex takes place in the Transcendental Aesthetic and the argument that HTMR cannot account for the unity of the proposition takes place in the Metaphysical Deduction. This is important, because it allows Landy to adjudicate on the issue of whether or not the Transcendental Deduction begins with a premise which would have been accepted by Hume. Landy contends that it does not, because the starting point for the Transcendental Deduction is the claim that we are capable of representing a single subject of experience persisting through time, contra HR3. But this isn’t a problem, because Hume’s reasons for endorsing HR3—namely HTMR—have already been rejected.
If Landy is right, Kant’s arguments by the start of the Transcendental Deduction have shown us that HTMR must be rejected. And since Hume uses HTMR to support HR1–3, this shows that Hume’s reasons for endorsing HR1–3 are no good. But of course it doesn’t follow from the fact that Hume’s reasons for endorsing HR1–3 are no good, that HR1-3 are false: people can have bad reasons for believing true things. So if Kant wants to show that HR1–3 are false, he needs to do more than show that HTMR is false. We need reason to think, further, that HR1–3 are false.
So where does Kant argue for the falsity of Hume’s claims? Landy holds that the falsity of HR1 and HR2 are shown by the Transcendental Deduction, perhaps when combined with the Analogies. In broad terms, Landy sees Kant as arguing that the combination of Kant’s theory of mental representation together with the observation that we can represent ourselves as the identical subject of our representations, entails that we must possess the capacities to represent necessary connection and the external world. But the starting point of this argument is the claim that we can represent ourselves as a single subject of experience persisting through time. That is, the Transcendental Deduction starts with the negation of HR3. In this way, the argument against HR1 and HR2 depend on the assumption that HR3 is false.
Landy doesn’t think this is problematic. He writes, ‘if we take that theory of mental representation to have been refuted by the time we reach the Transcendental Deduction, Kant is by that time free to reject Hume’s conclusion that we can have no such idea [of a single subject of experience persisting through time]’ (p.111). In a sense that’s right, since once Kant has shown that Hume has no reason to think that HR3 is true, he’s free to start with the falsity of HR3 when constructing the argument of the Transcendental Deduction. Arguments have to start somewhere, after all. But showing that HTMR is false should not be conflated with showing that HR3 is false, and until Kant justifies this claim, the arguments against HR1 and HR2 appear rather precarious.
Now Landy does think that Kant provides an argument for HR3—it’s just that this comes in the Paralogisms where Kant argues that we must represent our mental representations as belonging to a single subject of experience which persists through time, a conclusion which entails the falsity of HR3. So the overall structure of the case against Hume looks as follows: Hume takes HR1–3 to follow from HTMR. Kant argues against HTMR in the Transcendental Aesthetic and the Metaphysical Deduction, and suggests there that KTMR is to be preferred. He then argues in the Transcendental Deduction that the negation of HR3 together with KTMR entails the negation of HR1 and HR2. And finally in the Paralogisms he argues for the negation of HR3.
I hope this is the right way to understand Landy’s account of Kant’s case against Hume. And though I shall go on to raise some questions about that account, let me raise one issue here which doesn’t rise to the level of a question, but is more the expression of puzzlement. The structure of this case against Hume looks slightly odd. Kant’s concern is to reject HR1–3. His argument against HR1 and HR2 start from the falsity of HR3. Yet the argument against HR3 doesn’t appear until after the argument against HR1 and HR2, and in a section of the book where one might have thought that Kant had his eyes on other targets. Why doesn’t Kant start the Deduction with his argument for the falsity of HR3, and move from that to the falsity of HR1 and HR2? As I say, I’m not sure this is a question for Landy. But his presentation of the case against Hume does make me wonder why, if it is right, Kant arranged his book in such a topsy-turvy manner.
2. Possession and Justification
Any formulation of a case against Hume needs to be clear about the nature of the charge around which the case will be built. Landy presents Hume’s problem as primarily one of possession: how is it that we possess a certain set of concepts, in particular those that would allow us to represent necessary connection, the external world, and a single subject of experience? Call this the Possession Problem. We can contrast the Possession Problem with the problem of justification which opens the Deduction: how is it that we are justified in using a certain set of concepts, in particular those involved in our judging about necessary connection, the external world, and a single subject of experience? Call this the Justification Problem. Landy recognises that the Transcendental Deduction aims to solve the Justification Problem: in this sense, the Possession Problem and the Justification Problem are both addressed in the First Critique. But he takes Kant’s case against Hume to concern primarily issues of possession.
Hume certainly raises an issue about our possession of certain concepts in the Treatise and the Enquiry. But he also raises concerns about our justified use of those concepts. And although it is somewhat artificial to separate the problems, in the Treatise—the text which Landy draws on most extensively in setting out his account of Hume’s theory of mental representations—it is the Possession Problem which seems to be in the foreground: “[W]e must consider the idea of causation”, Hume tells us, “and see from what origin it is derived” (THN 220.127.116.11). In contrast, the Enquiry tells us that “we must enquire how we arrive at the knowledge of cause and effect”, with Hume suggesting that “as a general proposition, which admits of no exception […] the knowledge of this relation is not, in any instance, attained by reasonings a priori” (EHU 4.5–6). And although this also raises an issue about our possession of the concept of causation analogous to that raised in the Treatise, it looks principally to raise a problem about our grounds for believing that there are any causes in the world, whether as a general claim about events, or in the context of any particular causal judgement. This problem challenges our right to use the concept of causation since it raises the possibility that, like the concepts of fate and fortune, the concept of causation circulates with universal indulgence but without a clear legal ground to its entitlement.
Kant’s framing of the problem posed by Hume in the Prolegomena appears to recognise that problems of possession and justification trace back to Hume. He writes:
Hume started mainly from a single but important concept in metaphysics, namely, that of the connection of cause and effect […] and called upon reason, which pretends to have generated this concept in her womb, to give him an account of by what right she thinks: that something could be so constituted that, if it is posited, something else necessarily must be thereby posited as well. (Prol, AA 4:257)
One issue raised here is that of accounting for the possession of our concept of cause: we need to understand “how the concept of such a connection could be introduced a priori“ (Prol, AA 4:257). But we are also asked by what right we think with this concept and “what [its] objective validity is based on” (Prol, AA 4:260; cf. A84/B116). This is a problem of explaining our entitlement to use the concept of cause.
How do the two problems relate to one another? There’s a sense in which a solution to the Justification Problem entails a solution to the Possession Problem, since if we can show that we are justified in using a certain concept, then it follows that we possess that concept. But the converse is not true: a solution to the Possession Problem does not entail that we are justified in using any concepts. And even if solving the Justification Problem entails that we possess certain concepts, showing that we possess a concept is not the same as showing how we do so. For this reason it is more useful to think of the Possession and Justification problems as distinct since an adequate response to the Possession Problem will require removal of the obstacles which seem to stand in the way of our possessing certain concepts, and those obstacles may be distinct from the ones which seem to stand in the way of our being entitled to use those concepts.
That Kant thought of these tasks as distinct is suggested by his distinction between the quid juris and the quid facti at the start of the Transcendental Deduction (A84/B116) and his comment in the Critique of Practical Reason that:
[t]he Critique has shown by that deduction [of the categories], first that [the categories] are not of empirical origin but have their seat and source a priori in the pure understanding, and second that, since they are referred to objects in general independently of intuition of these objects, they indeed bring about theoretical cognition only in application to empirical objects. (KpV, AA 5:141)
I take these two claimed accomplishments to correspond to responses to the Possession Problem and Justification Problem respectively: the Critique shows first that we possess such concepts, and then that we are justified in using them as (and only as) applied to empirical objects. And the phrasing of the sentence suggests that Kant took his responses to each problem to be distinct.
If this is right then, in addition to HR1–3 above, we should also take Kant to be arguing against the following Humean claims about justification, at least when understood as concerning assertoric acts of representing:
HJ1. We are not justified in representing that there is necessary connection.
HJ2. We are not justified in representing that there is an external world.
HJ3. We are not justified in representing that there is a single subject of experience persisting through times.
That is, the problem that Hume raises for Kant is not simply a problem about the possession of certain concepts, amongst them the concepts of necessary connection, external world, and a single subject of experience, but also one of our entitlement to use such concepts, and any case against Hume will have to address both problems.
Once we distinguish the Possession Problem from the Justification Problem, and correspondingly, once we distinguish HR1–3 from HJ1–3, we can raise two questions for Landy’s inferentialist reading. First, does it explain our possession of the concepts of necessary connection, external world, and single subject of experience persisting through times, and thus show that HR1–3 are false? And second, does it explain our justified use of these concepts, and thus show that HJ1–3 are false? Since Landy takes issues related to our concept of a single subject of experience persisting through times to arise in the Paralogisms, I’ll put HR3 and HJ3 to one side, and focus only on these questions as they arise for the concepts of necessary connection and external world. In the next section I’ll say something about the first question, and in the final section I’ll say something about the second.
3. Possession and Genealogy
Here is the broad outline of Landy’s response to the Possession Problem concerning the concepts of necessary connection and external world (pp. 175–6). Start from the fact that we can represent ourselves as the single subject of experience persisting through time. This is possible, according to Landy’s Kant, only if we can combine the manifold of intuitions in one consciousness. Combining the manifold of intuitions in one consciousness requires the use of object-concepts. An object-concept is one which represents the manifold of intuitions as being necessarily connected to one another. Object-concepts can represent these necessary connections precisely because concepts function as inferential rules: they represent necessary connections among the manifold of intuitions at the object-level by licensing certain kinds of inferences—material inferences—at the meta-level, where material inferences are understood as “inferences that are valid in virtue of something other than merely their logical form” (p. 184).
Here’s an example to illustrate the proposal. We have a series of intuitions: this green leaf; this brown trunk, this conker shell. To unite them in a single cognition we have to combine them under an object-concept, say that of horse-chestnut tree. In combining them under such a concept, we represent the distinct intuitions as being necessarily connected to one another. And the concept horse-chestnut tree represents these necessary connections through licensing certain material inferences: what it is to combine a manifold of intuition under the object-concept horse-chestnut tree just is for certain material inferences to be permitted at the meta-level. In this case, thinking of these intuitions as falling under the object-concept horse-chestnut tree permits me to infer from ‘I am looking at the front of a horse-chestnut tree’ to ‘if I move around the tree, I will see the back of a horse-chestnut tree’.
How does this help explain our possession of the concepts of necessary connection and external world? I think the idea is this: the kinds of material inferences which are permitted by the combination of the manifold of intuition under object-concepts are inferences which commit us to representing the objects of our representations as necessarily connected with one another and distinct from our perceptions of them. So the rules of material inference which underlie our subsumption of intuitions under object-concepts—a subsumption which is necessary if we are to represent ourselves as the single subject of experience persisting through times—require us to represent necessary connections and a world independent of our perception of it, contrary to HR1 and HR2 above. This comprises a response to the Possession Problem.
How effective is this response? One issue concerns the role that genealogy plays in Kant’s response to Hume. Hume’s reasons for thinking that we cannot represent necessary connection or external world seems to turn on concerns about the origin of such capacities: it is because we cannot find an origin for the concept of cause that Hume denies our possession of it. As he puts it in the Treatise, “[t]here is no impression convey’d by our senses, which can give rise to that idea” (THN 1.3.14). This is a concern about the genealogy of our concepts of necessary connection and external world and it is a failure to provide an appropriate genealogy which damns our claim to possess the concepts.
Landy’s inferentialist response to this problem doesn’t seem to accord genealogy a central role in providing an account of our capacity to represent necessary connection and external world. That is, it doesn’t respond by accepting that we ought to provide an appropriate genealogy and finding one elsewhere—perhaps by tracing our concept of cause back to some non-impression like origins. Rather, on the inferentialist story, there is nothing more to representing necessary connection than our being able to make certain material inferences. And once we see that our making such inferences is a condition on our representing ourselves as a single subject of experiences persisting through time, then we have all we need to answer Hume’s scepticism about our possession of these concepts. A genealogy of our concept of cause doesn’t play a central role in this story.
But how does this fit with Kant’s use of genealogical metaphors in setting out his response to Hume’s Possession Problem? Kant often describes the nature of the Metaphysical Deduction in genealogical terms in a way which suggests that he takes the Metaphysical Deduction to find an origin for the categories which is not in the messy world of sense-impressions: “In the metaphysical deduction the origin of the a priori categories in general was established…” (B159); the table of categories is an “ancestral registry of the understanding” (A81/B107). And in the Second Critique Kant states that the first task of the Deduction has been to show that “[the categories] are not of empirical origin but have their seat and source a priori in the pure understanding” (KpV, AA 5:141).
That we explain our possession of the categories by attending to their origin in the pure understanding is further suggested by Kant’s famous response to Eberhard:
The Critique admits absolutely no implanted or innate representations. One and all, whether they belong to intuition or to concepts of the understanding, it considers them as acquired. But there is also an original acquisition (as the teachers of natural right call it), and this of that which did not yet exist at all, and so did not belong to anything prior to this act. (ÜE, AA 8:221)
We have reference here to an act of original acquisition, one which works through “reflection on the occasion of experience” (V-Met-L1/Pölitz, AA 28:233). This looks like an attempt to explain our possession of the categories through an alternative genealogy, one which doesn’t bottom out in sense-impressions as demanded by Hume. So issues about genealogy look to be central to Kant’s response to the Possession Problem.
Can the inferentialist reading account for this genealogical terminology? It remains true, of course, on the inferentialist reading that there is a link between the categories and the table of judgements—the pure concepts are just those whose form intuitions must take if they are to play a role in certain judgements (pp. 116–17)—so to the extent that the table of judgements has its origin in the understanding, then the categories, even understood as inferential rules, will also have an origin of some kind in the understanding. But I wonder whether this suffices to explain Kant’s repeated use of genealogical imagery, and whether the inferentialist reading can make sense of the suggestion given in the text that it is the provision of a non-Humean acquisition story which explains our possession of the categories—and thus the falsity of HR1 and HR2.
4. Justification and the Deduction
What about the Justification Problem? Landy holds that Kant’s strategy for answering the Justification Problem is “to demonstrate this justification by showing that this activity [of applying pure a priori concepts] is an essential part of another activity, which is itself justified” (p. 110). That other activity is the activity of “conceiving some representations as belonging to oneself” (p. 110). In outline, the structure of the argument is as follows: in order to conceive of some representations as belonging to ourselves we must combine the manifold of intuitions so as to represent an object; representing an object requires representing it as a complex of its parts; representing objects as complexes of their parts is only possible through the application of a concept to an intuition; this application of a concept to an intuition is a judgement; the categories are the logical functions of judgements and thus employed in every judgement; so if we are to conceive of some representations as belonging to ourselves, it is necessary that we employ the categories (from pp. 118–20).
This interpretation takes the Deduction to argue for the following conditional: Conceiving of some representations as belonging to oneself requires that one employ the categories. Landy takes it that we are justified in conceiving of some representations as belonging to ourselves. So, on the grounds that if we are justified in acting a certain way, then we are justified in doing anything which is a necessary condition on acting in that way, Landy takes the Deduction to show that we are justified in employing the categories in judgement. But merely employing the categories in a judgement is distinct from correctly applying the categories in a judgement: I employ the concept of cause when I judge that Umrao caused the fire, even if my judgement is mistaken. And from the fact that we are justified in employing the categories, it does not follow that the applications we thus make are thereby correct. Employment of a concept is compatible with that ascription being mistaken.
The distinction between mere and correct employment has been raised before as an issue for interpretations of the Transcendental Deduction. But it presumes that there is a distinction to be drawn between our necessarily employing the categories and the categories correctly applying. And it may be that this distinction disappears in the inferentialist framework. I’ll turn to this issue in a moment. For now, it is worth considering whether an argument which supported the conclusion only that we must employ the categories could answer Hume’s problem about justification.
Let us start by granting Landy’s argument concerning the possession of certain concepts. In the context of the Deduction, this means granting that the inferentialist reading of the Deduction shows that we possess certain categorial concepts, amongst them the concept of a necessary connection and those used in our thinking about the external world. So understood, the inferentialist reading has shown the falsity of HR1 and HR2 above. What about the issue of justification? Does the inferentialist reading show the falsity of HJ1 and HJ2?
Here are the reasons for thinking not. The conclusion of Landy’s reconstructed argument is that we must employ the categories (p. 120). And Landy thinks that if this is a condition on something which we are justified in doing—namely, taking some representations to belong to ourselves—then we are justified in employing the categories. But from the fact that we must employ the categories, it doesn’t follow that we must correctly employ the categories. So Landy’s argument cannot show that we are justified in correctly employing the categories. And it is this latter claim which needs defence if there is to be a genuine response to Hume’s problem of justification.
This is a particularly pressing concern in the context of building a case against Hume because, absent the concerns about our possession of a priori concepts, it is not obvious why a Humean should object to the conclusion that we must necessarily employ the categories in judgement if it doesn’t follow from that conclusion that we must sometimes employ them accurately. Consider that strand in Hume’s thinking which accords Custom and Habit a special role in supporting chains of reasoning which cannot be supported by intuition or demonstration (EHU 5.1). It would be open for a contemporary Humean to accept Landy’s argument but to hold that this shows only that the way in which we are compelled to think about the world necessarily requires us to make use of notions of necessary connection and external world. That would show that Hume was wrong to think that we lack such concepts—which is to say, Landy’s response to the Possession Problem would still remain—but it wouldn’t show that Hume was wrong to challenge our grounds for thinking that anything in the world corresponds to those notions. And without that stronger conclusion, HJ1 and HJ2 remain standing.
To belabour the point: Landy’s account of the Deduction takes the conclusion of the argument to be that we must necessarily employ the categories in our thought and perception of the world. He takes this to suffice for showing that we are justified in using the categories. But it doesn’t follow from the fact that we are necessarily required to think and perceive in accordance with certain concepts that those concepts pick out genuine features of the world. And if we can show only that we are required to think and perceive this way, and not that our thinking and perceiving this way is, at times, accurate, then we have fallen short of legitimising our use of these contested concepts. Landy’s inferentialist reading leaves us without a response to Hume’s worries about justification.
All of this is prefaced on the assumption that there is a distinction to be drawn between our necessarily employing the categories and our correctly employing the categories. And it is this assumption that we now need to examine. For it may be that this distinction disappears on the inferentialist reading—in which case, to show that we must employ the categories is ipso facto to show that we can in principle be correct in any such employment. And this would seem to suffice for showing that we are justified in using the categories, contrary to HJ1 and HJ2. So if the inferentialist reading can deny the distinction between our necessarily employing the categories and our correctly applying the categories—or better, deny that it could be the case that one must employ the categories without at least some of those applications being correct—then the above objection would be without merit.
Here is a reason to think this is possible. On the inferentialist reading, to say that something is an object is not to pick out some special sort of entity which has a particular nature. It is rather to say that something falls under an object-concept, where falling under an object-concept is to be understood in terms of the rule of inference which are permitted by that concept. The categories are just those “meta-conceptual rules that determine what counts as a representation of an object” (p. 252). So if we are to understand what is involved in employing the categories we have to remember that their employment is to be understood in terms of the inferences they licence.
Say that we must employ the categories to the manifold of intuition. Then we must subsume the manifold of intuitions under object-concepts which permit us to make certain inferences, in particular those which involve positing the objects of intuition as necessarily connected and distinct from our perceptions of them. Could our application of such an object-concept be mistaken, in such a way that the inferences permitted by the categories turn out to be mistaken? Landy’s most extensive discussion of this comes in his account of transcendental idealism. There are two ways, he says, to interpret the question of “whether the properties we attribute to the ‘objects’ of our representation are ‘really’ properties of those things” (p. 280). Either the question is an empirical one, in which case it is to be “answered by applying the conceptual and intuitional norms that govern our use of our current conceptual scheme”. Or it is a question about the adequacy of that scheme itself, in which case it is answered by “comparing the adequacy of our current conceptual scheme with that of a proposed successor scheme” (p. 280).
Here’s a suggestion of how to put this in inferentialist terms. To represent some object o as being F is to engage in certain inferences concerning o. The question of whether or not o is F is to be understood as a question about whether those inferences are the right kinds of inferences to be engaged in. Empirical considerations might tell against those inferences—perhaps some of the licensed inferences turn out to be incorrect—or a new conceptual scheme might licence improved inferences. But, either way, there is no way to judge the question of whether or not o is F without going back to the kinds of inferences which are licensed by the concept of being F and assessing whether those inferences are the right kind of inferences to make.
In the case of the categories, this means that if we are to decide whether there is a gap between showing that we must necessarily employ the categories and the categories thereby applying, then we have to look at the kinds of inferences which are permitted by object-concepts. There are no empirical considerations which could tell against the inferences which are structured by the categories, since those inferences set the framework within which empirical considerations can be brought to bear. But nor could there be a successor scheme which improved on our conceptual scheme without requiring the employment of the categories, since the necessity of our employing the categories entails that they must be used in any conceptual scheme. (I take this to be guaranteed by the connection to the table of judgements.) So there is no gap of the sort assumed by the above objection: if we are justified in employing the categories, then we are justified in principle of correctly employing the categories. Hume’s problem about justification is answered.
There’s something very neat about this proposal. Transcendental idealism is often invoked as a way of bridging the gap between it being the case that we must apply the categories and the conclusion that the categories must apply, but the most natural invocations of this strategy involve idealist commitments which seem in tension with Kant’s professed empirical realism. Landy’s representational form of transcendental idealism—which he glosses as the claim that “that which we represent using object-concepts is necessarily subject to the forms that such representations will necessarily take for creatures like us” (p. 279)—bridges the gap without any objectionable metaphysical penalty. This is a benefit of the proposal.
5. Subjective and Objective Necessity
But I have a niggling worry. At the very end of the B-edition Transcendental Deduction, immediately after he has completed his proof of the objective reality of the categories, Kant raises an objection to preformation systems of pure reason. Preformation systems of pure reason are those on which
the categories were neither self-thought a priori first principles of our cognition nor drawn from experience, but were rather subjective predispositions for thinking, implanted in us along with our existence by our author in such a way that their use would agree exactly with the laws of nature along which experience runs. (B167)
And Kant complains of such views that
in such a case the categories would lack the necessity that is essential to their concept […] I would not be able to say that the effect is combined with the cause in the object (i.e., necessarily), but only that I am so constituted that I cannot think of this representation otherwise than as so connected; which is precisely what the skeptic wishes most. (B167–8)
It seems likely that the immediate impetus for this addition to the Deduction was Johann Schultz’s 1785 review of J.A.H. Ulrich’s Institutiones Logicae et Metaphysicae. Kant introduces preformation systems of pure reason as entailing a pre-established harmony between pure concepts and appearances, and it is in this review that Schultz—later described by Kant as his best interpreter (Br, AA 12:367–8)—raises the possibility of pre-established harmony as a way of explaining our right to use the categories when thinking about the world (see Sassen 2000:214). Kant references Schultz’s review in the Metaphysical Foundations of Natural Science, promising to revise the Deduction to
make up for this deficiency […] so that the perceptive reviewer may not be left with the necessity, certainly unwelcome even to himself, of taking refuge in a preestablished harmony to explain the surprising agreement of appearances with the laws of the understanding, despite their having entirely different sources from the former. (MAN, AA 4:476n)
The footnote in the Metaphysical Foundations of Natural Science may confirm Schultz as the immediate impetus for the addition of this objection, but he doesn’t exhaust its target. One target is Crusius who, Kant tells us in the Prolegomena, held “that a spirit who can neither err nor deceive originally implanted these natural laws in us” (Prol, AA 4:320). And another, as unlikely as it may seem, is Hume. Schultz’s appeal to pre-established harmony comes, after all, in the context of his formulation of the problem that he thought Hume posed to Kant (Sassen 2000:214). And Hume himself writes in the Enquiry that his account of Custom and Habit is
a kind of pre-established harmony between the course of nature and the succession of our ideas; and though the powers and forces, by which the former is governed, be wholly unknown to us; yet our thoughts and conceptions have still, we find, gone on in the same train with the other works of nature. Custom is that principle, by which this correspondence has been effected. (EHU 5.2.21)
Hume’s status as a target is made explicit in the Metaphysical Foundations of Natural Science footnote. Pre-established harmony, Kant writes, makes things “only subjectively necessary, but objectively merely contingent, placing together, precisely as Hume has it when he calls this mere illusion from custom” (MAN, AA 4:476n). This charge also appears in the Prolegomena where Kant claims that Hume “passes off […] subjective necessity (i.e. habit) for an objective necessity (from insight)” (Prol, AA 4:257–8). Hume’s view is a form of pre-established harmony of the pure concepts and appearances, except with custom characteristically replacing a divine being as the source of agreement. This means that the argument against preformation systems of pure reason must be understood as part and parcel of Kant’s case against Hume.
What is the problem with preformation systems of pure reason? In each of these works—§27 of the Deduction, the footnote in the Metaphysical Foundations of Natural Science, and the Prolegomena—Kant makes the same point: that preformation systems establish only a mere “subjective necessity, arbitrarily implanted in us” (B168), whereas what is wanted is “the objective necessity that characterizes the pure concepts of the understanding” (MAN, AA 4:476n). One way to put the charge is that preformation systems make the forms of thinking independent of the laws of nature, even though both necessarily coincide. This is not a worry about the contingency of the agreement between our application of the categories and their applicability, as is sometimes suggested (e.g. Hanna 2006:309; Land 2015:37). On preformation views, it is genuinely necessary that we employ the categories, but the necessity of our doing so is ultimately grounded in aspects of our own subjective dispositions and not on the relation between the categories and the objects of experience. And if we cannot show more than this merely subjective necessity, then we cannot justify our use of the categories: Hume’s scepticism about justification is sustained.
What is the relevance of Kant’s criticisms of preformation systems of pure reason for the inferentialist reading of the Deduction? To recap, the initial objection to the inferentialist reading was that it could show only that we were justified in employing the categories, not that we were justified in correctly employing the categories. And it is the latter which is needed, I claimed, for Kant to have a case against the problem of justification posed by Hume. I noted that this worry could be avoided because the distinction between our necessarily employing the categories and our correctly applying the categories turns out, on the inferentialist account, to be a distinction without a difference: as meta-conceptual rules of inference, the categories structure what it is for something to be an object such that no sense can be attached to the question of whether objects really are as we necessarily represent them as being.
But the inferentialist account is able to make this move only by grounding the correspondence between our employment of the categories and their applicability—between, that is, the forms of thinking and the laws of nature—in the kinds of inferences which we are so constituted as to make. And this looks like the kind of grounding base which Kant warns us against §27 of the Deduction. Note that it doesn’t matter here that we are necessarily constituted so as to make these inferences: the worry about preformation systems isn’t about their contingency, but about their account of the ultimate ground for the agreement between the employment of the categories and their correct applicability. The inferentialist grounds this agreement in a certain aspect of our subjective constitution: namely, in the kinds of inferences which we are constrained to make. But this allows us to say “only that I am so constituted that I cannot think of this representation otherwise than as so connected” (B168). And that falls short of establishing the objective necessity which Kant thought to be required by the Deduction.
To conclude, I’ve suggested that Hume raises two problems for Kant: one about the possession of certain concepts, and one about the justification of certain concepts. It is the task of the Deduction to answer the second problem, and an adequate solution must end by showing not only that we are justified in employing the categories, but that we have grounds for thinking that some of the judgements we make using the categories are actually true. The inferentialist can bridge this gap and reach the stronger conclusion only by minimising the difference between our necessarily applying the categories and the correctness of those necessary applications. But if this is achieved by grounding the agreement between the forms of thinking and the laws of nature in a certain aspect of our subjective disposition—namely, our capacity to engage in certain kinds of inferences—then the end result looks to secure no more than a subjective necessity of the kind that Kant warned us against. If the inferentialist reading cannot do more than this, the case against Hume remains unproved.
Acknowledgements: Thanks, as usual, to Andrew Stephenson and Colin McLear for their very helpful comments.
Invited: 27 October 2015; received: 21 October 2016.
 Citations of Hume’s works are to the Clarendon Edition of the Works of David Hume. References to A Treatise of Human Nature (THU) are by book, part, section, and paragraph; references to the An Enquiry concerning Human Understanding (EHU) are by section and paragraph.↩
 See Cassam (2007) for an account of the structure of ‘how possible’ questions which emphasises the obstacle-dependent nature of such questions.↩
 See Gomes (2010; 2014), Van Cleve (1999) and the classic discussions in Cassam (1987), Rorty (1970) and Stroud (1968).↩
 In Gomes (forthcoming) I argue that the same problem befalls Lucy Allais’s nonconceptualist reading of the Transcendental Deduction.↩
Brandom, R. (1998), Making It Explicit: Reasoning, Representing, and Discursive Commitment (Cambridge, MA: Harvard University Press).
——— (2015), From Empiricism to Expressivism (Cambridge, MA: Harvard University Press).
Cassam, Q. (1987), ‘Transcendental Arguments, Transcendental Synthesis and Transcendental Idealism, Philosophical Quarterly 37 (149): 355–78.
——— (2007), The Possibility of Knowledge (Oxford: Oxford University Press).
Gomes, A. (2010), ‘Is Kant’s Transcendental Deduction of the Categories Fit for Purpose?, Kantian Review 15 (2): 118–37.
——— (2014), ‘Kant on Perception: Naïve Realism, Non-Conceptualism, and the B-Deduction’, Philosophical Quarterly 64 (254): 1–19.
——— (forthcoming), ‘Nonconceptualism, Hume’s Problem, and the Deduction’, Philosophical Studies.
Hanna, R. (2006), Kant, Science, and Human Nature (Oxford: Oxford University Press).
Land, T. (2015), ‘Nonconceptualist Readings of Kant and the Transcendental Deduction’, Kantian Review 20 (1): 25–51.
O’Shea, J. (2012), Kant’s Critique of Pure Reason: An Introduction (London: Acumen).
Rorty, R. (1970), ‘Strawson’s Objectivity Argument’, Review of Metaphysics 24: 207–44.
Rosenberg, J. F. (2005), Accessing Kant: A Relaxed Introduction to the Critique of Pure Reason (New York: Oxford University Press).
Sassen, B. (2000), Kant’s Early Critics: the Empiricist Critique of the Theoretical Philosophy (Cambridge: Cambridge University Press).
Sellars, W. (1968), Science and Metaphysics: Variations on Kantian themes (London: Routledge & Kegan Paul).
Stroud, B. (1968), ‘Transcendental Arguments’, Journal of Philosophy 65 (9): 241–56.
Van Cleve, J. (1999), Problems from Kant (New York: Oxford University Press).
© Anil Gomes, 2017.
Anil Gomes is Fellow and Tutor in Philosophy at Trinity College, Oxford and a CUF Lecturer (Associate Professor) in Philosophy in the Faculty of Philosophy in the University of Oxford. His main research interests are in the philosophy of mind and Kant’s theoretical philosophy and, in particular, on issues which arise at their intersection. He has published widely in, among other venues, Mind, Philosophical Studies, Kantian Review, Kant-Studien, Philosophical Quarterly, British Journal of Aesthetics and Erkenntnis. The edited volume Kant and the Philosophy of Mind, co-edited with Andrew Stephenson, is forthcoming from Oxford University Press.