site hit counter

 James Kreines

Department of Philosophy,
Claremont McKenna College

850 Columbia Ave
Claremont, CA 91711
Office: Roberts North 211
(909) 607-6845



Kant on the Laws of Nature:
Laws, Necessitation, and the Limitation of Our Knowledge[1]

Please cite the published version, European Journal of Philosophy 17 (4):527-558. What follows is the final draft:


Consider the laws of nature—the laws of physics, for example. One familiar philosophical question about laws is this: what is it to be a law of nature? More specifically, is a law of nature a regularity, or a generalization stating a regularity? Or is it something else? Another philosophical question is: how, and to what extent, can we have knowledge of the laws of nature? I am interested here in Kant’s answers to these questions, and their place within his broader theoretical philosophy during the period spanning from the first to the third Critique.

A simple approach is suggested by two claims widely associated with Kant. First, Kant appears to hold that a law of nature is not a contingent regularity, but is distinguished by a kind of necessity. For example, consider this discussion of causality and a causal law:

The concept of cause … requires that something A be of such a kind that something else B follows from it necessarily and in accordance with an absolutely universal rule.

Second, Kant famously holds that we can only have knowledge of necessity where we can have a priori knowledge. And so it is no surprise that this passage seems to add that we cannot have empirical knowledge of such laws of nature, insofar as they involve necessity:

Appearances may well offer cases from which a rule is possible in accordance with which something usually happens, but never a rule in accordance with which the succession is necessary… The strict universality of the rule is therefore not any property of empirical rules… (A91/B124)

The passage requires further interpretation, to be sure, but it does allow formulation of one possible and simple interpretive approach, along these lines: Kant holds that all laws of nature are distinguished by a kind of necessity; and so Kant denies the possibility of purely empirical knowledge of laws.

Many recent interpreters argue that Kant in fact proposes a very different account. The basic idea of the recently popular ‘best system’ interpretations (as I will call them) is this: Kant certainly holds that some laws of nature in general, such as the law that every alteration must have a cause, are supposed to require a priori defense. But Kant distinguishes such laws of nature in general from the particular laws of nature, or the laws of interest in the different branches of the natural sciences—for example, the laws of physics. The particular laws of nature are, borrowing Phillip Kitcher’s formulation, a subset of the true generalizations: ‘for Kant, the laws of nature are just the generalizations that would figure in the best unifying system in the limit of rational inquiry’ (1996: 412). Any one of these generalizations, ‘taken individually’ is ‘empirical and contingent’.[2] And there is no reason to associate the particular laws with a priori knowledge. There is a sense in which particular laws involve ‘necessity’, but only in a sense wholly derived from or constituted by their place in the best system of empirical knowledge.[3]

I reject best system interpretations (section 1 below). In this respect I follow Michael Freidman, from whom I borrow my way of framing the issue with the initial citation above (1992a, 161-2). But I defend an interpretation that contrasts with Friedman’s as well. Kant does not, I argue, account for what it is to be a law of nature in terms of an account of our knowledge of laws—neither in terms of an ideal limit of empirical inquiry (as in best system interpretations), nor in terms of any kind of derivation or deduction from a priori principles (as in Friedman’s alternative). Kant’s view rather rests on the intuition that an explanation must provide information about an underlying condition on which an explanandum depends. To state a particular law of nature, then, is not just to summarize or describe a regularity in nature; it is to identify a kind on whose nature some regularity depends, in the sense that it is necessitated by the nature of that kind. I call this Kant’s ‘necessitation account’, and I call these laws ‘necessitation-laws’ (section 2).

But it is crucial that I am not arguing here that Kant is any less concerned than we would expect with the epistemological restrictions commonly taken as central to his break with pre-critical metaphysics: Kant does not allow empirical knowledge of natural necessity; nor does he think that we can somehow deduce purely a priori the content of all the particular laws of physics, chemistry, etc. On the contrary, the overlooked key to the consistent position presented throughout the critical-period texts on laws is the way in which Kant’s specific epistemological commitments, concerning our dependence on sensible intuition, sharply restrict our knowledge of laws. To begin with, the texts emphasized by best system interpreters in fact defend this view: the natural sciences seek in empirical inquiry particular necessitation-laws; we have reason to take there to be such laws for the purposes of inquiry, and empirical inquiry can improve in approximation to knowledge of them; but we cannot generally achieve knowledge of particular laws (section 3).

Friedman emphasizes that Kant does attempt to derive knowledge of particular laws in the specific case of the laws of motion, by applying a priori principles of the understanding to the empirical concept of matter. But regardless of the prospects for this attempt, Kant here too—contra Friedman—argues that there is an ineliminable limit to our knowledge. For attempting such a derivation makes sense only in a special kind of case. In general, we do not seek to reduce all scientific knowledge to knowledge of a single basic kind, like matter; insofar as we seek knowledge of the diversity of nature, we generally seek knowledge of particular laws governing interactions between specifically distinct kinds. And Kant is right to conclude, given his epistemological commitments, that we can never hope to derive knowledge of this sort of law, in any way, from any a priori principles of our understanding. In the main, then, we can investigate particular laws only in empirical inquiry, which can make progress but cannot attain knowledge of particular laws of nature (section 4).

Best system interpreters often claim support from a comparison between Kant’s rejection of ‘transcendental realism’ and Putnam’s rejection of ‘metaphysical realism’. But while some versions of this comparison may be warranted, the specific version that would conflict with my reading here is not. And Kant’s combination of a necessitation account with a restriction of our knowledge fits well the broader ambitions of his theoretical philosophy to defend a kind of epistemological modesty (section 5).

Finally, once we understand Kant’s account we can see that it is more philosophically compelling than the hybrid position attempting to marry Kant’s own views to a contemporary best system account of the laws of nature. And we can appreciate the lasting philosophical strengths of Kant’s position without trying to assimilate it to currently popular views (section 6).

1. A Best System Account of the Laws of Nature in Kant?

As noted above, the law that every alteration is causally determined is not supposed to admit any empirical justification; it is supposed to require the a priori defense of the Second Analogy, focused on consideration of the conditions of the possibility of experience in general. But Kant distinguishes such laws of nature ‘in general’ from ‘particular laws’ (B165) governing particular kinds of things, or ‘empirical laws’ (P 4:320; KU 5:181-6). I will not address here debates about whether the results of the Second Analogy are supposed to include the conclusion that all alterations are governed by regular and repeatable particular laws. Rather, I simply begin with the fact that Kant refers to particular natural laws; and I ask: What is it to be such a law? How, and to what extent, can we have knowledge of them?

Best system interpretations find an account of particular natural laws in Kant’s discussions of empirical scientific inquiry. In the Critique of Pure Reason, the central discussion is found in the ‘Appendix to the Transcendental Dialectic’. Here Kant argues that the ‘faculty of reason’ provides methodological principles required for the guidance of our empirical inquiry into nature. The points of greatest concern in best system interpretations are these: Our principles require us, first, to unify our knowledge by reducing the number of recognized basic kinds and laws. And they require us to simultaneously seek to discover an ever-greater range of natural diversity.[4] Taken together, these principles are supposed to direct us toward the guiding goal of a systematic unity of empirical knowledge of nature.[5] But reason’s principles have a special status: they are legitimate only as ‘regulative’ or guiding principles for our research, and never as grounds on which to assert knowledge of a real systematic unity of nature.[6] So we cannot actually reach the goal set by reason; the goal is ‘projected’ (A647/B675) by reason, and we can only approach the guiding principles and ideas of reason ‘asymptotically, as it were, i.e., merely by approximation, without ever reaching them’ (A663/B691).

Defending best system interpretations of this material requires a step that is not always clearly noted. Kitcher is the clearest about this: Kant’s discussion of empirical inquiry is most explicit in proposing ‘an epistemological thesis about the conditions under which we are justified in counting something as a law’; but such an epistemological thesis ‘does not provide a characterization of what laws are’ (1986: 210). Best system interpreters must argue that this material also proposes what Kitcher calls an ‘ontological correlate’ of the epistemological claims (1986: 214). Perhaps others would prefer a different description of it, but all best system interpretations require this step. For all seek to draw conclusions about Kant’s account of what it is to be a law of nature—for example, about Kant’s answer to the familiar question of whether a law of nature is a generalization stating a regularity. Best system interpretations hold that Kant’s particular laws are generalizations—more specifically, those generalizations that would fit into the best system of empirical knowledge. Buchdahl (1965), Brittan (1978) and Kitcher (1986) develop such interpretations. Allison (1996) defends Buchdahl’s approach. And elements of the approach show up in other recent work, especially the idea that Kant conceives of what particular laws are in a manner that is supposed to allow empirical knowledge of them.[7] But once we note the extra step required by best system interpretations, the way is clear to begin to see that Kant consistently holds a very different account of what it is to be a law. 

2. Kant’s Necessitation Account of Particular Laws

In his defense of a best system interpretation, Kitcher first considers a contrasting necessitation account of laws, according to which ‘laws express objective natural necessities’. On this account, science seeks knowledge of ‘objective dependencies’ and ‘natural necessities’. Kitcher recognizes the intuitive appeal of this necessitation account of laws, while also arguing that there are philosophical reasons why Kant should and does reject it: the necessitation account ‘honors a straightforward vision of the aims and practice of science at the cost of epistemological opacity’ (1986: 201-203). I will return to contest Kitcher’s philosophical case concerning epistemological costs below. In this section I want to begin arguing for the interpretive point that Kant himself embraces a necessitation account of laws and the straightforward intuitions supporting it.

The intuitions in question play a prominent role throughout the Transcendental Dialectic of the first Critique—the broader context of the discussion of empirical inquiry noted above and heavily stressed by best system interpreters. The broader section is concerned with the ‘faculty of reason’. Kant argues that we do not rest content with knowledge, provided by ‘the understanding’, that such-and-such is the case. We seek to explain, or to know why such-and-such is the case. In order to account for this interest, the faculty of reason must itself be characterized by an ‘aim’ (Zweck) or an ‘interest’ (Interesse): knowledge of the underlying conditions on which things depend.[8]  Reason demands that we follow a ‘regress from the conditioned to its condition’ (A523/B521). As noted above, reason’s demands are legitimate only as guiding or regulative principles: they direct us to seek underlying conditions, but they do not license us to assert knowledge that there must be such underlying conditions and in particular any complete series of conditions, or anything absolutely and completely explanatory.[9]

For our purposes, the most immediately important point here is that Kant conceives of explanation in terms of dependence: an explanation must provide information about an underlying condition on which an explanandum really depends. I will call this ‘the simple intuition’. To see the simple intuitive appeal, consider the asymmetry of explanation (Kitcher 1986: 203). A common example is this: From falling barometer readings we could infer that the atmospheric pressure is dropping, and vice-versa. But appeal to changing barometer readings cannot explain why the atmospheric pressure is changing, while appeal to the pressure changes can explain why the barometer readings are falling. The simple intuition provides a simple account of this: there are asymmetries of explanation because there are asymmetries of dependence. For example, the barometer reading really depends on the atmospheric pressure, but the reverse is not true. Perhaps an explanation must also convey information about an underlying condition in a manner that addresses the interests of a particular audience, so that what counts as explanatory would vary with context. If so, then the simple intuition would still hold that explanation is also constrained by the way the world is, and more specifically by the ways in which some things in the world really depend or do not depend on others. I take it that the simple intuition is part of what Kitcher calls the ‘straightforward vision’ on which science aims to discover not just ‘useful information about general regularities’ but ‘objective dependency’ in nature (1986: 202-3).

I do not claim that it is intuitive to hold that all explanations must convey information about a condition that necessitates an explanandum. The necessitation of interest here comes into view only with consideration of explanations appealing specifically to laws. The crucial feature of laws is that they carry implications about any number of instances. And the simple intuition about explanation will exert pressure against some ways of understanding this feature. For example, consider the idea that a natural law is a regularity or a generalization stating a regularity. Can we explain why B’s regularly follow A’s by appeal to the regularity that B’s always follow A’s? Not if the simple intuition is correct. For the proposed account would merely restate the regularity; it would not provide information about any underlying condition on which the regularity truly depends.[10]

Now consider an account that faces no such pressure from the simple intuition: to state a law of nature is to identify a kind on whose nature some natural phenomenon depends. For example, imagine for the moment that ‘salt is water soluble’ is a law of nature. If so, then there would be a regularity, and the generalization ‘all salt is water soluble’ will be true. But to state the law would not be to refer to the regularity or to make the generalization. It would rather be to provide information about an underlying condition on which the regularity depends, namely, the nature of the kind. So the simple intuition would allow that appeal to such a law could explain the regularity.

To understand the sense in which such an account might naturally hold that laws are distinguished by necessitation, consider again explaining the solubility of the white stuff coming out of my shaker by appeal to the nature of the kind salt. If there is such a law, then it implies not only the universal generalization summarizing all actual cases. It implies, for example, that a gold ring would have been water soluble if it had been made of salt—given the nature of salt, it would have to be. Similarly, one might show that this is not a law by showing that the nature of salt allows for a possible kind of salt that would not be water soluble—even it happens that this other kind of salt has never and might never actually exist. A law of nature, on such an account, allows no possible exception, or implies a universal generalization summarizing actual and even all possible cases. That is the sense in which a natural law, on such an account, would involve necessitation and strict or absolute universality.[11]

We can now recognize in some initial texts from Kant both the simple intuition about explanation and this last necessitation account of laws. Consider again Kant’s discussion of a causal law:

The concept of cause …requires that something A be of such a kind that something else B follows from it necessarily and in accordance with an absolutely universal rulethe effect does not merely come along with the cause, but is posited through it and follows from it. (A91/B124)

Note how each of the central points from above is present in this passage: (i) The simple intuition, applied to the specific case of causality: causal explanation requires information about how the explanandum really depends on a condition in the specific sense that it ‘follows from’ or is ‘posited through’ a cause, as opposed to merely happening to ‘come along’. (ii) Wherever appeal to a law can explain, there must be something general—a general kind—that serves as a condition on which an explanandum depends. For example, if there is a causal law connecting A’s and B’s, then A must ‘be of such a kind’ that some B follows. (iii) If something follows from the nature of a general kind, then it will follow not only in actual cases, but in all possible cases of the kind. For example, for any possible instance of A’s kind, some B will follow. So where there is a law, something follows ‘necessarily and according to an absolutely universal rule’.

Kant himself sometimes makes these points by considering the possibility of an accidental regularity, for example, good weather follows stork-sightings. Imagine that some more precise version of ‘good weather follows stork-sightings’ were a true generalization. It might still fail to truly state a law of nature. For perhaps the generalization is true by chance, so that no instance of good weather really depends on (or follows from or is posited through or has a ground in) anything about stork-sightings. And if there were no such dependence, then no appeal to such considerations could legitimately explain the weather. And there would be no sense in which the weather would have been nicer if only I had managed to find a stork, or if I had arranged to have one brought to town. In Kant’s terms:

There are cases where something is posited, and another thing is posited after, yet where the one is not a ground of the other. Eg, when the stork comes, good weather follows. But to posit does not mean something follows the other accidentally; for the stork could also be brought on the mail coach. (LM 314-5; Ak. 28:549)

This last passage is presented as contrasting with a case in which there is the necessity and universality characteristic of a law:

Ground is that upon which something else follows in a wholly necessary way; or ground is that upon which something follows according to universal rules; basically it amounts to the same thing. (LM 314; Ak. 28:548) 

So consideration of accidental regularities helps to bring out the sense in which natural laws are not regularities; where there is a natural law there is necessitation and absolute or strict universality.[12]

Note that there is room for further argument over whether necessitation-laws might also, in other senses, be contingent. For example, perhaps absolutely all possible instances of the actual kind salt must necessarily be water soluble; there would be room for argument about whether the nature of the kind could have been different in this respect. Those who favor contemporary necessitation accounts of laws, and reject best system accounts, often hold necessitation-laws to be contingent in such a further sense—or take laws to be ‘contingent necessities’ (Armstrong 1983). Watkins argues that Kant’s account of laws is similar to Armstrong’s.[13] And perhaps Kant’s own reference to a ‘contingent unity of particular laws’ (KU 5:386) does suggest the idea of a system of laws and kinds which could have been otherwise arranged. But I take no stand here on whether Kant can, should, or does also assert some compatible sense of contingency. The important point here is that a particular law, taken individually, involves necessitation, so that ‘salt is water soluble’ is a law if water solubility is necessitated by the nature of salt, regardless of how this law might relate to others in any system; any further and compatible sense of contingency would be no indication of the very different best system account of what it is to be a law.[14]

Finally, this section is greatly influenced by powerful accounts of Kant on causal laws both by Friedman and by Watkins. For better or for worse, however, I am looking for a different path to a different destination. I do not appeal to the kind of derivation by application of a priori principles emphasized by Friedman.[15] Nor do I attribute to the critical Kant, as Watkins does, a cosmology and a metaphysics of substance continuous with pre-critical rationalism and Kant’s pre-critical work.[16] I seek a more direct argument against best system interpreters and a different alternative to their views by: first, looking instead to Kant’s claims about the aims of empirical inquiry—claims central to precisely the critical-period texts emphasized by best system interpreters; and second, by turning below to the importance of the epistemological restrictions emphasized by best system interpreters—restrictions designed to limit our knowledge of necessity, and to ground Kant’s turn away from pre-critical philosophy.

3. Empirical Inquiry: Progress without Knowledge of Natural Laws

Some ways of limiting our knowledge of laws would be unsurprising. For example, Watkins attributes to Kant this view: we seek to know particular laws on the basis of experience and regulative principles, but this ‘can presumably never be established with absolute certainty, since one can never rule out the possibility that future evidence might require a revision’ (2004: 483-4). This is surely true. But it is not a surprising limit, and it does not rest on any of Kant’s distinctive epistemological commitments. On the face of it, empirical evidence leaves me without ‘absolute certainty’ of nearly anything, except perhaps whatever I am currently observing directly. For example, I am fairly certain that my dining room table is still in the next room, but while in the study I cannot be absolutely certain even of this. So it is unsurprising that empirical evidence also leaves us without absolute certainty about whether this or that is a law of nature. But I will argue what Kant has to say about particular laws of nature during the critical period is shaped throughout by his more distinctive and demanding epistemological restrictions.

During the critical period, Kant argues that we can only have knowledge within ‘the bounds of experience’. His basic line of argument is this: First, to expand our knowledge we must go beyond merely reflecting on our concepts; we require access to intuition. Analytic judgments cannot amplify our cognition (e.g. A7-8/B12), and synthetic judgments ‘are possible only by the relating of a given concept to an intuition’ (C 11:39). Second, we have access to intuition only where this is provided by sensibility—that is the key limitation of our merely ‘discursive’ understanding (e.g. B135; P 4:288). And this places limits on the extent of our knowledge: 

No concept can have its objective reality be secured, save insofar as it can be presented in a corresponding intuition (which for us is always sensory), so that beyond the bounds of sensibility and thus of possible experience, there can be no cognition whatever, that is, no concepts of which one is sure that they are not empty. (UE 8:188-9)

So we can know whether something satisfies a concept—and can generally extend our theoretical knowledge—only within ‘bounds of sensibility and thus of possible experience’: only where we can have ‘cognition’ (Erkenntnis) requiring not only concepts but also ‘corresponding intuition (which for us is always sensory)’.[17]

Part of the point of this limit is to sharply restrict our access to informative knowledge of necessity: It is well known that Kant allows knowledge of necessity only where we can have a priori knowledge (e.g. B4). So given the need for intuition, extending our knowledge of necessity would require access to a priori intuition. And Kant narrowly constrains our access to a priori intuition: given our dependence on sensibility we can have a priori intuition only of the forms of our own sensible intuition, space and time.[18] This constraint is supposed to rule out, for example, pre-critical rationalist assertions about the existence of an absolutely necessary highest being.[19] Knowledge of necessity is supposed to remain possible, for example, where synthetic a priori knowledge of mathematics is made possible by our a priori intuition of space and time.

It would be a mistake to think that these epistemic restrictions should apply only to claims about certain otherworldly of objects of interest to pre-critical metaphysics. Kant does not just stipulate that certain pre-critical metaphysical claims are illegitimate. He argues. That is, he defends restrictions concerning sensible intuition that apply to all claims to theoretical knowledge, and argues that pre-critical metaphysical claims are ruled illegitimate by these restrictions. As Allison puts the point, ‘it is precisely because of the impossibility of providing an intuition answering to the concepts that Kant holds the judgments of transcendent metaphysics to be ungrounded’ (2004: 95). And Kant’s position is stronger still insofar as he accepts in a principled manner all of the conclusions that follow from this line of argument.

With respect to particular laws, what follows is this: On the one hand, a priori intuitions of the forms of all sensible intuition in general are supposed to make possible synthetic a priori knowledge of the laws of nature in general—for example, consideration of the conditions of the possibility of time-determination establishes the law that all alternations are determined by a cause.[20] But Kant rules out the possibility of purely a priori knowledge of particular laws of nature—the laws of physics, chemistry, etc. One way Kant puts this point is by saying that ‘particular laws … can not be completely derived from the categories’ (B165). Sometimes Kant claims that there could be no knowledge of particular or ‘empirical’ laws without experience; as Friedman has emphasized, it is a mistake to read these passages as allowing purely empirical knowledge of particular laws.[21] For, on the other hand, Kant also rules out purely empirical knowledge of particular laws. Again, we cannot have empirical knowledge of necessity (e.g. B4). More specifically, empirical intuition from sensibility cannot serve as corresponding intuition in the case of judgments about necessitation: ‘no necessity of their connection is or can become evident in the perceptions themselves, since apprehension is only a juxtaposition of the manifold of empirical intuition’ (B219/A176). As noted in the initial citation above concerning the causal law governing a particular kind, we cannot have empirical knowledge of ‘a rule in accordance with which the succession is necessary… The strict universality of the rule is therefore not any property of empirical rules’ (A91/B124).

Granted, empirical inquiry is supposed to be able to reach knowledge of objects that we cannot perceive. One of Kant’s examples is this: ‘we cognize the existence of a magnetic matter penetrating all bodies from the perception of attracted iron filings’ (A226/B273). Contemporary necessitation accounts of laws such as Armstrong’s often hold that we can similarly infer from observed regularities to the necessitation-laws on which they are supposed to depend, giving us epistemic access to necessitation-laws comparable to our access to presently unobserved material objects.[22] Such views have their attractions, but they are at odds with Kant’s epistemological restrictions and with what Kant has to say about laws in the first Critique, the third, and in other texts published in this period. Kant holds that we can reach knowledge of unobservables only where these are such as could be given in sensible intuition, in accordance with the forms of space and time, if only our sense organs could detect ‘finer’ things. For we can reach knowledge by drawing inferences about unobservables only where we can know a priori that the laws of nature in general apply (e.g. that every alteration in time is causally determined). Or, only for an object such that

…in accordance with the laws of sensibility and the context of our perceptions, we could also happen upon the immediate empirical intuition of it in an experience if our senses, the crudeness of which does not affect the form of possible experience in general, were finer. (A226/B273)

But Kant’s worry concerning necessitation-laws is that we cannot have for judgments about necessitation any corresponding sensible intuition—the size or duration of whatever might be necessitated is immaterial to this worry. So Kant’s specific restrictions, stemming from our dependence on sensible intuition, place no restriction on our access to knowledge of presently unobserved objects, nor even on our access to knowledge of objects too small or events too brief for our sense organs to detect, but they cannot allow in any similar manner knowledge of particular laws of nature.[23]

What hope can there then be for empirical investigation of particular laws? I would explain Kant’s general view as follows: Consider all of the observations that we can make of different substances dissolving in liquids. With nothing but initial observations to go on, we might be distracted by innumerably many initially observed regularities. For example, we might initially notice a correspondence between whether white powders dissolve and the movement of some planet; when that correspondence breaks down one could always look for more complex patterns linking solubility and movements of planets. Fortunately, then, we are not limited to initial observations. The principles guiding empirical inquiry direct us to narrow our focus, seeking specifically those regularities that would be explained by the simplest system of basic laws, kinds and forces. Our guiding principles might direct us to single out, for example, the rule or statement ‘salt is water soluble’. But such guidance falls short of establishing knowledge of a law; rather, we regard the general rule as a law for the purposes of further research. And we might well discover thereby that there is no such law. If so, then we might distinguish lower-level kinds like the underlying elements and state regularities concerning their interactions. For the reasons noted above concerning sensible intuition, this still cannot yield knowledge of laws, nor any justification for concluding that we have reached particular laws—not even if we modestly add that one can never be absolutely certain. But if we can account in these new terms for the behavior of salt and other compounds, then we have made a real advance, and we do have reason to conclude that our theories are improving or that we are making progress, or that we are improving in approximation to knowledge of particular laws. One might object that it does seem possible that we could come to focus our attention on statements that do in fact state natural laws. Still, empirical inquiry could not provide evidence that we have reached laws (as opposed to making progress toward them)—it can never achieve knowledge of a particular law as such.[24]

We can find this combination of a necessitation account with a limitation of our knowledge, surprisingly, in precisely the texts emphasized by best system interpreters of Kant. Those interpreters claim the most direct textual support from passages in the published Introduction to the Critique of the Power of Judgment (KU). This material differs in many respects from the Appendix to the Dialectic in the first Critique, especially insofar as it portrays empirical inquiry as guided by principles of ‘reflective judgment’ and in particular a ‘principle of purposiveness’.[25] But what is important here is that this material embraces the necessitation account of laws, and consequently limits our knowledge of laws. For example, one passage stressed by best system interpreters is found at KU 5:184-5. Buchdahl intersperses interpretation and quotation of this passage on empirical or particular natural laws as follows: ‘these laws, being only “empirically known,” are “as regards the understanding, contingent”’ (1965: 200). Kitcher cites the same passage and comments: ‘such laws of nature must be apprehended, and apprehended as laws, on the basis of experience’ (1986: 208). But the text makes no suggestion of allowing empirical knowledge of laws; Kant’s philosophical point here is precisely the opposite—precisely that we cannot have such knowledge. Kant carefully writes that we can have empirical knowledge of rules, but not laws. [26] He refers to

…a certain order of nature in its particular rules [Regeln], which can only be known to it [our understanding – JK] empirically and which from its point of view are contingent. These rules, without which there would be no progress from the general analogy of a possible experience in general to the particular, it must think as laws [Gesetze] (i.e., as necessary), because otherwise they would not constitute an order of nature, even though it does not and never can cognize their necessity. (KU 5:184-5)

Our guiding principles allow us to single out and ‘think as laws’ the empirically-known rules, specifically for the reason Kant emphasizes: without doing so we could make ‘no progress’ in investigation of the particularities of nature. But Kant holds here that particular natural laws involve precisely the sort of necessity which is independent of the conditions under which we can have knowledge so as to preclude empirical inquiry reaching knowledge of laws: given the ‘point of view’ of our understanding we ‘never can cognize their necessity’.

Allison takes a nearby passage to be decisive in his defense of Buchdahl’s general approach. Here Kant refers to laws that

may seem to be contingent in accordance with the insight of our understanding, but which, if they are to be called laws (as is required by the concept of nature) must be regarded as necessary on a principle, unknown though it be to us, of the unity of the manifold. (KU 5:179-80)

But—contra Allison—the principle referred to here is not said to play any role in constituting what it is to be a law or in ‘the grounding of genuine laws’, nor any role in grounding ‘the grounding of necessity’ that distinguishes laws.[27] Here too, the importance of the principle is that it allows us only to regard the rules we abstract from empirical inquiry as statements of laws. But what is it to be a law? Here too the answer involves a sense of necessity independent of our guiding principles in a way that rules out empirical knowledge of laws. Even if we were to hit on a hypothesis that does happen to state a law, we cannot have empirical knowledge that this is a law, for it would always ‘seem to be contingent in accordance with the insight of our understanding’. 

A similar view is present back in the Appendix to the Transcendental Dialectic. Kant here argues that empirical inquiry seeks knowledge of laws characterized by some kind of universality, specifically under the guidance of principles provided by the faculty of reason. Kant says: ‘this use of reason is only regulative, bringing unity into particular cognitions as far as possible and thereby approximating the rule to universality’ (A647/B675; my emphasis). There is no suggestion here that we should reconceive of what it is to be a particular law in terms of some kind of universality that would bring laws within reach of the principles guiding our empirical inquiry—some kind of universality that would merely approximate the strict universality of necessitation-laws. On the contrary, the point is that we seek in empirical inquiry knowledge of a kind of universality that is independent of the conditions under which we can have knowledge, or out of reach, so that empirical inquiry can only approximate to knowledge of the sort of universality that it seeks.[28] 

It would be interesting to compare and contrast this account of the limitation of our knowledge of laws to Locke’s and Hume’s views, but there are of course many controversies about their views that I will not attempt to resolve.[29] What is most important here is this: Kant consistently holds a necessitation account of what it is to be a law. It follows from this, given his general epistemological commitments, that empirical inquiry cannot achieve knowledge of particular laws of nature. And that is why Kant holds that empirical inquiry can only continually improve our approximation to such knowledge.

4. Seeking Knowledge of Laws Not Derivable From A Priori Principles

There is a very different discussion of our knowledge of particular laws in Kant’s 1786 Metaphysical Foundations of Natural Science (MAdN), for Kant here attempts to establish knowledge of some particular laws. He does so specifically in the case of the laws governing the motion of all matter, a case emphasized especially by Friedman (1992a). But there are important and easy to overlook connections between this material and the discussion of empirical inquiry in the third Critique: both discussions turn on the non-epistemic necessitation account of laws presented above; and both discussions argue that an ineliminable barrier prevents us from achieving knowledge of particular laws outside of one limited kind of special case.

MAdN begins with a discussion of what would be required for knowledge of particular laws. This discussion turns on Kant’s non-epistemic necessitation account, according to which there is a particular law where something is (i) necessitated by (ii) the nature of a particular kind. First, the reference to necessitation is the reason why we cannot have purely empirical knowledge of particular laws. In Kant’s terms, if a science is to achieve knowledge of particular laws then it requires a ‘pure part’:

Since the word nature already carries with it the concept of laws, and the latter carries with it the concept of the necessity of all determinations of a thing belonging to its existence, one easily sees why natural science must derive the legitimacy of this title only from its pure part. (MAdN 4:468-9)

Second, because knowledge of a particular law would require knowledge of the ‘particular nature of this or that kind of thing’, we can never have purely a priori knowledge of laws, or a science with knowledge of laws would require an ‘applied part’ (MAdN 4:468).

Granted, it is extremely difficult to understand how Kant could hope to meet the above constraints in establishing knowledge of particular laws; for how could such knowledge be partly a priori and partly empirical? Friedman proposes an interpretation of Kant’s attempt to establish, for the laws of motion, knowledge with just such a ‘peculiar mixed status’ (1992a, 174). But I will here neither interpret nor evaluate Kant’s attempt. And I will not address interpretive controversies concerning this attempt, including Friedman’s account of how Kant applies the a priori principles of the understanding to the empirical concept of matter seeking to transition from empirically observed regularities to knowledge of laws of motion.[30] My topic is not this specific case, but particular laws more generally. So I am more interested here in Friedman’s claims about this topic. And I reject Friedman’s claim that Kant also seeks to account for what it is to be a particular law, in general, in terms of the attempt to establish knowledge of the laws of motion specifically. Friedman’s proposal is that to be a particular law is to be derivable by further application of the a priori principles of the understanding: particular laws are characterized by ‘empirical necessity’, and this necessity ‘can derive from nowhere else than an a priori grounding in the principles of the understanding’ (1992a, 190).

I have already offered evidence above that Kant accounts for what it is to be a particular law, in general, in terms of a kind of necessitation not itself defined in terms of—not derived from or grounded in—any account of our knowledge of laws. But there is also a problem for Friedman’s proposal in the MAdN texts that he emphasizes: MAdN denies that all of the particular laws of which we seek knowledge can be derived in any way from a priori principles of our understanding; so clearly Kant cannot here conceive of what it is to be such a law in terms of the possibility of such derivation. And Kant is right to impose this limit: his account of our epistemic limits rules out such derivations of particular laws generally.

To see why this is so, recall that what limits our knowledge is our dependence on sensibility if we are to have, for our concepts, ‘intuition corresponding to them’ (A50/B74) or ‘corresponding intuition’ (UE 8:188-9). Empirical intuition from sensibility cannot provide corresponding intuition in the case of laws involving necessitation. So for knowledge of particular natural laws, a priori intuitions of space and time would have to provide corresponding intuition. And MAdN puts the point in precisely these terms: given the empirical concept of a particular kind, ‘it is still required that the intuition corresponding to the concept be given a priori’ (MAdN 4: 470). We can here catch at least a glimpse of why Kant hopes that the empirical concept of matter might present a special case: a priori intuitions of space and time are supposed to correspond to the empirical concept of matter insofar as everything that can show up for us in outer sense (whose form is space) is corporeal (4:470), and more specifically matter in motion (4:476). The hope is that a connection between our a priori intuition and the concept of matter could make possible the application of mathematics to establish knowledge of the laws of motion.[31]

But regardless of what the prospects for this project might be in this specific case, we can clearly see that Kant is right to rule out the possibility of similar derivations for particular laws generally. For example, MAdN is skeptical about the possibility of knowledge of particular chemical laws, for ‘the principles of chemistry are merely empirical and admit of no presentation a priori in intuition’ (MAdN 4:471). What is crucial and easy to overlook here is that chemistry is an example illustrating a general and ineliminable restriction of our knowledge. Chemistry is a case in which we naturally expect laws governing not a single basic kind like matter, but rather relations between specifically distinct natural kinds. For example, perhaps water, earth, etc. are distinct kinds whose interactions are governed by natural laws; or perhaps hydrogen, helium, etc. But no concept of such a kind specifically distinct from others could ever apply to everything that shows up in outer sense, or could correspond to a priori intuition of the forms of our sensibility in this way, for it must be distinct from the other kinds with which it is supposed to interact in a law-governed manner. Granted, Friedman is right to say that Kant can allow the possibility of chemistry advancing and achieving knowledge of chemical laws.[32] But such an advance would require chemistry to become more like Kant understands physics to be—the distinct chemical kinds and laws governing them would have to be reduced to something like a single basic kind (like matter). In Kant’s terms, we cannot make the advance unless for the ‘chemical actions of matters on one another’ it is possible that ‘their motions and all the consequences thereof can be made intuitive and presented a priori in space’.[33]

No matter what surprising advances a science like chemistry might make, then, we will still always lack knowledge of particular natural laws governing the interactions between fundamentally distinct or diverse kinds. And science will remain interested in such laws insofar as science seeks knowledge of the diversity of nature.

That scientific inquiry seeks knowledge of laws governing distinct or diverse kinds is emphasized in the published Introduction to the third Critique. Kant there refers to ‘such a manifold of forms in nature’, or to ‘so many modifications of the universal transcendental concepts of nature that are left undetermined by those laws that the pure understanding gives a priori’ (KU 5:179). Shortly thereafter Kant gives precise articulation to this limitation of our knowledge:

Specifically distinct natures, besides what they have in common as belonging to nature in general, can still be causes in infinitely many ways; and each of these ways must … have its rule, which is a law, and hence brings necessity with it, although given the constitution and the limits of our faculties of cognition we have no insight at all into this necessity. (KU 5:183)

Kant refers to cases in which distinctions between basic kinds rule out the possibility of deriving knowledge of governing laws from any a priori principles of our understanding. Clearly Kant does not here conceive of what it is to be a particular law in terms of the possibility of the sort of derivation or grounding that he here rules out. Friedman sees this threat, and takes the point of this passage to be that ‘the vast majority of empirical laws have not yet received this kind of grounding’ (1992a, 190). But this cannot be right. Kant’s epistemological commitments justify the claim that he in fact makes here: the limit is in principle and ineliminable; our lack of knowledge stems not from the state of science at some particular time but from our limited access to a priori intuition, or from ‘the limits of our faculties of cognition’.

One could try to argue that we can somehow expand our knowledge of natural necessity without a priori intuition, or somehow expand our access to a priori intuition, but this would be to take leave of the critical Kant and his sharp restrictions on our access to knowledge of necessity. Granted, Kant does sometimes argue that we have a limited ability to conceive of a form of intellect able to bypass the mediating conditions of our sensible intuition and achieve immediate knowledge of reality.[34] We cannot have theoretical knowledge that there is such an intellect. Still, insofar as ‘we have no insight at all into this necessity’ of laws governing distinct kinds, thinking of this sort of particular law is thinking of something that we could only know if we enjoyed some such superior sort of a priori ‘insight’.[35] But this is not to define what it is to be a law is in terms of any kind of knowledge, nor say that we have any possibility of deriving such laws from the principles of our merely discursive understanding.

It is also true that, if our epistemic limits prevent knowledge of natural laws governing specifically distinct kinds, then we must lack even knowledge that there are such laws. First of all, so far as we can know nature is both diverse or complex and yet also law-governed: ‘the objects of empirical cognition are still determined or, as far as one can judge a priori, determinable in so many ways … that specifically distinct natures’ are governed by necessitation-laws (KU 5:183). Second, it is crucial that our empirical inquiry is not guided only by a demand to simplify and unify; if it were, then we might be prone to falsifying the apparent diversity of nature by seeking to treat all sciences like the science of motion, or seeking to reduce the natural kinds of interest in all sciences to a single basic kind, like matter. The first Critique rather specifies that the natural sciences also seek knowledge of ‘manifoldness and variety’ (A654/B682). And the third Critique specifies that we require guidance by the idea of a system of laws governing diverse or specifically distinct kinds: ‘the understanding … requires in addition a certain order of nature’, even though we cannot have knowledge of or ‘cognize’ the necessity of the natural laws supposed to comprise this order (KU 5:184-5). So empirical inquiry into the diversity of nature requires us to think or ‘assume’[36] for the sake of inquiry that there are natural laws governing specifically distinct kinds, and to seek knowledge of such laws—even though we can thereby only improve our approximation to knowledge of them.  

In sum, Kant holds one consistent view about what it is to be a particular law of nature during the period discussed here, spanning from the first Critique to the third. Kant distinguishes particular laws in terms of necessitation, or a form of dependence, and not in terms of any account of our knowledge. So he is free to both rule out the possibility of purely empirical knowledge of particular natural laws (as emphasized by Friedman), while also holding that particular natural laws generally cannot be given any derivation from a priori principles (as emphasized by best system interpreters). The overlooked key to Kant’s consistent position is his distinctive account of our epistemic limits, from which it follows that we can generally only seek knowledge of the particular laws of interest in the sciences in empirical inquiry, and that we cannot generally achieve knowledge of the laws we seek.

5. The Broader Significance within Kant’s Theoretical Philosophy

Kant’s broad project in theoretical philosophy aims at supporting ‘transcendental idealism’. Kant sometimes says more specifically that he rejects ‘transcendental realism’ in seeking to defend ‘empirical realism’ (e.g. A369ff.). Some see in this broad theoretical project a similarity between Kant and Putnam’s more recent rejection of ‘metaphysical realism’ in favor of ‘internal realism’. And some, including Kitcher, see this connection as support for a best system interpretation of Kant’s account of the laws of nature.[37]

Much here depends on the precise connection proposed. One way of making the connection with internal realism would be to attribute to Kant this view: the objects of our experience are not independent of or “external” to the conditions of the possibility of experience. To advocate this view would have to involve explaining more precisely the sort of dependence or connection asserted here. But advocates could look to Kant’s claim that ‘the conditions of the possibility of experience in general are at the same time conditions of the possibility of the objects of experience’ (A158/B197). And to the way in which the Second Analogy argues from the conditions of the possibility of time-determination to a principle concerning alterations themselves—namely, that ‘all alterations occur in accordance with the law of the connection of cause and effect’ (B232).

I neither oppose nor advocate such a comparison to internal realism, but only note that it would not conflict with my interpretation here. The denial of independence formulated in the previous paragraph is restricted in that it applies to objects of which we can have knowledge or objects as known by us. True, I hold that Kant does not define what it is to be a particular law in general in terms of any account of our knowledge of laws. But in specific cases there might be particular kinds and laws with different sorts of connection to or independence from the conditions under which we can have knowledge. Consider specifically the particular natural laws of which Kant allows knowledge: the laws of motion. Though I do not pursue the nature of the connection here, I have argued that Kant takes his epistemic limits to allow his attempt to derive these laws because he sees some kind of connection between the kind matter and the a priori forms of our own sensible intuition.

My interpretation would be ruled out, however, if a different comparison held up. Putnam at one point proposes what looks like an unrestricted claim about all objects: ‘“objects” do not exist independently of conceptual schemes’ (1981: 52). Though I make no claim to interpret Putnam here, it is worth considering the possibility that Kant holds this unrestricted form of internalism (hereafter UI): the very idea of anything independent of the conditions under which we can have knowledge is unintelligible, and unintelligible in a sense demonstrating that there cannot be any such thing. I have portrayed Kant as holding that we take there to be, for the purposes of inquiry, laws governing specifically distinct kinds, although we cannot achieve knowledge of such laws. Proponents of UI would hold that there cannot be such unknowable laws, and that the very idea of them is unintelligible. So if Kant accepts UI then this would require a very different account of the laws of nature—an account of what it is to be a law in terms of an account of our knowledge of laws, perhaps in terms of the best system of knowledge at the limit of empirical inquiry.

But attributing UI to Kant would be an extremely radical step, insofar as there are many passages that are inconsistent with UI. Consider one of Putnam’s own claims: ‘Kant has, in a way, two philosophies’, because sometimes Kant claims that ‘we cannot really form any intelligible notion of a noumenal thing’ and yet sometimes Kant claims that ‘there is God, Freedom and Immortality’ in a real ‘noumenal’ world (1987: 41-2). If the former view amounts to UI, then this passage suggests two philosophical possibilities: assert knowledge that there can be nothing beyond the bounds of experience; or assert knowledge that there are things beyond those bounds. If these were the only philosophical possibilities, then perhaps it would make sense to read Kant as follows: Kant’s critical epistemological claims certainly rule out asserting theoretical knowledge of objects beyond the bounds of experience; so these central claims commit Kant to taking the only other possible path, asserting knowledge that there can be nothing beyond the bounds of experience. We would then have to conclude, as Putnam does, that Kant sometimes violates his commitment. Now clearly Putnam’s reading raises many issues that go beyond my scope here; there can be no question here of providing an interpretation of Kant on noumena, God, freedom, immortality, transcendental idealism, empirical realism, etc. But we can at least see here that the two possibilities mentioned by Putnam do not exhaust the philosophical possibilities, so that there is no reason why we must see Kant as aiming to choose one or the other and as obviously failing to do so in a consistent manner.

To see that there are other possibilities, consider Kant’s own claims concerning the limitation of our knowledge. If one holds that our knowledge is limited or restricted by the sort of conceptual scheme to which Putnam refers above, then perhaps one should conclude we cannot intelligibly draw on that very conceptual scheme in order to conceive of objects beyond the reach of our knowledge. But Kant’s limit is very different: Kant does hold that knowledge requires concepts; but Kant holds that our knowledge is limited or restricted not by concepts or conceptual schemes but specifically—as we have seen—by our dependence for all intuition on sensibility and its specific forms. This point tends to be agreed in recent work focusing specifically on Kant’s ‘transcendental idealism’—for example by both proponents and critics of ‘metaphysical’ interpretations.[38] And this limit allows the claim, which Kant himself makes, that we can employ our concepts in thinking or conceiving things that we cannot know given our dependence on sensible intuition—so that we can know neither that there are such things nor that there are no such things. We must not assume, then, that Kant must seek either to assert knowledge that there are things beyond the bounds of experience, or else assert knowledge that there cannot be anything beyond those bounds. Kant says, for example: ‘I can think what I like, as long as I do not contradict myself … even if I cannot give any assurance whether there is a corresponding object within the sum total of possibilities or not’.[39] With respect to the pure categories of the understanding, ‘the categories are not restricted in thinking by the conditions of our sensible intuition, but have an unbounded field’ (B166n).

And it is misleading to say that it is ‘when Kant came to write his moral philosophy’ that he contradicts unrestricted forms of internalism (Putnam 1987: 41). Kant seeks to defend a theoretical philosophy, at odds with UI, with the theoretical virtue of epistemological modesty. Thus Kant specifically identifies as insufficiently modest, or pretentious, the claim that absolutely all things must conform to the conditions under which we can have knowledge—or the sort of unrestricted internalism formulated above in UI. In Kant’s terms, his claims about the merely regulative role of the ideas of reason is important,

partly in order to restrain the worrisome pretensions of the understanding, as if (in virtue of being able to furnish a priori the conditions of all things that it can cognize) it has thereby also confined the possibility of all things in general within these boundaries (KU 5:167-8).[40]

And Kant emphasizes that asserting knowledge of non-existence of something beyond the bounds of experience would be just as dogmatic as asserting its existence. For example, ‘empiricism itself becomes dogmatic in regard to the ideas (as frequently happens)’ when it ‘boldly denies whatever lies beyond the sphere of its intuitive cognitions’ (A471/B499).

To be sure, Kant does sometimes say that our concepts employed beyond the bounds of sensibility or experience lack significance or sense. Although there are many complications here, note that Kant refers to one very specific kind of significance. For example, with respect to the categories: ‘without descending to conditions of sensibility … all significance (Bedeutung), i.e., relation to the object, disappears’ (A241/B300). The distinction between schematized and unschematized categories makes room for Kant’s claims that another kind of significance remains: ‘even after abstraction from every sensible condition’ the categories retain ‘a logical significance’ (A147/B186). Clearly one could allow us to think or conceive unknowables while denying that such thoughts have significance in the specific sense of the sort of ‘relation to the object’ necessary for cognition and knowledge.[41] These passages, then, need not be read as endorsing UI and so as obviously inconsistent with so many of Kant’s other claims and ambitions, including his aim for the epistemological modesty noted above.

My own view is that the many unanswered questions about Kant’s transcendental idealism can be addressed in a manner extending my approach to Kant on laws. For example, Kant claims that our dependence on sensibility for intuition prevents us having knowledge of things as they are in themselves. On the one hand, I think that we should seek to understand how this is supposed to amount to a surprising and substantial restriction or limitation of our knowledge—not just the tautological claim, for example, that we cannot know what we cannot know.[42] On the other hand, I think that we should aim to make sense of this in a manner that is consistent with the epistemological commitments involved in Kant’s break with pre-critical metaphysics; we should not read Kant as asserting knowledge, with pre-critical metaphysics, that there are real unconditioned grounds, or purely intrinsic features of things—while denying only that we can have more specific knowledge of these.[43] My own view is that both demands can be met by attending to what Kant says about the aims of theoretical inquiry.

But, again, my point here is neither to develop nor to defend an interpretation of Kant’s claims about transcendental idealism, noumena, meaning, the categories, etc. My point is rather this: We should not assume that Kant aims to choose from the subset of philosophical options that happen to be allowed by contemporary opponents of metaphysical realism. And we certainly should not assume such a controversial position drawn from one side or the other in contemporary debates and then view Kant through this lens, discarding as inconsistent mistakes all of the many incompatible passages. Insofar as we seek to understand Kant’s broader theoretical philosophy we should focus on how he attempts to develop a coherent position of his own, in response to philosophical aims of his own.

Insofar as we seek to understand Kant’s account of the laws of nature and explanation, we should try to understand how the texts emphasized in rival interpretations in fact present a coherent philosophical position on these topics. In trying to do this, I have argued that Kant holds this view: an explanation must provide information about an underlying condition on which an explanandum depends. If so, then we can have explanatory knowledge only where we can have knowledge of such conditions—and certainly never beyond the limits of our knowledge. For example, we cannot have explanatory knowledge involving laws except where we have access to ‘known laws’.[44] But this is no reason to attribute to Kant the sort of unrestricted internalism that would define what it is to be explanatory entirely in terms of the conditions of the possibility of our knowledge, so that there could not be any intelligible question about explanation for which we cannot in principle know the answer.[45] Though there are many other complexities regarding their status,[46] Kant consistently holds that the principles guiding our inquiry have a regulative status; and we have just seen that part of the point of this status is to block the dogmatic claim that absolutely everything must be such that we can have knowledge of it. If that is dogmatic, then it would also be dogmatic to hold that every underlying condition and every form of dependence must be such that we can have knowledge of them. If so, then we should not seek to rule out a priori the possibility of intelligible questions about explanation for which we cannot know the answers. Nor should reconceive of what it is to be a law of nature so as to try to rule out a priori the possibility of intelligible questions about laws for which we cannot know the answers. Kant’s view, I have argued, is that science pursues such questions about laws, and inquiry is guided by such questions, even though we cannot in general achieve knowledge of the answers.

6. The Philosophical Advantages of Kant’s Necessitation Account of Laws

I have argued that the best system account discussed by Kitcher and other recent interpreters is not present in Kant at all. The view Kitcher favors, then, is a kind of hybrid, marrying elements from Kant with a recently popular best system account of what it is to be a law; for ease of reference, I will call the hybrid view ‘KBS’. I now argue that, whatever attractions contemporary best system accounts might have, Kitcher fails to demonstrate any philosophical superiority of the hybrid view, KBS, over the necessitation account that Kant himself consistently holds. And I argue that we can appreciate the unusual combination of philosophical strengths of Kant’s own position on laws, without trying to assimilate it to any currently popular view.

Kitcher defends KBS by appealing to two sets of philosophical costs and benefits. More specifically, KBS is supposed to offer a ‘via media’ between two extremes. One extreme is a necessitation account of laws, according to which ‘laws express objective natural necessities’. The other extreme is a ‘Mach-Duhem’ view, on which science aims only to organize our empirical observations in a manner that is useful for us. Both have ‘one great merit’ and face ‘one major difficulty’:

The Aristotelian approach honors a straightforward vision of the aims and practice of science at the cost of epistemological opacity, while the Mach-Duhem conception purchases its epistemological purity by renouncing what seem to be widely exemplified scientific virtues. (1986: 201-203).

So KBS, first, has epistemological benefits relative to the costs of necessitation account. Second, KBS has intuitive benefits, relative to the Mach-Duhem view, insofar as it better accommodates the intuitive or ‘straightforward vision’ of the aims of science. 

But, first, Kitcher fails to demonstrate that KBS enjoys any advantage with respect to epistemological costs and benefits. Perhaps one could try to criticize contemporary necessitation accounts on grounds that they fail to adequately explain how we can have empirical knowledge of the laws of nature. But Kant denies the possibility of empirical knowledge. Can KBS claim superiority here by allowing such knowledge and explaining how it is possible? No. Perhaps KBS seems superior here, given this intuition: knowledge of necessitation-laws would have to be a different kind than ordinary empirical knowledge, but we could have knowledge of best-system-laws in the sense that this would require only more or a greater degree of the ordinary empirical knowledge that we do have. Consider, however, what Kitcher says elsewhere, in his own name, about the ideal limit: ‘the relevant notion of idealization consists in liberating our actual, scattered, experience from its partialities and its spatio-temporal boundedness’ (1995: 662). One could simply assert that we are in a better epistemic position relative to a perspective liberated from spatio-boundedness than we are in relative to necessitation. But this would be ad hoc compared to Kant’s argument for a principled limitation of our knowledge. Kant argues that our knowledge is limited by our dependence on the faculty of sensibility, which is limited precisely by the forms of our sensibility—space and time. If Kant is right, then there can be no better candidate for a difference in kind than that between our experience and anything liberated from ‘spatio-temporal boundedness’. Perhaps a contemporary best system account could enlist support from a different epistemological argument about our epistemic limtis; but, if so, this would be no consolation for the attempt to combine a best system account with Kant’s philosophy—no consolation for KBS.

But Kitcher’s most powerful epistemological worry is this: Even if there are real relations of necessitation in nature, the generalizations that fit the most unified system of empirical knowledge might not correspond to them. On a necessitation account, then, ‘the rationality’ of our preference for unification ‘depends on contingent facts about the world, and there is only a contingent connection between theoretical understanding and unification’. Kitcher proposes eliminating the ‘gap’ between empirical inquiry and the laws it seeks: ‘Kant's problem about a possible gap between the search for causal law and the imposition of unity is resolved by making unity constitutive of explanatory dependence’ (1994: 263). Applying this solution to particular laws would yield the claim that ‘the imposition of unity’ is ‘constitutive’ of what counts as a particular law, or constitutive of the only sort of ‘necessity’ that truly distinguishes laws.

KBS, however, preserves the contingency and so the worry about justification. A best system account might guarantee that one would begin to approximate knowledge of laws as one approached the limit of inquiry. But perhaps the world is infinitely complex in the sense that our finite inquiry would never approach the end of all possible empirical observations, at every level of detail, resulting from all possible experiments. On the face of it, it is a contingent fact about the world whether or not this is so. And, as critics of best system accounts have noted, the results of following guiding principles will vary greatly given different initial beliefs.[47] On the face of it, then, it is a contingent matter whether or not our inquiry began with intuitive beliefs that have so far sent inquiry only away from the ideal. So why should it be rational for us to follow guiding or unifying principles that might—no matter how long of a finite period we devote to empirical inquiry—still be leading us astray? If such contingency amounts to a philosophical cost, then it is a cost KBS incurs.

Kitcher recognizes this worry about KBS. And he proposes that the worry is addressed by Kant’s third Critique account of a guiding ‘principle of purposiveness’:

If we presuppose the principle of purposiveness, then currently adopted explanatory dependences will be approximations to those that would emerge in the limit of inquiry (1994: 268; see also 266).

But however promising or unpromising this strategy might be, it will on the face of it be equally so when modified to apply to Kant’s own necessitation account, as follows:

If we presuppose the principle of purposiveness, then current beliefs about explanatory dependences will be approximations to the truth about real relations of dependence, including the truth about which regularities in nature are necessitated by the natures of which general kinds.

In sum, then, Kitcher does not demonstrate that KBS has an advantage with respect to epistemological costs over the Kant’s own combination of necessitation account with restrictions on our knowledge.

Second, Kitcher argues that KBS has intuitive benefits, relative to a Mach-Duhem view, insofar as it better accords with the ‘straightforward vision’ motivating necessitation accounts—the vision of science as aiming for knowledge of ‘natural necessity’ and ‘objective dependency’ (1986: 203). But if this is a benefit, then surely the views that will reap the most benefit are necessitation accounts themselves.

And best system accounts will come out badly by Kitcher’s own measure here, for they conflict with the intuitions at stake. This is widely recognized even in recent defenses of best system accounts. For example, Schaffer’s (2007) defense recognizes this example: ‘Imagine a relatively simple world with just a single electron moving in a straight line forever’. Intuitively, fixing what happens in this world leaves open many potential answers to questions about what the governing laws are. Here are some intuitive potential answers: ‘(i) this world is governed by Newton’s three laws; (ii) this world is governed by the single law that all things move in straight lines forever’ etc. But on a best system account, counter-intuitively, fixing everything that happens must already determine what the laws are. So if we accept the intuition that ‘there can be differences in lawhood without differences in history’ (Schaffer 2007, 94), then we must reject best system accounts of laws.

Note that this last difficulty stems directly from a conflict between best system accounts and the intuition about ‘objective dependence’. The intuition, applied to the laws of nature, is this: what happens is governed by or dependent on the laws of nature. Best system accounts conflict with this intuition: on such accounts what the laws are depends on what happens or on history, rather than the reverse. Even contemporary defenses tend to concede, then, that best-system-laws do not ‘govern’ but rather ‘describe’ or ‘summarize’ what happens; these contemporary defenses recognize that they must attack intuitions about governing laws, or the dependence of events on laws, arguing that these intuitions are incoherent and/or unacceptably rest on the idea of a God’s legislation or governance.[48] But there is no reason to find in this contemporary strategy any consolation for the proposal to marry a best system account with Kant’s philosophy, or for KBS; for Kant embraces the intuitions in question, and Kitcher agrees insofar as he appeals to the dependence intuition in his case for KBS.

No best system account can hope to instead find support by accommodating intuitions about dependence. True, Kitcher’s KBS is supposed to make ‘unity constitutive of explanatory dependence’ (1994: 263). So KBS can hold that, in seeking knowledge of the laws of nature, science seeks knowledge of a kind of ‘explanatory dependence’. But just naming the reconstituted notion ‘explanatory dependence’ does nothing to accommodate the intuition that, in seeking knowledge of a law, science is seeking knowledge of something on which regularities really depend. We have just seen that best system accounts must deny this.

True, a best system account can hold that laws of nature involve a kind of necessity, so long as it holds that this ‘necessity’ is also constituted by the unity of a system. Perhaps contemporary accounts can show that this notion of necessity has philosophical virtues. But no best system account can accommodate intuitions about necessity and dependence by holding that best-system-laws ‘legislate for unactualized possibilities’ (Kitcher 1986: 219 and 231). The word ‘legislate’, like ‘govern’, suggests that what can happen depends on what the laws are. This cannot be true on a best system account. On a best system account, the laws of nature are a subclass of the true generalizations; whether or not such a generalization is true in a possible world depends on what happens there, or on ‘history’ as Schaffer puts it; what happens in an actual or possible case cannot be legislated by or dependent on best-system-laws.[49]

We can now summarize the costs and benefits: First, Kitcher demonstrates no epistemological benefits of KBS over the necessitation view that Kant himself holds. Second, KBS is inferior to necessitation accounts with respect to our intuitions about explanation and dependence—assuming, with Kitcher, that conflict with such intuitions is a philosophical cost. So focusing on the costs and benefits recognized by Kitcher’s argument supports the conclusion that the attempt at a marriage between Kant and best system accounts is philosophically less compelling that the necessitation account that we have found present in Kant’s text. Contemporary best system accounts might have their advantages, but not advantages that could fit with Kant’s philosophical commitments: Kant embraces intuitions about explanation and dependence that conflict with best system accounts; and Kant’s epistemological commitments prevent claiming any epistemological advantage for a best system account.

Kitcher elsewhere worries that Kant is sometimes driven by an ‘aprioristic yearning’ that ‘can no longer be seen as an appropriate model for our own philosophical engagement with science’ (1996: 411). Perhaps one might try to argue on these grounds that there is something obsolete about Kant’s attempt to derive the laws of motion specifically. But this is no reason to worry about the general account of the particular laws of nature that has been my topic here. For while Kant may well yearn for certainty, we have seen that he has entirely independent support for his necessitation account of what it is to be a law—namely, the simple intuition that science seeks to explain in the sense of seeking underlying conditions on which something depends.

Finally, all this suggests a respect in which Kant’s general approach to particular laws is not obsolete, but can actually combine central philosophical advantages claimed by rival contemporary accounts. On the one hand, Kant’s account of what it is to be a law can accommodate intuitions about explanation and dependence—and this is a central advantage claimed by recently necessitation accounts over best system accounts.[50] On the other hand, I have argued that we should not try assimilate Kant’s position to contemporary necessitation accounts, because Kant’s position is shaped by his distinctive and demanding restrictions on our knowledge. This epistemological point is not a slight difference in tone, for it is crucial to the philosophical advantages that Kant himself seeks in accounts throughout his theoretical philosophy. Kant seeks, we have seen, epistemological modesty. And opponents of contemporary necessitation views still today claim the advantage of such modesty: for example, Kitcher says that ‘notions of natural necessity, causation, and objective dependency have worried scrupulous epistemologists at least since Hume’ (1986: 203). If adherence to such scruples is a philosophical advantage, then Kant’s position on laws can claim much of this advantage as well, insofar as he so sharply restricts our access to knowledge of natural necessity. Of course, what makes possible this unusual combination of benefits is Kant’s willingness to draw the conclusion that we cannot generally achieve knowledge of the laws of nature. Perhaps Kant would argue that this is a philosophically attractive form of epistemological modesty, even if some today might see this conclusion as a philosophically costly form of skepticism. But what the precise costs of Kant’s approach are, and whether or not they might be worth paying given the unusual combination of benefits—these seem to me to be philosophical questions worth pursuing. So I do not see any reason in the concerns raised by Kitcher for thinking that Kant’s necessitation approach to the laws of nature is philosophically obsolete.[51]

In sum, then, instead of trying to assimilate Kant’s position to a recently popular account of the laws of nature, we should recognize what is surprising and unusual about Kant’s view—his combination of a necessitation account of what it is to be a law of nature with the claim that we cannot in general achieve knowledge of the particular laws. This position is philosophically superior to the hybrid view resulting from attempts to marry Kant’s philosophy to contemporary best system accounts of laws, even if judged in terms of the philosophical considerations emphasized by best system interpreters such as Kitcher. And this is the position Kant consistently maintains, even in the texts emphasized by best system interpreters, from the first Critique to the third.



James Kreines

Claremont McKenna College

850 Columbia Avenue

Claremont, CA 91711


Primary Texts/Abbreviations

I use the translations of Kant’s works in the Cambridge Edition of the Works of Immanuel Kant, citing Ak. vol:page except with the standard A/B references to the first Critique. Abbreviations are as follows: A/B=Critique of Pure Reason (Guyer-Wood translation 1998); Ak=Deutsche Akademie der Wissenschaften edition of Kants gesammelte Schriften (Berlin: Walter de Gruyter, 1902ff). ; KpV=Critique of Practical Reason. MAdN=Metaphysical Foundations of Natural Science (Friedman 2002); KU=Critique of the Power of Judgment (Guyer-Mathews translation 2000); LM=Lectures on Metaphysics; NF=Notes and Fragments (Bowman et al. translation 2005); P=‘Prolegomena…’ and UE=‘On a Discovery…’ (from Theoretical Philosophical after…); TP= ‘On a Recently Prominent Tone…’ (Heath translation 2002 in Theoretical Philosophical after…).

Other Works Cited

Allais, L. (2003), ‘Kant's transcendental idealism and contemporary anti-realism’, International Journal of Philosophical Studies, 11, 4 369-392.

Allison, H. E. (1996), Idealism and Freedom. Cambridge: CUP.

Allison, H. E. (2004), Kant's Transcendental Idealism Revised and Enlarged Edition. New Haven: Yale University Press.

Ameriks, K. (1990), ‘Kant, Fichte, and Short Arguments to Idealism’, Archiv für Geschichte der Philosophie, 72,  63-85.

Ameriks, K. (2003) Interpreting Kant’s Critiques (Oxford: Clarendon Press).

Armstrong, D. M. (1983), What Is A Law of Nature? Cambridge: CUP.

Beebee, H. (2000), ‘The Non-Governing Conception of Laws of Nature’, Philosophy and Phenomenological Research 81:571-94.

Brittan, G. (1978), Kant's Theory of Science. Princeton, Princeton University Press.

Buchdahl, G. (1965), ‘Causality, Causal Law and Scientific Theory in the Philosophy of Kant’, Brit. J. Phil. Sci., 16, 187-208.

Buchdahl, G. (1969), ‘The Kantian “Dynamic of Reason”, with special reference to the place of causality in Kant’s system’, in L. W. Beck (ed.), Kant Studies Today. La Salle, Ill.: Open Court, 341–74.

Buchdahl. G. (1971), ‘The Conception of Lawlikeness in Kant's Philosophy Of Science’, Synthese, 23: 24-46.

Dretske, F. (1977), ‘Laws of Nature’, Philosophy of Science 44:  248–68.

Friedman, M. (1992a), ‘Causal Laws and Natural Science’m In The Cambridge Companion to Kant. Paul  Guyer (ed.),  Cambridge: CUP.

Friedman, M. (1992b), Kant and the Exact Sciences. Cambridge: Harvard University Press.

Friedman, M. (2001), ‘Matter and motion in the Metaphysical Foundations and the first Critique: The Empirical Concept of Matter and the Categories’, In Kant and the Sciences. Ed. E. Watkins. 53-69. (New York: Oxford University Press).

Ginsborg, H. (1990), The Role of Taste in Kant's Theory of Cognition. New York: Garland Publishing.

Gloy, K. (1976), Die Kantische Theorie der Naturwissenschaft. Berlin.

Grier, M. (2001), Kant's Doctrine of Transcendental Illusion, Cambridge: Cambridge University Press.

Guyer, P. (1990), ‘Kant's conception of empirical law’ in Proceedings of the Aristotelian Society, Supplementary Volume 64:  220-242.

Hanna, R. (1998), ‘A Kantian Critique of Scientific Essentialism’, Philosophy and Phenomenological Research 58 (3), 497-528.

Janiak, A. (unpublished), ‘Nonsense and things in themselves: Kant on logical and real meaning’.

Kim, J. (1994), ‘Explanatory Knowledge and Metaphysical Dependence’, Philosophical Issues 5: 51-69.

Kitcher, Ph. (1986), ‘Projecting the order of nature’ in Kant’s Philosophy of Physical Science, ed. R.E. Butts. Dordrecht: Reidel.

Kitcher, Ph. (1994), ‘The Unity of Science and the Unity of Nature’, Kant’s Epistemology and Theory of Science, P. Parrini ed. Dordrecht, Kluwer.

Kitcher, Ph. (1995), ‘Author’s Response’, Philosophy and Phenomenological Research, 55 (3): 653– 673.

Kitcher, Philip. (1996), ‘Aprioristic Yearnings’, Erkenntnis 44: 397-416.

Langton, R. (1998), Kantian Humility. Our Ignorance of Things in Themselves, Oxford University Press, Oxford.

Loewer, B. (1997), ‘Humean Supervenience’, Philosophical Topics 24: 101-126.

Longuenesse, B. (2005), Kant on the Human Standpoint. Cambridge: Cambridge University Press.

Nayak, A. C. and Sotnak, E. (1995), ‘Kant on the Impossibility of the “Soft Sciences”’, Philosophy and Phenomenological Research 55: 133-151.

Okruhlik, K. (1986), ‘Kant on realism and methodology’ in Kant’s philosophy of physical science, ed. R.E. Butts. (Dordrecht: Reidel).

Plaass, P. (1965), Kants Theorie der Naturwissenschaft. Göttingen: Vandenhoeck and Ruprecht.

Putnam, H. (1981), Reason, Truth, and History, Cambridge University Press.

Putnam, H. (1987), The Many Faces of Realism. La Salle, Ill.: Open Court.

Schaffer, J. (2007), ‘Causation and Laws of Nature: Reductionism’, in Contemporary Debates in Metaphysics, eds. Hawthorne, Sider, and Zimmerman (2007): Basil Blackwell.

Strawson, G. (2000), 'David Hume: Objects and Power', in The New Hume Debate, (ed.) R. Read & K. Richman, London: Routledge.

Warren, D. (2001), Reality and Impenetrability in Kant's Philosophy of Nature, London: Routledge.

Watkins, E. (2004), ‘Kant’s Model of Causality: Causal Powers, Laws, and Kant’s Reply to Hume’, Journal of the History of Philosophy: 449-488

Watkins, E. (2005), Kant and the Metaphysics of Causality (New York: Cambridge University Press)

Winkler, K. (2000), ‘The New Hume’, in The New Hume Debate, (ed.) R. Read & K. Richman, London: Routledge.

Wood, A. W. (1984), ‘Kant’s Compatibilism’, in Self and Nature in Kant’s Philosophy. (ed.) Allen W. Wood. Ithaca: Cornell University Press, 73–101.



[1] I want to thank first of all Gordon Brittan for extensive and extremely helpful comments. I also thank participants in the Pacific study group of North American Kant Society at UCLA and the Yale Philosophy Department work in progress colloquium. I also thank for comments and discussion Andrew Chignell, Troy Cross, Michael Della Rocca, Michael Friedman, Des Hogan, Paul Hurley, Andrew Janiak, Houston Smit, Matthew Smith, Peter Thielke, and Ken Winkler.

[2] Kitcher (1986: 209). In Buchdahl’s influential version, where there is a particular causal law, Kant denies that ‘A, qua A, is necessarily connected with B’ (1969: p. 359 n. 48).

[3] Buchdahl: ‘the conception of lawlike necessity is … defined by the character of the theoretical system’ (1969: 342). Similarly in Brittan: ‘it is not that laws are necessary but that they must be regarded as necessary … insofar as they constitute the representation of the systematic unity of nature’ (1978: 185).

[4] ‘To the logical principle of genera which postulates identity, there is opposed another, namely that of species, which needs manifoldness and variety’ (A654/B682)

[5] Kitcher 1986, 209; 1994, 254; Freidman 1992a, 191. For a more detailed account of Kant’s discussion of the principles here, see Guyer 1990, 23-27. Sometimes Kant also speaks of a combined principle which itself articulates that ideal—the ‘law of the affinity of all concepts’ or the ‘continuity of forms’.

[6] Kant denies that we can have knowledge of any final or complete systematic unity of nature (A661/B669; A644/B672). And the principle of affinity, discussed in the previous note, is ‘legitimate and excellent regulative principle’ (A668/B696).

[7] The claim that empirical laws can be known by empirical induction is, Friedman argues, part of the ‘virtually unanimous opinion’ ‘at least in the English-speaking world’ (1992a, 165 and 193 note #7).

[8] With respect to the connection between being conditioned and being dependent, see for example Kant’s formulation of the question of ‘whether something is conditioned and hence externally dependent’ (A481/B509).

[9] Reason ultimately demands that we seek knowledge of a complete series of conditions, or knowledge of something unconditioned (A307/B364). In terms of explanation, there is a ‘rational prescription to seek the ultimate explanation’ (Grier 2001, 2). Or, reason demands that one ‘never stop seeking conditions until one gets them all, never rest satisfied with an explanation that leaves something unexplained’ (Allison 2004, 31).

[10] Compare contemporary arguments that regularity accounts provide laws that cannot explain, see Dretske 1977, 262; see also Armstrong 1983, 40. And see Watkins account of these contemporary arguments and their relevance to Kant 2005: 42.

[11] See also Watkins: ‘Kant holds that the laws of nature are based on the natures of substances, and that substances must act in accordance with those natures’, and his link between this view and a conception on which laws ‘govern’ events (2005: 18).

[12] Compare also Reflexion 5215: ‘That which is the condition under which we posit something in accordance with a rule is the cause. If I arrange for a stork to fly in wintertime, it does not become warm. Hence this is not the cause’ (NF p. 221; Ak. 18:120). And Watkins on accidental generalizations (2005, 346). Section 6 below argues that best system accounts cannot succeed in their attempt to distinguish laws from accidental generalizations (Brittan 1978, 183; Kitcher 1986, 221) in any manner accommodating the simple intuition about explanation and dependence. Also note that the necessitation supposed to distinguish laws is not the epistemic necessity involved in an inference from a general rule (‘all F’s are G’s’) in a particular case (‘this is an F’) to a conclusion (‘this is a G’). For such inference follows just as certainly in any case of a true generalization, even if there is no law of nature and it is only accidentally true that all F’s are G’s.

[13] Watkins (2005: 18, 401-7); see also Langton (1998, 118-20).

[14] So perhaps particular laws ‘are not true in all worlds which we, given our, cognitive faculties, could experience’ (Kitcher 1994: 265; see also Buchdahl 1965, 201). But this is no reason to prefer a best system account. See also Hanna on Kant on the contingency of governing laws (1998: 524). Note that Kant also discusses contingency with respect to laws in discussing whether it is only contingently true that nature is such that we can make progress toward knowledge of laws; on this topic see especially Ginsborg (1990: 175ff.)

[15] I am especially influenced by this take on causal laws in Friedman: ‘if event A causes event B, we know that this relation is universal: Events of the same kind as A are necessarily followed by, or result in, events of the same kind as B’ (1992a, 163).

[16] ‘Kant retains many of the central features of his cosmology in the Critical period’ (Watkins 2005: 182). The views I have attributed to Kant so far involve no assertion of knowledge of necessitation-laws—not even knowledge the sort of necessitation-laws that we seek really exist, nor knowledge of an ‘ontological framework’ of laws not yet filled with content; compare Watkins (2005: 290). 

[17] We might have ‘cognition’ of an object while lacking knowledge of it, in the sense that ‘cognition is false if it does not agree with the object to which it is related’ (A58/B83). But we cannot extend our theoretical knowledge beyond the bounds within which we can have this relation to an object involved in ‘cognition’. For example, Kant insists that metaphysicians should acknowledge ‘that it is not allowed them even once to guess [mutmaßen], let alone to know [wissen], something about that which lies beyond the boundaries of all possible experience, but only to assume something about it … as is possible and even indispensable for the guidance of the understanding and will in life’ (P 4:278).

[18] There is ‘only one way possible for my intuition to precede the actuality of the object and occur as an a priori cognition, namely if it contains nothing else except the form of sensibility, which in me as subject precedes all actual impressions through which I am affected by objects’ (P 4:282).

[19] Even if we have an innate concept of such a being, theoretical knowledge of whether such a being actually exists, as synthetic judgments about existence would require intuition that we lack (A597/B625).

[20] On the synthetic status of the principles of the understanding, and the connection to the a priori intuition of time, see Allison (2004: 225-8).  

[21] Interpreters on different sides in recent debates tend to point to the passage at B165: see Buchdahl (1965: 190 and 1969, 357), Kitcher (1994: 258) and Friedman’s response (1992a, 166). For other passages see for example A216/B263, A207/B252; MAdN 4:468; and the discussion of MAdN below.

[22] E.g. Armstrong 1983: 99-107. Specifically: ‘if there is no good reason to move beyond the regularities of the world to laws which are more than regularities, then there is equally no good reason for him to move beyond his present observations and/or sensations to a regular world’ (1983: 106).

[23] Note that this restriction also means that we cannot infer from contingent and dependent things to knowledge of an absolutely necessary ground on which everything is supposed to depend: pre-critical cosmological arguments misuse ‘the transcendental principle of inferring from the contingent to a cause, which has significance only in the world of sense’ (A609/B637). Armstrong notes that his view threatens to support an inference to a metaphysically necessary ground explaining why the laws of nature are as they are: on the face of it the inference should “enlist the sympathy of anybody who, like myself, looks to an account of laws which treats them as the explanations of regularities” (1983: 159).

[24] I much influenced here by Hanna’s comparison of Kant’s ‘skeptical worry about the limitations of empirical scientific inquiry’ (1998: 511), though I would not say that the limitation concerns inaccessible ‘physical microstructures’ (1998: 509) since this could suggest the possibility of knowledge with sense organs able to detect smaller spatio-temporal objects. And I am much indebted here to Ginsborg’s account of the progress of our empirical inquiry, though instead of saying that the sciences seeks more fundamental ‘properties which are in principle discoverable’ (1990: 187) I would emphasize that we generally cannot in principle know when we have reached kinds governed by laws of nature. I am also indebted to Longuenesse on the role of our guiding principles (2005: 175); to Warren (2001: 55) on the explanatory ‘regress’; and to Wood on Kant’s Aristotelian way of conceiving or thinking about the unknowable ‘intelligible world’ (1984: 88).

[25] Issues about the relations between the earlier and later accounts require and deserve separate attention; see Guyer (1990) on the differences, and especially Ginsborg (1990) on the Introductions to the KU. 

[26] On this contrast between ‘laws’ and ‘rules’: ‘One can very well produce rules empirically, but not laws (Reflexion 5414; NF 230; Ak 18:176). And see the discussion in Gloy (1976: 19ff.).

[27] Allison 1996, 91. Allison does suggest a less extreme version of a best system view, on which a place in the most unified system of empirical knowledge is a necessary but not sufficient condition for something being a law. But the passage gives the principle a role in guiding inquiry, not in grounding necessity. And Kant’s view is that laws of nature are not regularities, so they cannot be regularities that meet some combination of conditions; to state a law is not to refer to many instances but rather to the nature of a kind.

[28] Kitcher does note Kant’s ‘dark utterances about how such laws are ‘approximated to universality’’ and reads these passages as claiming that, ‘because the practice within which the ‘laws’ are established might itself be overthrown, we cannot speak of proving those ‘laws’’ (1986: 230). But this way of explaining how a best system account could be made to cohere with Kant’s denial of the possibility of empirical knowledge of laws undercuts the interpretive case for the claim that any of Kant’s texts suggest a best system account in the first place. For the interpretive case turns on construing Kant as holding that our guiding principles do allow empirical knowledge of particular laws.

[29] I am much influenced by Hanna’s (1998) comparison with Locke’s denial of knowledge of real essence. With respect to Hume, my criticisms of best system interpreters are similar to worries about the ‘standard’ view of Hume pressed by proponents of the ‘new’ Hume—see e.g. Strawson on the confusion of the epistemological and the ontological (2000, 33); but there are also similarities between Kant’s view and Winkler’s version of a ‘standard’ Hume, insofar as Kant refrains from asserting knowledge that there are unknowable particular laws of nature (2000, 53).

[30] Friedman 1992a, 175ff. Just some of the literature covering Kant’s attempt: Plaass (1965), Gloy (1976) and Friedman’s (2001).

[31]‘Since in any doctrine of nature there is only as much proper science as there is a priori knowledge therein, a doctrine of nature will contain only as much proper science as there is mathematics capable of application there’ (MAdN 4:470). 

[32] Friedman argues that Kant’s very late unpublished work claims that chemistry might allow derivations of laws because it involves the single basic kind of ‘aether’ or ‘caloric’ (1992b, 220ff.). I address here only the period from the first Critique to the third.

[33] MAdN 4:470-1. Here I agree with the reading of this point in terms of reduction in Nayak and Sotnak (1995: 147-150). Also on MAdN on chemistry see Gloy (1976, 186).

[34] In the first Critique, see for example A49/B72, A252/B309, A279/B335.

[35] As Kant puts the point elsewhere: ‘The rules of uniform appearances are called laws of nature (e.g., mechanical laws) only when they are either cognized really a priori or (as in the case of chemical laws) when it is assumed (annimmt) that they would be cognized a priori from objective grounds if our insight went deeper’ (KpV 5:26). Note that the principles of our understanding are principles of our understanding. An intellect not limited by the a priori forms of our sensibility, including time, would not be an intellect not requiring principles of a priori time determination, or the principles of the Analogies. See also Allison on the connection between the principles, the schemata, and a priori intuitions of space and time (2004: 225ff.).

[36] Kant says that we are guided by an assumption about laws we cannot know at (Ak. 20:206) in the KU first Introduction. He also refers to an assumption about laws in a Reflexion: ‘One can very well produce rules empirically, but not laws, as Kepler did in comparison to Newton; for to the latter there belongs necessity, hence that they can be cognized a priori. Yet one always assumes of the rules of nature that they are necessary, because it is on this account that it is nature, and that there can be a priori insight into them’ (Reflexion 5414; NF 230; Ak 18:176).

[37] Kitcher 1984: 214 and especially 1994: 262-3 and 271 note 20. See also similar considerations but without the comparison in Buchdahl 1969: 352. For more detailed comparison to ‘internal realism’ see also Okruhlik 1986, 322 ff.

[38] Ameriks has long argued for the importance of this point (1990). Allison acknowledges Ameriks’ criticism on this score, and accepts the point: ‘we can only know objects as they appear because we can know them only insofar as they are given in sensible intuition’ (1996: 7).  

[39] Bxxvi note. Some ideas of unknowable objects generate contradictions. But it is crucial to Kant’s project, as he emphasizes, that this claim is restricted to the ideas of the spatio-temporal world as a completed, unconditioned whole: ‘The first two antinomies … are founded on such a contradictory concept’ (P 4:341). That claim does not apply to the ideas central to the second two antinomies. Nor does it apply to psychological and theological ideas (A673/B701).

[40] Also cited in Allais’ (2003) argument that legitimate comparisons with anti-realism can apply only specifically to Kant’s account of appearances, and not more broadly.

[41] See especially Adams (1997: 807-8); also Clark (1985) and Janiak (unpublished).  Kant does say that merely ‘intelligible objects … without any schema of sensibility’ are ‘impossible’ (A286/B342). But even right here he continues to introduce a sense of noumenon that is ‘problematic’ in this sense: it is a ‘representation of a thing of which we can say neither that it is possible nor that it is impossible’ (A286-7/B343).  

[42] I side with Langton in her criticism, along these same lines, Allison’s epistemological or methodological reading of transcendental idealism (1998: 10).

[43] I also side with Allison in his criticism, along these lines, of Langton: while Kant sometimes characterizes “the concept of a thing in itself” as if things in themselves would be ‘Leibnizian monads (substances with intrinsic properties)’, still ‘it hardly follows that Kant remained committed to the substantive metaphysical view that reality is composed of such substances’ (Allison 2004, 10).

[44] ‘For the explanation of given appearances no other things and grounds of explanation can be adduced than those which are connected to the given appearances by already known laws of appearances’ (A772/B800). I thank Gordon Brittan for pressing me concerning this passage.

[45] Note that Jaegwon Kim attributes to Kitcher an ‘epistemic conception of explanation’ or ‘explanatory internalism’ (1994: 63 and 57).

[46] For example, our guiding principles are not merely convenient heuristic principles (A653/B681; Kitcher 1986, 213) and without them we would lack a criterion of empirical truth (A651/B679; Kitcher 1986, 213; 1996, 412). But we must not confuse these points with a retreat from the claim that these principles are regulative. See for example Allison (2004: 431ff.) on how such principles can have a transcendental status while remaining regulative.

[47] See van Fraassen’s worry about Lewis’ best system account: ‘‘the evolution of science as a whole is historically conditioned by its starting-point’ (1990: 59).

[48] For defenses of regularity accounts that recognize the conflict with intuitions about governing laws, seeking to attack that intuition, see Loewer (1997), Beebee (2000) and Schaffer (2007). That best-system-laws do not govern is also noted in Watkins’ interpretation of Kant (2005: 402).

[49] Can one hold that best-system-laws legislate for unactualized possibilities in the sense that they would hold in the nearest possible worlds? As Forrest points out, this would be circular, for ‘it is only if we think of Ramsey-Lewis regularities as (manifestations of) necessary laws that we shall insist that they hold in all nearby worlds’ (1985: 5). Lewis does try to offer independent reason for thinking of laws as holding in nearby possible worlds: we hold the laws to be important. See Forrest’s further rejoinder (1985: 5) and Armstrong’s: if we happened to find the laws of nature less unimportant then there would be no sense in which they would hold in nearby worlds (1983: 69).

[50] For example: ‘the fact that every F is G fails to explain why any F is G’ (Dretske 1977, 262). And see Armstrong’s appeal to explanation (1983: pp. 40-1), against best system accounts specifically see 66ff.

[51] I do not discuss difficulties concerning probabilistic laws here. Best system accounts face such difficulties as well, and the best system interpretations of Kant discussed here do not claim advantage in this regard.