2. The Problem

Kant's project in G is to provide ‘a supreme principle of morality’ (4: 392).Footnote 1 The various formulations of the categorical imperative are different ways of specifying this supreme principle. Here are the first two formulations of the categorical imperative:

1. 1. Formula of universal law (FUL): Act only according to that maxim whereby you can at the same time will that it should become universal law. (4: 421)

2. 2. Formula of the law of nature (FLN): Act as if the maxim of your action were to become through your will a universal law of nature. (4: 421)

In G, Kant's definition of a ‘maxim’ is very brief, found in a note at 4: 420–1: ‘A maxim is the subjective principle of acting and must be distinguished from the objective principle, viz., the practical law. A maxim contains the practical rule which reason determines in accordance with the conditions of the subject (often his ignorance or his inclinations) and is thus the principle according to which the subject does act.’ Just how to understand a Kantian maxim is a matter of persistent controversy, to which I will return. For now I shall follow the definition and examples found in G.

The way to use FUL and FLN is to take one's subjective principle of volition and to determine whether, when generalized, it yields any contradictions.Footnote 2 There are two sorts of contradiction to test for. If you cannot even think of your maxim as a universal law – i.e. if it yields a conceptual contradiction – then this principle of willing is contrary to a perfect duty. Kant's most famous example is the making of false promises (4: 422).Footnote 3 If you can consistently think of your maxim as a general law of nature, but you could not will it to become such a law, then your principle of willing is contrary to an imperfect duty. Kant's most famous example of this concerns charity.Footnote 4 Rather than a conceptual contradiction, here the person running the generalization test encounters a practical contradiction.

The problem of maxim-fiddling arises when we pay further attention to the nature and source of the maxims which are to be tested via FLN and FUL. In G, Kant offers us little guidance on these matters. His examples of subjective principles of volition are uniformly neither so general as to abstract away from all details of the particular case nor so particular as to apply just to the agent's own action and not to others.Footnote 5 For instance, this is his characterization of the maxim to be examined in the false promise case: ‘The maxim of his action would then be expressed as follows: when I believe myself to be in need of money, I will borrow money and promise to pay it back, although I know that I never can do so’ (4: 422). Moreover, he seems to assume that all actions are produced in accordance with such principles, but he does not provide a method for identifying or formulating them. All of this suggests that agents can play a legitimate role in designing maxims, and this gives rise to the maxim-fiddling problem. The natural context for this problem is one in which a maxim has already failed the test. Why not reformulate your maxim so that it can be universalized, thereby, despite the first result, rendering your action permissible by the standards of CI? However, such post-failure maxim revision is inessential to the issue. Once one realizes that careful formulation will guarantee that one's maxim will pass the universalization test, there is no need ever to fail it in the first place. This is the problem of maxim-fiddling. This objection has been made by various commentators in various ways. Alisdair MacIntyre, in his widely read A Short History of Ethics, presses this objection in a few sentences:

Fred Feldman presses the same objection against Kant, with more attention to the nature of maxims of action. Feldman proposes that we understand maxims as describing first a general sort of situation and then a form of action for that kind of situation (Feldman Reference Feldman1978: 99). Accordingly, this is the form of particular principles of volition:

Whenever I am ––––, I shall ––––. (Feldman Reference Feldman1978: 101)

Feldman also provides the generalized form of maxims, to be tested via FLN and FUL:

Whenever anyone is ––––, she will ––––. (Feldman Reference Feldman1978: 102)

After criticizing Kant's own examples (106–14), Feldman evaluates the CI test with some of his own. He makes the maxim-fiddling objection with an example of academic dishonesty. Consider a student who buys a term paper to submit for a course. Using Feldman's schema, here is the maxim:

When I need a term paper for a course and I do not feel like writing one, I shall buy a term paper and submit it as my own work. (Feldman Reference Feldman1978: 115)

The generalized maxim is, accordingly:

Whenever anyone needs a term paper for a course and does not feel like writing one, she will buy one and submit it as her own work. (Feldman Reference Feldman1978: 115)

This fails the CI test straightforwardly: instructors would cease to require term papers and students would be evaluated in other ways. This would deprive the current student of her opportunity for academic dishonesty. However, according to Feldman, it is also easy to modify the maxim so that cheating in this way passes the CI test. The maxims need one small addition: a clause specifying that no change in the system will occur by submitting a purchased paper:

When anyone needs a term paper for a course and does not feel like writing one, and no change in the system will occur by submitting a store-bought one, she will buy one and submit it as her own work. (Feldman Reference Feldman1978: 116; emphasis added)

After ‘fairly trivial’ (Feldman Reference Feldman1978: 115) fiddling, this generalized maxim will pass the CI universalization test. The academic dishonesty in question, when performed in accordance with this principle of volition, appears to be permissible by the standards of G, despite the result of the first application of the universalization test.

In sum, the maxim-fiddling problem is this: the CI universalization test is designed to distinguish between morally permissible and impermissible actions. But when run in a particular way, it fails to do this. This means that this test fails by its own standards. The scope of this problem is very wide: it applies to every application of the CI universalization test for permissible maxims. The implication is that a maxim that will pass the CI universalization test can be found for each and every action. By extension, this implies that, if this objection is sound, no maxim, and hence no action, need fail the universalization test: absolutely everything can be shown to be permissible by the standards of this test. This would undermine the Kantian project of specifying the supreme principle of morality. To vindicate the Kantian project, a Kantian response to the problem of maxim-fiddling is required.

3. A New Kantian Response to Maxim-Fiddling

This section contains a response to the problem of maxim-fiddling. The demonstration that it is a new one must wait until the next section, which surveys extant approaches to this problem. I shall start with some preliminaries.

First, my aim is to provide a ‘Kantian’ solution to this problem. Just what it is for a position to be ‘Kantian’ deserves attention. I mean three things by this. First, and negatively, by this term I mean to signal that I am not representing this solution as Kant's own. Kant did not address the problem of maxim-fiddling.Footnote 6 As we shall see, nor did he explicitly provide the sorts of tools that commentators have generally thought to be necessary to solve this problem. So, whatever can be said of the solution suggested here, it is unwarranted to see it as Kant's. The second thing I mean by ‘Kantian’ is that the position here is built upon a foundation of famous ideas from Kant found in, arguably, his most important ethical work – G. Specifically, I argue that the maxim-fiddling problem can be solved by reflexive application of the CI universalizability test. This test is a distinctively Kantian contribution to ethics if anything is, so it is warranted to see anything that uses this as its foundation as Kantian. Put together, these two points constitute a project aimed at offering something that is not Kant's but that is in the general spirit of Kant's ethics.Footnote 7 These points also indicate the standards of success and failure for such a project. Insofar as I am offering something that is in the spirit of Kant, the project must be tied to Kant's texts. But insofar as what is offered is not to be regarded as Kant's own, creative construction beyond textual exegesis is at work. The project is successful if it fits into a recognizably Kantian space of ideas even if it is neither Kant's own nor independently plausible. This brings us to the third and most weak condition of ‘Kantian’ positions. I hope that the position developed here is Kantian in the sense of contradicting no other of Kant's distinctive contributions to ethics. I take seriously, however, the possibility that this might be an impossible constraint to meet. The reason is that Kant's own overall position on ethics might contain contradictions. Arnulf Zweig contends that, rather than Kant, commentators find in his ethics various ‘Kants’, and that the diverse interpretations made of Kant's ethics do not entirely rest on misinterpretations (Reference Zweig2009: 255). If this is correct, then it will be no surprise if a position, such as the present one, that draws on a specific Kantian position turns out to conflict with some other distinctive Kantian position.

This point brings us to the second preliminary issue. My response to the problem of maxim-fiddling draws solely from the resources of G. Specifically, I follow Kant's vague definition of maxim as a principle of volition and his examples as widely as possible: anything that might possibly count as a subjective principle of volition will, for the purposes of argument, be treated as a possible maxim of action. Given this interpretation of ‘maxim’, I will argue that there are, nonetheless, G resources for resisting the problem of maxim-fiddling. This deserves some explanation. Are there good reasons for restricting our focus to this part of the Kantian corpus? I shall present one reason for this approach in this section. Three more reasons are added in the next section, in connection with the survey of extant approaches to this problem.

The first reason is that the present approach matches the grounds on which the maxim-fiddling objection is typically made. Such objectors as MacIntyre and Feldman make this objection on the basis of G alone. If a response to the problem of maxim-fiddling can be provided using G alone, it will all at once be a Kantian response that cuts off the problem of maxim-fiddling on exactly the same grounds that typically give rise to it. Indeed, to address the problem of maxim-fiddling with as wide an understanding of ‘maxim’ as possible is to address the problem in the form most favourable to the objectors. To defeat the objection in its strongest form is to answer the problem in as close to a decisive way as possible.

As we shall see in section 4, the standard Kantian responses to the problem of maxim-fiddling seek a Kantian definition of ‘maxim’. In lieu of this, a Kantian response to this problem can be formulated by examining the CI universalizability test itself. The purported problem is that the CI fails by its own standards. This point is reached by presuming that one can legitimately design a maxim to pass the test. This presumption deserves scrutiny. If maxim-fiddling is not a legitimate way to run the CI universalization test, then the objection does not get off the ground.

The crucial idea that I shall use to argue that maxim-fiddling is not a legitimate way to run the CI universalization test is this: running the CI test is itself an action. It is something that we do, it can be done intentionally, to achieve one or more ends, just like preparing coffee or making a false promise. This means that the running of the CI test is itself subject to the CI universalization test. Presumably Kant did not intend us to run the CI test before every intentional action, as this would be both practically and logically impossible. It is practically impossible because we are incessantly doing things intentionally; running the CI test all the time would paralyse us. It is logically impossible since running the CI test is something we do intentionally. So, to run the test, we would first have to run the test, and to do this we would first have to run the test about this running of test, ad infinitum. But Kant does hold that the FUL underlies the moral evaluation of all of our actions: ‘We must be able to will that a maxim of our action become a universal law: this is the canon for morally estimating any of our actions’ (4: 424). Since running the CI test is an action, our maxims in running this test are subject to assessment via FUL and FLN just like any other maxims. This is what offers the possibility of a new Kantian, G-based response to the problem of maxim-fiddling.

I shall use Feldman's schematic representation of maxims, given that he is one of the objectors in question. The specific form is

Whenever I am ––––, I shall ––––. (Feldman Reference Feldman1978: 101)

The present situation concerns someone who is considering running the test with a maxim that is designed specifically to pass this test. Accordingly, here is Feldman's schema filled out for this case:

Whenever I am in the position of running the CI test, I shall use a maxim designed specifically to pass the universalizability test for moral permissibility.

Let us call this the maxim-fiddling maxim. Recall that Feldman also offers the generalized form of maxims, to be subject to the CI test:

Whenever anyone is ––––, she will ––––. (Feldman Reference Feldman1978: 102)

Filled out, here is the appropriate maxim to test via FLN and FUL:

Whenever anyone is in the position of running the CI test, she will use a maxim designed specifically to pass the universalizability test for moral permissibility.

Let us call this the generalized maxim-fiddling maxim. In order for maxim-fiddling to be permissible by Kantian standards, the generalized maxim-fiddling maxim must pass the CI universalizability test. When we attempt to conceive of this maxim as a universal law of nature, do we encounter any conceptual or practical contradictions? Showing that a particular maxim generates contradictions when subjected to the CI test is notoriously difficult, so I will not attempt to provide a knock-down argument. Instead, here are two prima facie arguments. The first shows that there is reason to think that a conceptual contradiction is encountered when we run the CI universalizability test with the generalized maxim-fiddling maxim. The second shows that, if we disregard the conceptual contradiction and proceed as if we could coherently conceive of the generalized maxim-fiddling maxim as a universal law of nature, we nevertheless encounter a practical contradiction.

Let us start with the case for a conceptual contradiction. To think of a maxim as a universal law of nature, one must think of it as a structuring law of the world, like the physical laws describing the speed of light and the force of acceleration due to gravity. In the case of maxims of actions, to do so means two things:

1. (a) to think of everyone acting in the requisite way whenever the occasion arises, just as objects fall in accordance with the law of gravity whenever the occasion arises;

2. (b) to think of this pattern of action being in place from the beginning of time, just like laws of nature.

This cannot be done with the maxim-fiddling maxim. The concepts of moral permissibility and impermissibility are internal to the institution of morality. Morality is a complex social institution that has developed and persisted over time.Footnote 8 To be able to think with moral concepts, one must be part of this institution. For this institution to develop and persist, people must be able to make principled use of the concepts of moral permissibility and impermissibility. This requires that not all ways of acting be permissible; some must be impermissible. Otherwise there is no job for the distinction between moral permissibility and impermissibility to do, and the distinction will disappear. If the generalized maxim-fiddling maxim were a universal law of nature, then all ways of acting would be permissible. This would undermine the role of the distinction between moral permissibility and impermissibility, and it would disappear. There would now be no such distinction – perhaps not even an institution of morality – and it would be impossible to formulate the present maxim. This is just to say that a conceptual contradiction is, prima facie, encountered when we subject the generalized maxim-fiddling maxim to the CI universalizability test.

We can add to this argument. The generalized maxim-fiddling maxim requires that we are familiar with the CI universalization test. The point of the CI universalizability test is to distinguish between permissible and impermissible actions. This requires that it be possible for an action to turn out to be impermissible through running the test. If the generalized maxim-fiddling maxim were a universal law of nature, this would be impossible. If we knew that any action would ultimately, with ingenuity, pass the CI universalizability test, we would stop running the test. This would undermine our current familiarity with it. The formulation of the maxim-fiddling maxim requires the concept of the CI test, and its universalization simultaneously undermines the possibility of us having this concept. That is to say, in making the generalized maxim-fiddling maxim into a universal law we encounter, prima facie, a conceptual contradiction.

It is important to note that this is a prima facie argument. Just how persuasive it is will depend on whether the possibility of thinking the relevant thoughts depends on the development and persistence of particular social ways of behaving. If it does, then this sort of Kantian argument can be persuasively constructed. If the possibility of possessing such concepts is not tied so closely to such social forms of behaviour, then it is much less likely that such a Kantian argument can be constructed.

Let us turn to the prima facie case that a practical contradiction is encountered here. Let us put the first argument aside and assume that we can coherently conceive of the generalized maxim-fiddling maxim as a universal law of nature. Nevertheless, a rational person could not will it to be such a law. Inevitably we must interact with other people. As part of this, we will desire other people not to do certain things to us. Generally, we will want others to respect our rationality. Overall, what this means is that we will want other people to see certain ways of acting as impermissible. In a Kantian spirit, we would wish always to be treated as ends, not merely as means. We would wish actions that treat us merely as means to be thought of as impermissible. If the generalized maxim-fiddling maxim were a universal law of nature, then no actions would be impermissible, as every action would pass the CI universalizability test for permissibility through maxim revision. A rational being would not put itself into this situation: it would not wish for a certain sort of treatment and simultaneously undermine the possibility of receiving it by its own will. This is just to say that we encounter a practical contradiction when we consider whether we could by our will make the generalized maxim-fiddling maxim into a universal law.

To clarify these arguments I shall address some possible worries. First, in G, a conceptual contradiction indicates that your action is contrary to a perfect duty, and a practical contradiction signifies that it is contrary to an imperfect one. Roughly speaking, a perfect duty is one the force of which is completely independent of our inclinations, whereas agents have legitimate elbow room in living up to imperfect duties.Footnote 9 To use the Kantian examples: our feelings have no weight against the prohibition against killing ourselves, but we need only perform acts of charity some of the time, to some extent, and our feelings can have a role in determining the appropriate times and extents. I have just constructed prima facie arguments suggesting that we encounter both conceptual and practical contradictions when subjecting the generalized maxim-fiddling maxim to the CI universalizability test. One might worry that there is something contradictory about this result: it seems to suggest that we must both always and not always respect and perform the duties attached to maxim-fiddling. Let me defuse this worry. First, we must not ignore the condition under which the second argument works: the practical contradiction was generated only after explicitly disregarding the argument for a conceptual contradiction. So the argument is not that we encounter both sorts of contradiction. It is instead that when we run the CI test, we encounter a conceptual contradiction, and that if we disregard this result, we encounter a practical contradiction. Second, and more importantly, the worry indicates a genuine ambiguity about the situation being assessed. Agents must refrain from maxim-fiddling whenever they run the CI test. That is, since every time this test is performed there is a duty to refrain from maxim-fiddling, we should see this as a perfect duty. But agents cannot perform the CI test for everything they do; this would result in practical paralysis. So agents can exercise legitimate discretion in respecting the duty to refrain from maxim-fiddling to exactly the same extent that they have such elbow room in running the CI test at all.

The second possible worry is that the present argument could itself be undermined by maxim-fiddling. Why not formulate one's maxim-fiddling maxim to ensure that it passes the CI universalizability test? This worry stems from a misunderstanding of the structure of the generalized maxim-fiddling maxim. This is a maxim about maxims – i.e. it is a higher order maxim. But it is not a maxim of a specific higher order, nor is it about a specific lower order. It is, for example, neither a second order nor a third order maxim, etc., nor does it apply specifically to first order ones. Instead, it is generally higher order: it applies to any maxim. This also makes it reflexive: it applies to itself. Consequently there is no opportunity for higher order maxim-fiddling that escapes the scope of the present arguments. Revision of the generalized maxim-fiddling maxim is problematic for exactly the same reasons as revision of ordinary first order maxims of action.

Note that the present results do not prohibit maxim revision, nor repeated performance of the CI universalizability test.Footnote 10 The contradictions arise not from maxim revision but from maxim formulation that undermines the possibility of, and thereby the import of, the conceptual and practical distinction between moral permissibility and impermissibility. So long as maxim formulation respects the possibility of these distinctions – i.e. so long as running the CI test preserves the possibility of one's action turning out to be impermissible – no contradictions of the sort sketched in the present prima facie arguments will be encountered. However, other contradictions might be the result, showing one's action, as conceived even through a revised maxim, to be impermissible.

In summary, the new Kantian response to the problem of maxim-fiddling focuses on the running of the CI universalizability test with maxims designed specifically to pass this test as an action to which the CI test applies, just like any other action. When we subject the appropriate maxim to assessment via FUL and FLN, we discover conceptual and/or practical contradictions. This signifies that such maxim formulation is impermissible by these familiar Kantian standards. As a consequence, the problem of maxim-fiddling is answered: the CI universalizability test cannot be legitimately undermined through maxim-fiddling because such fiddling itself fails this test. As we shall see in the next section, this is a more decisive response to this problem than standard ones because its foundation is the broad understanding of Kant's moral philosophy that is used to formulate the objection in the first place. It is nonetheless a genuinely Kantian response because it consists in a redeployment of the conceptual resources provided in G.

4. Standard Kantian Responses to Maxim-Fiddling

Generously speaking, there are two standard responses to the problem of maxim-fiddling. First, some commentators dismiss the problem in a few sentences, as if it is a small concern. Second (taking the notion of a ‘response’ broadly), other commentators ignore it entirely. These standard Kantian responses stem from differences within a shared project. The shared project is to provide an account of what a ‘maxim’ is, using the Kantian corpus beyond G. These two responses to the problem of maxim-fiddling tend to line up with different accounts of what a Kantian maxim is.

A useful entry-point to the issue of what a maxim is, both in general and for present purposes, is the well-known ‘action description problem’.Footnote 11 FLN and FUL are about both actions and maxims; Kant's classic examples illustrate the connections between actions and maxims. Famously, actions are describable in many ways, perhaps indefinitely many.Footnote 12 To use a universalization test to assess the permissibility of an action, it seems that one must use the correct action description, or at least an appropriate one. Kant directs us to look at an agent's maxims to describe actions in the appropriate way for moral evaluation. Given the scant discussion of the nature of maxims in G, commentators have sought to construct Kantian notions of ‘maxim’ on a broader textual basis and hence to solve this problem.

It is not automatically the case that a constructed Kantian notion of ‘maxim’ that answers the action description problem will also answer the maxim-fiddling problem. This will depend on the details of the particular definition of ‘maxim’. In general, if a definition of ‘maxim’ allows either multiple maxims for an action or leeway for an agent to consciously participate in maxim formulation, then theoretical space will remain for the maxim-fiddling problem even if guidance has been provided for the action description problem. If instead a definition of ‘maxim’ restricts maxims to one per action and eliminates active agential participation in maxim formation, then such a definition will provide an answer to the maxim-fiddling problem. Whether it is a satisfactory answer must be assessed independently.

Herman identifies two dominant approaches to the definition of ‘maxim’ and hence to the action description problem (1993: 219).Footnote 13 One approach interprets a maxim as a deep principle. These deep principles are not themselves typically directly expressed in our particular intentions to act. Rather, they regulate the formation of particular intentions. O'Neill makes this case with the example of trying to welcome a visitor by preparing a cup of coffee. There are lots of short-term actions performed to do this, and presumably lots of particular intentions, about getting the cream and sugar, choosing from various mugs, etc. O'Neill thinks none of these intentions is itself the agent's maxim. Instead, it is the principle of making a visitor feel welcome. In the absence of coffee, this principle could be realized in a completely different way, with distinct particular intentions in acting. In O'Neill's words, ‘Maxims are those underlying principles or intentions by which we guide and control our more specific intentions’ (1990: 84). Maxims are a sort of intention, but not just any intention is a maxim. As it happens, on this interpretation it is typically much clearer what particular intention(s) an agent had in acting – to get the sugar from the cupboard, to select the good mug, etc. – than it is what maxim the agent acted from. Nevertheless it is this deep principle that we should use to describe the action for moral evaluation.

Jens Timmermann offers a similar interpretation of ‘maxim’. According to Timmermann, maxims are principles which codify an agent's deep ends. The rules which can describe our particular intentions in acting can be very trivially related to an agent's life. By contrast, maxims are rules that describe an agent's commitments and thereby her character (2007: 179–80).

On the basis of this sort of interpretation of ‘maxim’, O'Neill and Timmermann dismiss the maxim-fiddling problem in a few sentences. The spirit of their response is this: when an agent revises her ‘maxims’ in the wake of an unfavourable result from the CI test, she is not really revising her true maxim of action. That is, she is not revising her deep commitments, her character, or her deep intentions. Rather, she is just revising superficial intentions in acting. O'Neill: ‘The fact that we can formulate some universalizable surface intention is no embarrassment to a universality test that is intended to apply to agents’ maxims, and offers a solution to the problem of relevant descriptions’ (1989: 87). Timmermann echoes this thought:

On this view of the nature of maxims, maxim-fiddling is itself impossible. What appears to be maxim revision actually does not affect an agent's maxims. It affects just the agent's superficial intentions, to which the CI universalizability test does not apply.

Briefly, here are two difficulties with this approach to maxim-fiddling. The first is that it ties this problem to a contentious view of the nature of maxims. This will become evident as I discuss two more interpretations of this notion. The second problem is that it risks failing to answer the maxim-fiddling objection. There are two crucial questions here. The first is whether agents can exercise control over their character, their commitments, their deep intentions. By most standards, it seems that they can. The second is whether agents can have idiosyncratic and even very specific commitments. Again, by most standards it seems that they can. The second point implies that agents could adopt such ends as ‘lying to so-and-so in such-and-such a situation’, to follow Timmermann, or as ‘buying a term paper when no change in the system will occur by submitting a store-bought one’, to follow Feldman. The first point implies that agents could exercise control over their adoption of such ends. Together these ideas hold out the possibility of ‘true’ maxim-fiddling, even by the standards of the view of maxims offered by O'Neill and Timmermann. In other words, instead of resolving the problem of maxim-fiddling, O'Neill and Timmermann merely and inadvertently move it from the level of specific intentions in action to the level of ‘true’ maxims.

The second dominant approach to the definition of ‘maxim’ and hence to the action description problem identified by Herman was formulated by O'Neill in earlier work (Reference O'Neill1975). On this view, an agent's maxim is her ‘specific intention in acting’ (Herman Reference Herman1993: 219). O'Neill gives this as the schematic form of a maxim:

I (he, X) ought to (may, deserves to, etc.) do/omit –––– if … (1975: 35)

On this view, every intentional aspect of an agent's action has a maxim.Footnote 14 This means that actions have multiple maxims and descriptions, each of which is relevant to the application of the CI test (O'Neill Reference O'Neill1975: 41). O'Neill does not address the problem of maxim-fiddling in connection with this view of maxims, and the reason should be clear: it provides a natural foundation for the formulation of the problem. Just as an agent can intend to do X, she can intend to do X under certain conditions, and these conditions might be tailored to ensure that the action passes the CI universalizability test.Footnote 15

Herman offers a third account of the nature of maxims. Her view is closer to the permissive approach of the earlier O'Neill than the restrictive one of Timmermann and the later O'Neill. She emphasizes the variation in ‘levels of generality’ with which Kant offers examples of maxims as a problem for the two dominant views, and she notes that textual evidence seems insufficient to settle this issue. She takes the central interpretative task to be representing the way an agent wills. Accordingly, ‘the maxim should include all aspects of both action and end that the agent would offer as justification for her acting as she intends to act’ (1993: 221). The formulation of an agent's maxim should specify why she sees the action as good or choice-worthy. The things that might be relevant to specifying this can depend on idiosyncratic features of an agent's motives, such as how the action contributes to her interests, her goals, or the production of ‘collateral effects also of interest to her’ (Herman Reference Herman1993: 221). These subjective elements relevant to maxim specification might be evident to the agent, but they need not be. Agents can be mistaken about the maxims on which they act. As a consequence, ‘Maxim specification is a dynamic process that the agent will be drawn into as her actions as she describes them seem at once justified and in conflict with principles she accepts’ (Herman Reference Herman1993: 223; original emphasis). Herman does not address the problem of maxim-fiddling and, as with the second dominant view of the nature of maxims, the reason should be clear. The combination of the role for agents in maxim specification with the place for subjective, idiosyncratic interests in the contents of maxims provides a natural foundation for the problem of maxim-fiddling. In an explicitly Kantian context, a particular way of thinking about an action might be choice-worthy precisely because the action so conceived will pass the CI universalizability test. Maybe the agent has other more primary interests, but the ‘collateral effect’ of passing the CI test may, by Herman's standards, be relevant to maxim specification. Since the agent may be part of the process of maxim specification, the choice of CI-successful maxims after initial failure – i.e. of maxim-fiddling – is a natural implication of Herman's view of the nature of maxims.Footnote 16

In summary: I have looked at three views of the nature of maxims and their respective implications for the maxim-fiddling problem. Two of these views provide natural foundations for the expression of this problem, and hence are not suitable for solving it. The third view, which happens to be the only one which explicitly addresses this problem, is unsatisfactory for both comparative and inherent reasons.

I can now add three more reasons in defence of using as general a notion of ‘maxim’ as possible to address the maxim-fiddling problem. The first is related to the reason offered in section 3. The present strategy will provide a decisive answer to the problem of maxim-fiddling, whereas the strategy of pursuing a specialized, truly Kantian definition of ‘maxim’ will not. As noted by Herman (Reference Herman1993), the textual evidence seems to be such that any particular specialized definition of maxim will always be contentious. This makes such definitions unlikely foundations for an adequate response to the problem. By contrast, a response which uses an interpretation of ‘maxim’ which is consistent with Kant's discussion in G and which is used by the objectors themselves offers the promise of a definitive solution.

The second reason is in the spirit of Herman's approach: Kant's examples of maxims in G seem just to be generalized ways of acting and of thinking of one's own actions. They seem to be neither deep commitment-defining principles nor intricately specified intentions in action, contra Timmerman and O'Neill. So adopting a loose interpretation of ‘maxim’ allows us to retain the spirit of these examples. This strategy is also in the spirit of Kant's G definition of maxims as subjective principles of volition. We retain the deep ambiguity of this wording, perhaps desirably, by refraining from adopting a narrow understanding of what a maxim is. Overall, a response to maxim-fiddling using a loose notion of ‘maxim’ will match the spirit of Kant's own discussion in G.

The third reason pertains primarily to the definition of maxims as deep principles, as offered by O'Neill and Timmermann. O'Neill and Timmermann dismiss the problem of maxim-fiddling as ill-formulated: revised ‘maxims’ are really just revised specific intentions in actions, not maxims at all. One worry about this response is that it risks missing the point of the objection: if agents can exercise control over their commitments, and if they can have idiosyncratic commitments, then there is a problem of maxim-fiddling even by the exacting standards of this definition of ‘maxim’. By contrast, a response to this problem that avoids using such a specialized definition of ‘maxim’ avoids this risk of merely pushing the problem back one step. If a restrictive notion of ‘maxim’ is the true Kantian one, then a response to maxim-fiddling using a wider notion will apply to the true notion mutatis mutandis.

Overall, I have no deep quibble with the project of specifying a Kantian view of the nature of maxims. This is desirable both on its own and as a potential way of answering objections. However, the provision of a specific Kantian definition of ‘maxim’ seems not to be imminent. Nor are the options on hand particularly promising as a response to the problem of maxim-fiddling. I conclude that the solution offered in section 3 is both new and superior to the alternatives that have been offered.

5. Concluding Reflections on the Status of the Problem

If my arguments are correct, then the problem of maxim-fiddling is distinct from the problem of puzzle maxims with which it is typically connected. Puzzle maxims, or false negatives, are principles of volition which seem clearly morally innocuous, yet which fail the CI universalizability test. Examples are ‘to always be first through the door’ (Herman Reference Herman1993: 225) and ‘dining at a friend's place on Monday nights’ (Timmermann Reference Timmermann2007: 157). To deal with these, it is important to address the nature of ‘maxim’. The typical results are either that these turn out not to be maxims, or that they turn out to be impermissible when adopted as character-defining principles.Footnote 17

The maxim-fiddling problem appears to be the inverse of the problem of puzzle maxims, but it is not. The difference is that the maxim-fiddling problem arises in connection with a particular way of running the CI universalizability test, whereas puzzle maxims are just maxims that seem unjustifiably to fail this test. The maxim-fiddling problem is about the CI universalizability test for moral permissibility, whereas the problem of puzzle maxims is not. This implies that these need different treatments. Hence I have provided a treatment of maxim-fiddling distinct from puzzle maxims.

The inverse of the problem of false negatives is the problem of false positives. These are maxims which should be ruled out by the CI test but which pass. The maxim-fiddling problem differs from this problem more than one might think. The arguments in section 3 show that the issue is designing your maxim specifically to pass the CI test. Only some false positives are arrived at in this way. The present arguments do not apply to false positive maxims that have been arrived at accidentally. One implication is that the present arguments suggest that there is something suspicious about legitimate positives that an agent has because she desires to pass the CI test. I shall not pursue this issue here.Footnote 18

What remains to be shown is the deep problem indicated by these arguments. Failing the CI universalizability test shows that an action is wrong, but it does not by itself show why the action is wrong (Herman Reference Herman1993: 226–7). What is the deep problem with maxim-fiddling?

The import of this question becomes particularly clear when we compare maxim-fiddling with the four examples used by Kant to demonstrate the universalizability test (4: 422–4). Two involve duties to oneself and two concern duties to other people. Suicide is an offence against oneself, and specifically against one's own rational nature. Refraining altogether from charity is an offence against others. Maxim-fiddling fits neither case: we do not do maxim-fiddling to ourselves or to other people. This suggests that maxim-fiddling represents a special kind of moral problem, one that cannot be understood as an offence against either oneself or other people.

The key to this issue is a point raised at the end of section 3: that the maxim-fiddling maxim is a higher order maxim. It is about maxims and, specifically, about designing maxims to pass the CI universalizability test. Given these objects, we should see maxim-fiddling as an offence against morality itself. Although Kant's four examples do not exemplify such offences, their possibility is provided for later in G in the discussion of the distinction between price and dignity. In describing the kingdom of ends Kant claims that everything there has either a price or a dignity (4: 434–5). To have a price is to have a value that it is legitimate to assess quantitatively, at least figuratively. This implies that, if it can be determined that two things have equivalent prices, it is appropriate to substitute one for the other. By contrast, things with a dignity are ‘above all price’ (4: 434). The value of things with a dignity is not assessable quantitatively, not even figuratively, and hence it is illegitimate to speak of equivalent dignities, nor is it appropriate to try to replace one thing with a dignity with another. Most things in the kingdom of ends have merely a price. However, ‘morality and humanity, insofar as it is capable of morality, alone have dignity’ (4: 435). It is our rational nature that makes us capable of morality, and this is the source of the problems in Kant's four examples. He provides no case that exemplifies the dignity of morality itself, including how one might fail to respect this dignity. I submit that maxim-fiddling provides such a case.

Without getting into the nuances of understanding the kingdom of ends, here is a two-fold diagnosis of the deep problem with maxim-fiddling. The first aspect of this case turns on the relation of this test to our desires. The CI universalizability test is our tool for moral assessment of our maxims of action. This test assesses whether the sort of thing that one is considering doing can be abstracted from the particular case and shared with other people. By the standards of this test an action is permissible if it is the sort of thing that could be, in principle, performed by anyone at any time. Permissible actions are essentially shareable; impermissible actions are not, meaning that in performing them you make an exception of yourself and must will that others not act in the same way. To design maxims specifically to pass the CI universalizability test is intentionally to subvert the way this test assesses shareability. In some cases the specific details of actions are passed off as shareable; this is found in MacIntyre's description of maxim-fiddling. In other cases shareability is ensured by building test-passing conditions into the specification of the maxim; this is found in Feldman's discussion. In both cases, morality's tool for assessment of our actions is subverted for one's own interests. This is to treat morality as merely a means to satisfying our own desires, and this in turn is inconsistent with respecting its dignity.

The second aspect of the diagnosis of the deep problem with maxim-fiddling turns on further reflection on the grounds of dignity. What is it for something to have dignity rather than a price? Kant's answer is that ‘that which constitutes the condition under which alone something can be an end in itself has not merely a relative worth, i.e., a price, but has an intrinsic worth, i.e., dignity’ (4: 435). Morality is this condition for rational beings. For present purposes the important point is the contrast between ends in themselves and other ends. Other ends depend for their worth on our desires. To be an end in itself something must have a worth independent of desire. Kant's position is that morality makes this possible for rational beings. How? Without getting into detailed nuances, moral laws are laws we give ourselves as expressions of our freedom qua rational. When we act in accordance with such laws, the will is determined not by inclination but by reason. Herein lies the deep problem with maxim-fiddling. When we design maxims specifically to pass the CI universalizability test, we make all action, and the ends thereof, dependent on inclination. We hence rule out the possibility of there being, at least in practice, ends independent of desire, i.e. ends in themselves. Since, by Kantian standards, the grounds of this possibility have the special evaluative status designated as ‘dignity’, maxim-fiddling is an offence against the dignity of morality because of its threat to ends-in-themselves.

More needs to be said to develop this case. For now, it should be clear how and why the problem of maxim-fiddling differs from the problem of puzzle maxims. Puzzle maxims contravene no duty to morality itself. It should also be clear why the problematic nature of maxim-fiddling has been overlooked: the idea of duties to morality itself is an odd one, not even addressed by Kant himself in his examples. If we think of the domain of duties as exhausted by duties to people, then it should seem that maxim-fiddling contravenes no duties. But the maxim-fiddling maxim is not universalizable, and Kant's own position in G provides at least the beginning of an explanation of why this should be so. In both regards, the present response to the problem of maxim-fiddling is novel yet thoroughly Kantian.