No one before Kant had imported the concept of “autonomy” into ethics. It had always been a strictly political term, referring to a nation or people that is self-governing, or free from colonization or conquest. But at the end of the paragraph that begins with the statement quoted above (Kant 1997, 40/4:432), Kant adapted the oqnce strictly political term to refer to individual persons. He there also introduced autonomy’s antonym into ethics: “heteronomy.”

Heteronomy in the history of ethics. It is clear from the context where he introduced the terms “autonomy” and “heteronomy” that Kant wanted to emphasize, in a dramatic way, the contrast between his approach to ethics and all that had gone before. One way he did this was to distinguish between hypothetical imperatives and categorical imperatives. Principles of duty had always been presented as hypothetical imperatives, he claimed. But it is clear, or at least it was to him, that moral principles can be adequately expressed only by categorical imperatives. The difference is that, in principles of the first type, we would be told how to act in order to achieve some end. Yet if we did not care about the end presented by the principle, it would not give us any reason to do what it commands. For example, a hypothetical moral imperative might be: “Thou shalt not commit adultery, in order to be pleasing to God, and avoid punishment in a future life.” But if you do not believe in God, or in a future life, then the imperative does not give you any reason not to commit adultery. A categorical imperative, by contrast, would say simply that adultery is wrong, or that you ought not commit adultery. It would not offer any future condition to be realized by your obedience as the reason why you should obey. Ethical principles had been conceived historically as hypothetical imperatives, Kant said. What he suggested instead is essentially that the commands of morality are given categorically by each rational agent to him- or herself; and therein lies the reason for obeying this. This is the difference between the heteronomy of the moral imperatives found in the history of ethics leading up to Kant, and the autonomy implied by the categorical moral imperative as Kant introduced it (see Schneewind, 1998).

Autonomy and universality. Kant’s new understanding of autonomy must provide the answers to two difficult questions. First, why think the moral law that each legislates for him- or herself will be universal, in the sense of being the same for everyone? No one expects autonomous nation-states to have the same constitutions, and the same statutory laws. So why think autonomous individuals would all self-legislate the same law? Second, why think that the laws autonomous individuals give to themselves will be permanent, or universal for all time? If someone can legislate her obedience to a certain law at a certain time, she can just as easily, later, repeal that legislation. So why should we suppose that autonomous agents commit themselves to always obeying the moral law? The answer to both questions depends on some assumptions in the metaphysics of morals. These will not be worked out in any detail until Groundwork III—and even there the details are faintly sketched. But Kant has so far expressed at least some metaphysical principles, in a conditional way. That is, he has claimed that if there is a valid categorical imperative, then it must depend on something that exists as an end in itself. We of course have no scientific or empirical way to determine whether anything exists as an end in itself. So if we are convinced that the categorical imperative is a valid, moral principle for our conduct, then we are committed, in metaphysics, to something’s existing as an end in itself. If we can see ourselves going that far, Kant assumes, then we should not have any qualms about commitments to metaphysical principles that may be implied by the formula of autonomy. Chiefly, it will be required for this principle’s validity that each of us exists as a member of the same community, and outside of time. Our being united in a communal existence (the moral world) can explain how we all, together, legislate for ourselves the same law (the answer to the first question about autonomy). Our existence outside of time can explain how we are each committed to that legislation without possibility of later repeal—since if we are outside of time there is no before and after (the answer to the second question about autonomy). Again, the arguments of Groundwork III are supposed to address these more central, metaphysical issues behind Kant’s idea of autonomy. For the present, the community of self-legislators referred to here will be the subject of Kant’s next idea related to autonomy, which he calls a “kingdom of ends.”

Autonomous and heteronomous action. For many years it has been commonplace for writers on Kant’s ethics to apply the concepts of autonomy and heteronomy to the wills of moral agents, to types of moral principles, and to actions. But Kant himself applied the concepts to the first two only, never to the third. Hill confirms this with his observation that “contemporary writers refer to conscientious action as ‘autonomous’ and to desire-motivated action as ‘heteronomous.’ These are not Kant’s expressions and they misleadingly suggest that those who act wrongly, or act to satisfy innocent desires, lack autonomy (at least at the moment)” (Hill 1992, 85). That suggestion is misleading since Kant would not want to deny that a rational agent is autonomous because of choosing to act a certain way. Nevertheless, this misleading presentation of Kant’s conceptions of autonomy and heteronomy has become entrenched. It can be found in English critics of Kant as early as the late 19th century (see Sidgwick 1888, 412) and in more sympathetic writers since at least the middle 20th century (see Paton 1948, 215). Korsgaard, more recently, explains that “When you are motivated autonomously, you act on a law that you give to yourself; when you act heteronomously, the law is imposed on you by means of a sanction” (Korsgaard 1996, 22). She then goes on to illustrate her point with an example of someone paying taxes heteronomously, in order to avoid a legal penalty, contrasting this with paying taxes autonomously, in order to do what is morally right. The issue between Hill and Korsgaard on this point cannot be decided by consulting passages from the text, simply because Kant was not clear enough on this point. Korsgaard cites a text from the Groundwork that, to her mind, supports her interpretation (Korsgaard 2009, 81). But it is not clear that the passage she cites should be read that way. Kant wrote:

If the will seeks the law that is to determine it anywhere else than in the fitness of its maxims for its own giving of universal law... heteronomy always results.  (Kant 1997, 47/4:441)

This passage, in its context, is more plausibly read as explaining the condition making a principle of conduct heteronomous, rather than an action. The context of the passage is concerned with the difference between the heteronomy of all the moral principles that had been offered historically, and Kant’s categorical imperative as a principle of autonomy. So it does not seem correct to read the statement Korsgaard cites as evidence for Kant’s having recognized the possibility of acting heteronomously. The best way to adjudicate this dispute over the possibility of autonomous and heteronomous action, it seems to me, is to inquire what necessary role the distinction might play in Kant’s system. I do not think it is necessary at all. The distinction between autonomous and heteronomous action seems to be nothing but the distinction, in different terms, between acting from duty, with “moral worth,” and acting from inclination. That distinction, found much earlier in the Groundwork, eliminates any necessity to apply the concepts of autonomy and heteronomy to actions. So I think Hill is correct. Contemporary writers are mistaken, and misleading, in classifying actions as autonomous or heteronomous on the basis of their motives.

Heteronomy and unfreedom. A vexing problem about the freedom of heteronomous action is the main consequence of the preceding (mis)interpretation. As will become evident later, in the third section of the Groundwork, Kant saw the concept of autonomy as importantly connected to freedom of the will. Judging from some of his comments, it seems as if the two might be the same. Operating under the assumption that they are the same, critics have charged that if acting autonomously is acting freely, then acting heteronomously must be acting unfreely, and so no one should be held morally responsible for immoral actions, since they would be heteronomous (see Sidgwick, 1888). Commentators sympathetic to Kant have invoked a set of interpretive tricks in reply to this criticism (the “incorporation thesis” and the “Wille-Willkür distinction”), mainly because they have seldom if ever asked whether the ideas of heteronomous and autonomous action actually belong in Kant’s system. Presently, the line of response to the criticism that seems most widely agreed upon involves explaining autonomy as a kind of capacity or disposition of rational agents, which is the same or nearly the same as their free will, and which is retained even while acting heteronomously (see Allison 2011, 262). It is said that one is autonomous, free and therefore morally responsible for one’s actions because of having this capacity, regardless of whether one exercises it in acting on purely moral reasons, or acts heteronomously, even immorally. But the difficulty with this idea is that autonomy, if it is supposed to refer to “self-legislation,” does not appear to be a capacity. No one thinks of self-legislation as a capacity, any more than they think of self-nomination as a capacity, or self-incrimination. There can of course be capacities for self-legislation, or self-nomination. But self-legislation cannot be the same as the capacity for self-legislation. That is why autonomy cannot be the same as the capacity for acting autonomously. So seeking a solution to the problem of the unfreedom of heteronomous action in a “capacity of autonomy” seems misguided. If what has been explained above is mostly correct, then the best solution to the problem seems to be simply to deny that heteronomous action makes any sense in Kant’s theory. Heteronomy can still be equated with unfreedom; but acting from desire can no longer be equated with heteronomy.