Debate: Paternalism and Autonomy

I. DEFINING TERMS

Paternalism can be understood in many different ways, and the definitional challenges are daunting.⁶ For present purposes, let us understand the term to refer, very simply, to efforts taken by outsiders, including government officials, (1) to influence, alter, or interfere with people’s choices, (2) for their own good, (3) on the basis of a judgment that without the influence, alteration, or interference, people will make mistakes if they act on their own.⁷ “Government officiala” might refer to legislators; it might refer to those in administrative agencies; it might refer to courts. A paternalistic government does not believe that people’s choices will always promote their welfare, and sometimes it takes steps to influence, alter, or interfere with those choices to make their lives go better. Such steps will be my focus here.

We are not going to get bogged down by definitional issues, but there are of course alternative definitions. In an influential treatment, David Archard offers this one:

P behaves paternalistically towards Q iff:

• (1) P aims to bring it about that with respect to some state(s) of affairs which concerns Q’s good Q’s choice or opportunity to choose is denied or diminished;

• (2) P’s belief that this behaviour promotes Q’s good is the main reason for P’s behaviour;

• (3) P discounts Q’s belief that P’s behaviour does not promote Q’s good.⁸

This is a more restrictive definition than mine. It emphasizes that the paternalistic actor denies or diminishes choices or opportunities to choose. For that reason, it would appear to exclude efforts (merely) to influence or interfere with people’s choices for their own good. Seanna Shiffrin’s influential definition is not limited to cases of denying or diminishing choices.⁹ She says that paternalistic behavior by A is

• (a) aimed to have (or to avoid) an effect on B or her sphere of legitimate agency

• (b) that involves the substitution of A’s judgment or agency for B’s

• (c) directed at B’s own interests or matters that legitimately lie within B’s control

• (d) undertaken on the grounds that compared to B’s judgment or agency with respect to those interests or other matters, A regards her judgment or agency to be (or as likely to be), in some respect, superior to B’s.

This definition captures what many people have in mind with the term, and it fits with mine insofar as it points to “an effect on B,” rather than some denial of B’s choices, or a diminution of those choices. But the word “substitution” is ambiguous, and we might object to the word “legitimate” in (a) and to part of (c). Suppose that A stops B from smoking, on the ground that if B smokes, B will run serious risks, not justified by the benefits of smoking. Do we have to ask whether smoking is “legitimately within B’s control” in order even to know whether A is acting paternalistically? I hope not. We need not answer that question to assert that A is indeed acting paternalistically. But for my purposes here, a range of definitions will work, and we need not struggle to choose among them.

II. TWO FORMS OF PATERNALISM

Let us distinguish between two forms of paternalism: coercive and libertarian. Coercive paternalism is far more familiar.¹⁰ It can be found in mandates and bans, as in a requirement that people buckle their seatbelts. Economic incentives, positive or negative, are not bans, but they can be placed in the same category, as forms of coercive paternalism. Consider a fine or a tax, imposed on those who smoke, who do not buckle their seatbelts, who refuse to wear motorcycle helmets, fail to save for retirement, or who buy sugar-sweetened beverages. By contrast, libertarian paternalism can be found in nudges, which preserve freedom of choice, and which do not impose material incentives, but which steer people in particular directions.¹¹ A GPS device is a nudge. It respects people’s ends, but it steers them toward certain means of achieving those ends.

As forms of libertarian paternalism, nudges fall in two categories: educative and architectural. In response to the risk of human error (perhaps stemming from cognitive biases), educative nudges aim to teach people something and to promote learning, with the goal of influencing or steering them to do something. Consider a warning (“cigarette smoking causes cancer”), a reminder (“your bill is due”), or a label (“200 calories”). A fuel economy label is an educative nudge, and so is a “nutrition facts” panel. Graphic warnings on cigarette packages count as nudges. Educative nudges might be designed to overcome self-control problems, weakness of will, unrealistic optimism, inertia, limited attention, or present bias.

It would be possible to question whether educative nudges are paternalistic, and on some definitions, they are not. But if a planner, such as a government official, is seeking to steer people in a certain direction with the thought that choosers, unsteered, will not choose well (perhaps because of a behavioral bias, such as unrealistic optimism or limited attention), it seems reasonable to count the resulting action as paternalistic. After all, a label or a warning might be about any number of things, and if a government official mandates a label or a warning about one thing rather than another, it is likely to be because of a judgment that without it, people will choose poorly. The specification of a single product characteristic, as the focus of regulation, might be counted as a paternalistic act, designed to promote better choosing.

By contrast, some nudges are architectural, in the sense that they alter the social environment, or the background against which choices are made, without necessarily teaching people anything at all. Consider automatic enrollment in savings or health care plans, with the goal of overcoming inertia or procrastination on the part of choosers. Or consider redesign of a cafeteria, a website, or a store so as to encourage healthy choices, perhaps by making those choices salient to choosers, who might otherwise pay too little attention to health considerations. All over the world, we are seeing educative and architectural nudges, informed by behavioral science,¹² and occasionally facing serious objections from those concerned about autonomy.¹³ Notwithstanding those objections, there is evidence that both kinds of nudges attract strong majority support in many nations,¹⁴ though there is also evidence that people’s enthusiasm or lack of enthusiasm for nudges is much affected by whether they support or do not support the values reflected in the nudge.¹⁵

III. AUTONOMY, THICK AND THIN

Nudges can increase welfare, not least by reducing morbidity and mortality, and if we are welfarists, our support for or rejection of nudges will depend on whether they do that.¹⁶ Consider, for example, an effort to nudge people with heart problems to take prescribed medicines at the relevant time, perhaps with reminders via text message, perhaps with carefully designed packages that help people to remember. Or consider redesign of automobiles so as to nudge people to drive within the speed limit, not to change lanes recklessly, or to pay close attention when they are approaching other vehicles. These interventions, and many others, might be easy to justify on welfarist grounds.¹⁷

A nonwelfarist approach, prominently favored by Jeremy Waldron, David Enoch, and others, might seek to identify the right conception of autonomy, and explore whether one or another approach, including libertarian paternalism, is consistent with it.¹⁸ Those who object to paternalism on grounds of autonomy might understand their argument as a kind of trump card: Because paternalism intrudes on autonomy, it is unacceptable, period. On that approach, autonomy is a side constraint. On a more cautious view, autonomy is a value, not a trump card, but it must be given weight. It counts against paternalism. Those who hold that more cautious view might agree that if paternalism greatly improves people’s well-being, it can be justified, even if it reduces their autonomy. I will be dealing here largely with autonomy as a trump card, but the various points apply to the more cautious view as well.

IV. RESPECT

Autonomy can of course be specified in different ways, and different specifications might lead to different judgments about paternalism. Broadly emphasizing respect for persons, Kantians would be strongly inclined to be skeptical of coercive paternalism via law,¹⁹ though they might well be amenable to libertarian paternalism, depending on its nature and scope and why it works.²⁰ A central problem for Kantians is that coercive paternalism represents a failure of respect²¹ and to that extent a violation of rights.²² As Carl Fox puts it, “paternalism is insulting because one agent treats another as something less than a full moral equal and thereby calls her status as an equal into question.”²³ Some definitions of paternalism invite that conclusion.

The idea of an “insult” animates many objections to paternalism. On one view, what is wrong is the intention of the paternalist, which is insulting and demeaning to the object of paternalism, who is being treated as a tool or a fool. As we have seen, some people, including Shiffrin, build the idea of intention into the very idea of paternalism. But this does create a bit of a puzzle. Some kind of “choice architecture” cannot be avoided, and in that sense some kind of nudging is inevitable.²⁴ Any website, or office, has a design. A retirement plan must be designed with opt-in or opt-out, or perhaps with active choosing, by which people are required to choose in order to work at all. Is opt out a violation of autonomy? More than opt in? That is not obvious. Might active choosing be a violation of autonomy, if people do not want to choose? Forced choosing would seem to be paternalistic too, when choosers do not want to choose.

If a designer does not know that participation rates will be higher with opt-out (perhaps she has not read behavioral economics), some of those who are focused on autonomy would likely have no objection, because a paternalistic intention is what is objectionable. If, by contrast, a designer does know that, and chooses opt-out, those focused on autonomy will find some kind of insult. That does seem a bit awkward.²⁵ Is someone’s autonomy really affected by someone else’s intention, independent of the nature or effect of that someone else’s action?

Perhaps so, in which case everything does indeed depend on that intention. But if not, the problem may be that paternalistic actions impose an expressive harm, suggesting that their object is inferior or lesser. As Nicholas Cornell puts it, “instances of paternalism are objectionable because of their expressive content. Paternalism is suspect because it implies that the other party is not capable of making good judgments for herself.”²⁶ Note that Cornell is not drawing attention to the intentions of the paternalist; he is pointing to the expressive effect of the paternalistic act. We can certainly see the force of Cornell’s claim in the context of coercive paternalism, as with a regulation forbidding workers from running certain risks, or banning recipients of federal funds from using those funds to buy the foods that they prefer. It is true that even if we think that such actions impose an expressive harm, and are thus troublesome from the standpoint of autonomy, we might conclude that they are justified on balance. Autonomy may or may not be a trump card.

The question of libertarian paternalism must be approached in its own right. Consider a GPS device, which would be hard to reject on Kantian grounds. It does not fail to treat people with respect. It does not seem to impose expressive harm. The same could be said for educative nudges more broadly. A fuel economy label does not treat people disrespectfully or run afoul of Kantian strictures. Something similar is true for nutrition facts panels, designed to promote healthier eating. If people are reminded that they have a doctor’s appointment tomorrow or that a bill is coming due, no one has treated them as tools or fools, or as infants. Even adults can be forgetful.

Are architectural nudges disrespectful? Do they work expressive harm? Suppose that people are automatically enrolled in some health care program, on the ground that they are eligible, might not bother to enroll, and would certainly or almost certainly benefit from enrollment. If so, it is not so easy to identify a problem on Kantian grounds, or to say that people have been treated disrespectfully or with contempt. It is true that if people are automatically enrolled in a retirement plan in part on the ground that inertia is a powerful force, we might worry that some third parties (perhaps government officials) think that they know better than choosers do. But it is not exactly contemptuous to acknowledge that human beings are human and to try to make life easier for them, while also acknowledging that they might want to take their own path, and making it easy for them to do that (through simple opt-out).

V. CONTROLLING ONE’S DESTINY

Suppose that we adopt a distinctive conception of autonomy in accordance with which people are entitled to control their own lives, or to be authors of their own life’s narratives. Joseph Raz writes:

The ruling ideal behind the ideal of personal autonomy is that people should make their own lives. The autonomous person is a (part) author of his own life. The ideal of personal autonomy is the vision of people controlling, to some degree, their own destiny, fashioning it through successive decisions throughout their lives.²⁷

On Raz’ account, coercion endangers autonomy, and so does manipulation. If we accept Raz’ account, educative nudges would not seem to be a problem. On the contrary, they would promote autonomy. They help people to control their own destiny. They do not manipulate anyone.

What about architectural nudges? What about automatic enrollment in antipoverty programs, health care plans, or retirement plans?²⁸ Initiatives of this kind do not seem to run afoul of Raz’ “ruling idea”; people are fully entitled to control their own destiny. Nonetheless, some people object to libertarian paternalism of that form on autonomy grounds. Apparently focused on such nudges, Enoch writes plainly: “Everyone suspects – perhaps knows, but at least suspects that nudging offends against the nudged’s autonomy.”²⁹ (He adds a nice footnote, “Well, pretty much everyone,” referring to the present author as an exception.³⁰) At the same time, Enoch candidly notes that “it has proved rather difficult to say why.”

Let us spend some time with his account, on the ground that it is the most careful and subtle effort to bring together Kantian and Razian themes to show that libertarian paternalism does indeed offend autonomy. Enoch is aware that if people are free to opt out, the objection from autonomy is not so easy to vindicate. Still, Enoch urges that if people “end up with a specific – even advisable – savings plan merely because someone from nudging central-command made sure that was the default option,” there is a real problem from the standpoint of autonomy, even if there is no problem from the standpoint of welfare.

It might not be unfair to wonder whether the term “nudging central-command” is loading some dice. If we said, “a public official, after a process of public comment,” the conclusion might be less clear. Still, the point holds: A person ends up with a savings plan, not because they have chosen to do so, but because some outsider, perhaps a government official, has made it the default option. To explain what he sees as the widespread intuition that nudges violate autonomy, Enoch builds on Raz to make an illuminating distinction between two conceptions of that idea. The first is autonomy as sovereignty. The second is autonomy as non-alienation.³¹

Sovereignty means that you are able to exercise control over how your life goes (and to this extent Enoch clearly follows Raz). Suppose, for example, that you are told that you cannot eat brownies or sugar-sweetened cereals, or that you must buy an electric car, or that you are not allowed to smoke, to hunt, or to play tennis. If so, you have lost autonomy in the sense of sovereignty, whatever the effects on your welfare. Now suppose that you want to be a good and healthy person, but that you have a serious problem with alcohol, and that if alcohol is nearby, you will drink a lot of it and misbehave. If you are succumbing to temptation, Enoch thinks, your choice is a product of alienation, in violation of an important value: shaping your life in accordance with your own deepest commitments. Enoch defines this as the value of non-alienation.

Enoch acknowledges that libertarian paternalism need not offend sovereignty. An educative nudge is highly unlikely to do that. Architectural nudges might also respect sovereignty. If you choose an item because it is listed first on a website, you have still made the relevant call, and you could easily have chosen another item. What about non-alienation? An educative nudge need not offend that conception of autonomy. On the contrary, it might promote non-alienation (by, for example, reducing the power of temptation; consider a graphic health warning). We might reach the same conclusion for architectural nudges. If I choose a healthy item because it is listed first, or if I remain enrolled in a health care plan after automatic enrollment, Enoch agrees that there need be no problem from the standpoint of non-alienation. His objection is more intriguing and subtle than that. It is that when certain forms of libertarian paternalism are at work, “it can no longer be truly said that the decision manifests non-alienation because it manifests sovereignty.”³² That is, the relevant choice was made in part because of the nudge, and not solely by the chooser in an exercise of autonomy.

As Enoch has it, “The value of autonomy is fully manifested in a choice only if it manifests non-alienation precisely because it manifests sovereignty.”³³ If people choose to drink water rather than alcohol, there is no problem; choosers have manifested sovereignty in the interest of non-alienation. He objects that if, for example, automatic enrollment leads people to save more for retirement, both sovereignty and non-alienation are independently respected, “but the connection between the two will be severed.”³⁴ People save for retirement in part because they have been nudged, and that is true, Enoch thinks, “for any other nudging case.” That is bad, because “the right relation between sovereignty and non-alienation is also needed for the full manifestation of the value of autonomy.” Nudging, Enoch urges, “always upsets autonomy.”³⁵

This is a subtle, careful, and illuminating analysis. An initial question is straightforward: Is it really true for any other nudging case? If people are nudged, through a reminder, to pay their electricity bill on time, and if they then pay it, we have not compromised the right relationship between sovereignty and non-alienation. People remain sovereign over their choices, and nothing has been alienated, and their decision manifests non-alienation because it manifests sovereignty. Enoch seems to be focusing on architectural nudges, not on all nudges. He might agree with this point, of course, and simply narrow his claim to architectural nudges, which appear to be his main interest, and indeed the interest of all of those who worry over the relationship between nudging and autonomy.³⁶

A second concern would linger a bit over Enoch’s strategy here. He begins with what he sees as a widespread intuition, which is that nudges violate autonomy. (Existing empirical work raises serious doubts about that claim, because it finds widespread support for nudges in many nations;³⁷ but let us bracket that point.) Then he works very hard, very plausibly, and very cleverly to vindicate that intuition,³⁸ not only by identifying, or constructing, two conceptions of autonomy, but also by insisting that it is not enough to respect both of them. We must show the right relationship between the two to manifest, fully, the value of autonomy.

Maybe so. Still, we might consider the possibility that the relevant value turns out to be a heuristic for something else, which is well-being. That is, our intuitive judgments about autonomy, and our efforts to make sense of those intuitions, might reflect a heuristic for well-being, which is what really matters.³⁹ Consider an analogy. People use the availability heuristic to make assessments about probability; if some event comes readily to mind, people tend to think that it is probable, even if statistical analysis suggests that it is not probable at all. Intuitive judgments about factual questions, including the likely future, are often based on heuristics of just this sort.⁴⁰ Such judgments are frequently rapid and do work that would be better done, time permitting, through a more deliberative process. In the moral domain, consider the possibility that something similar happens.⁴¹ People make a rapid, intuitive judgment that something is moral or not, even though a deliberative process would override that judgment. This is so even if the intuitive judgment is firmly held and difficult to dislodge. Perhaps the focus on autonomy, which is often rapid and intuitive, is a heuristic, and perhaps people work long and hard to defend that focus when they would do better to go directly to well-being.⁴² This might be so even if people’s choices generally do promote their well-being, and even if frustration of choice-making often does produce a loss in terms of well-being.

It would take a lot of work to vindicate these claims, and they might not be right, so let us flag them without insisting on them. Let us consider the possibility that Enoch has developed the best understanding of autonomy and that it is a fair objection to libertarian paternalism. Is it? Suppose that someone chooses an option in part because it is the default, or in part because it is the first on a long list. To be sure, we can identify a conception of autonomy, identical or akin to Enoch’s, that the nudge violates. The conception is not without appeal, but it does seem a bit exotic. Ought we to accept it, or worry so much about its violation? Suppose that people do have freedom of choice, and suppose further that they can also insist on the primacy of their own values. The question is whether autonomy, specified in Enoch’s way, amounts to a serious objection.

There is reason to doubt it. Imagine that a government employer automatically enrolls people in a health care plan, knowing, on the basis of evidence, that the plan suits most or almost all people. Imagine too that people are allowed to opt out and go their own way. Their autonomy is doubly preserved. First, the plan is designed with close reference to their likely situations, and to what they would likely choose if they were adequately informed and sufficiently free from behavioral biases. Second, they are not required to stick with it. Why should we endorse a conception of sovereignty and non-alienation that is offended by this freedom-preserving plan? Why should we insist that people exercise sovereignty to vindicate their deepest values? Or suppose that children are automatically enrolled in a free school meals program, and that their parents can opt out if they like. Is it so terrible, from the standpoint of autonomy, that parents have not exercised their sovereignty here?

Here is another way to put it. Surely we can produce a conception of autonomy that all or most architectural nudges violate. Actually there is (surely again) more than one. Enoch identifies an intelligible and in some ways appealing one. Should we accept it? Should we condemn, as objectionable or as presumptively objectionable, acts of libertarian paternalism that run afoul of it? I do not see why we should, so long as freedom of choice is fully preserved, and so long as those acts are based on plausible judgments about what informed choosers would choose.⁴³

VI. AGENCY

There is an overlapping objection to libertarian paternalism, pressed by antipaternalists concerned with autonomy rather than welfare. Like Raz and Enoch, those who make this objection tend to associate autonomy with agency.⁴⁴ In some forms, the objection is that nudges fail to respect autonomy because they take advantage of behavioral biases and do not engage people’s deliberative capacities, which are crucial to the exercise of personal agency.⁴⁵ Some of those who offer this objection think that nudges are manipulative, and that manipulation is a form of paternalism, offensive to autonomy.⁴⁶

With respect to educative nudges, the objection again seems to miss the mark. Such nudges engage deliberative capacities and promote agency. It is true that a fuel economy label might attempt to make fuel economy salient, and that it might be responsive to a judgment that people show limited attention. But insofar as that is true, a fuel economy label increases autonomy; it does not violate it.

By contrast, architectural nudges might not engage deliberative capacities, and they might work because of behavioral biases. Automatic enrollment might work in part because of the power of inertia and inattention. If we identified autonomy with an affirmative exercise of agency, we might therefore object to automatic enrollment. But does autonomy really require exercises of agency?⁴⁷ If it did, life would become pretty exhausting, and in a hurry. People often choose not to exercise agency, and reasonably so. That is one way that they exercise their agency. Call it second-order agency, which might be exercised explicitly (through a delegation) or implicitly.⁴⁸ Of course we would have to investigate the context to know whether an architectural nudge captures or tracks people’s explicit or implicit exercise of second-order agency. For example, automatic enrollment in a pension or health care plan might well reflect people’s choice not to choose. And to the extent that it does not, people can always exercise their agency by opting out. There are insufficiently resolved issues here about choosing not to choose, autonomy, and libertarian paternalism.

The concern about manipulation is different. To come to terms with it, we need to know what manipulation is.⁴⁹ The basic point might be that nudges count as manipulative, and violate autonomy, because they are tricks, treating people as objects, and leading them in certain directions without respecting their capacity for deliberative choice.⁵⁰ Manipulation can indeed be a form of paternalism, less brutal to be sure, but more insidious. (For those who like Greek gods, or Marvel Comics: manipulative paternalists are like Loki, while coercive paternalists are like Thor.)

One more time: Educative nudges do not run afoul of this objection. Whether architectural nudges do so depends on exactly what they are. Automatic enrollment in an excellent health care plan, with easy opt-out, does not manipulate anyone; it respects people’s capacity for deliberative choice and preserves their agency. At the same time, graphic health warnings, designed to produce strong emotional reactions, might cross the line, depending on exactly why they work.⁵¹ Certain kinds of “dark patterns,” as with shrouded terms, verge on deception, and even if they do not count as that, they are unquestionably manipulative.⁵² They threaten both welfare and autonomy, and should be forbidden for that reason. They do not fit standard definitions of nudges, but if they are characterized as such, they are nonetheless illicit.

VII. A VERDICT

Coercion paternalism raises serious objections from the standpoint of autonomy. To know whether libertarian paternalism or nudging violates autonomy, we have to ask hard questions about how and why it works. One more time: If nudges work by informing people, there would seem to be little basis for complaint, even or especially if we give autonomy pride of place. If they free people up to spend time on matters that interest and most concern them (consider automatic enrollment), the complaint would also seem to be weakened. Autonomy is increased if people can manage their time as they like.⁵³

The conflict between libertarian paternalism and autonomy often disappears on reflection and after engagement with particulars. So long as coercion is avoided, it is reasonable to insist that educative and architectural nudges preserve autonomy. Still, it is true that freedom-preserving architectural nudges can intrude on some conceptions of autonomy, perhaps because they do not call for personal agency, properly understood, or perhaps because they do not promote, and may bypass, deliberative and reflective choices on the part of those whose lives are involved. For those for whom those conceptions are primary, that is a plausible concern. The point raises an objection to libertarian paternalism, at least where we can avoid it. (Recall that we often cannot.)

The question is whether we should accept those conceptions of autonomy. In my view, we can, and should, find them intelligible, without taking them to be convincing arguments to libertarian paternalism.

ACKNOWLEDGEMENTS

Many thanks to two anonymous reviewers and Robert Goodin for valuable comments on a previous draft. Thanks too to Richard Thaler for many discussions and to Lexie Gagnon for superb research assistance.

COMPETING INTERESTS

The author declares that he has no competing interests.