The Libertarian Paternalistic Justification of pro-Self Nudges
There may exist different kinds of normative justification for nudges; for example, in situations where more freedom-interfering interventions like prohibitions are justified, nudges might also be justified. However, we will concentrate in this part on the question whether the governmental use of pro-self nudge can be ethically justified on the basis of the specificity of the tools that are used—namely, pro-self nudges, rather than financial incentives or commands as in classical paternalism.
For this, it is important to dissociate the different dimensions of nudges. There seem to be little ethical objections against informing nudges, informing being generally considered as ethically unproblematic. And as a matter of fact, the bulk of the literature has concentrated on heuristics-triggering nudges; those are the ones whose normative justification we are going to investigate here.
Nudges and Rationality
Libertarian paternalism points to the purported evidence from psychological research for the systematic irrationality of human decision makers, in order to justify the use of nudges. Because a person B’s decision that is marred by irrationality might be considered to not be truly B’s decision at all, the usual arguments against paternalism do not apply: “to whatever extent B’s apparent choice stems from ignorance, coercion, derangement, drugs, or other voluntariness-vitiating factors, there are grounds for suspecting that it does not come from his own will, and might be as alien as the choices of someone else” (Feinberg 1986). Nudging a systematically irrational agent thus might be justifiable because it helps the agent realize her own will.
When this argument is used to justify heuristics-triggering nudges, it relies on an outcome-oriented, rather than process-oriented (Charland 2014), conception of rationality: what matters is that the choice fits some notion of rationality (typically satisfaction of preferences), irrespective of whether the process by which this choice was produced is rational. As some authors have suggested, this might be a consequence of the reliance of nudges on as-if models, and the neglect of underlying mental processes (Berg and Gigerenzer 2010; Grüne-Yanoff et al. 2014). On the opposite, heuristics-blocking nudges may be seen as relying on a process-oriented conception of rationality that would remove supposedly irrational applications of heuristics in certain circumstances.
However, one needs to be cautious with this diagnosis of irrationality. Gigerenzer (this issue) argues that many decisions that appear to be irrational on a superficial analysis are in fact ecologically rational—that is, they lead to adequate outcomes in the appropriate environment.
The case of time-discounting appears to be especially problematic. As detailed by Guala and Mittone (2015, this issue), because of agents’ temporal myopia, it is not clear which self is rational: the present one, or the future one? (see also Hill 2007). This is especially true if agents cannot recognize that their future selves are fully the same persons as there present selves, as argued by Lecouteux (2015, this issue): in such a case, it may not be irrational for them to strongly (i.e., hyperbolically) discount the interests of their future selves.
But the fact that the diagnostic of irrationality is sometimes difficult to establish does not imply that it is always impossible: as emphasized by Sunstein (2015, this issue), one should be wary of the risk of over-generalizing. In some situations like akrasia or addiction, it might be possible to diagnose irrationality (Guala and Mittone 2015, this issue; Barton 2013). There is thus hope that nudges might help, in such cases, to restore people’s rationality.
The Restoration of Preferences
Libertarian paternalists, in contrast to other forms of paternalism, take individuals’ own preferences seriously. Nudges, so they claim, steer people’s behaviour in a private welfare-promoting direction—that is, in agreement with their personal preferences. In particular, libertarian paternalism claims not to be a form of ‘ends paternalism’, that would impose some goals to the nudgees, but rather a form of ‘means’ paternalism, which accepts people’s goals and aims at steering people’s behaviour towards those goals (Sunstein 2014). However, libertarian paternalists also acknowledge that people’s preferences are not always well-formed. Consequently, they interpret private welfare as the satisfaction of the agent’s true preferences - those the agents would have in ideal conditions of complete information, no lack of self-control and unlimited cognitive abilities (Sunstein and Thaler 2003).
Various authors have pointed out that in practice, nudge applications rarely follow this interpretation of private welfare as the satisfaction of true preferences (Fateh-Moghadam and Gutmann 2013; Grüne-Yanoff 2012; Rebonato 2012; Rizzo and Whitman 2009). A libertarian paternalistic planner faces a knowledge problem when attempting to identify people’s true preferences. Whitman and Rizzo (2015, this issue) elaborate on this argument by suggesting that people’s underlying preferences may not even exist in some situations—like the ones in which they manifest temporal discounting, hot-cold empathy gaps, or framing effects. Moreover, Sunstein and Thaler (2003) definition of ideal conditions for true preferences (namely complete information, unlimited cognitive abilities and no lack of willpower) might appear quite ill-defined, as suggested by Sugden (2008)—who argues that even if they could be clarified, it would make them presumably impossible to know for the libertarian paternalistic planner. If people’s true preferences do not exist or are unknowable, it seems to undermine the goal of steering the agents’ behaviour towards satisfying such preferences.
Moreover, even if some nudges may steer people to act according to their true preferences, they would not do so for all agents: preferences are heterogeneous in virtually any large enough population. Some people may prefer to eat unhealthy but tasty food over healthy food, or smoke willingly, even when taking into account the risks involved. Because of this hetereogeneity, the libertarian paternalist faces what Nagatsu (2015, this issue) calls the ‘objection from coherence’ (see also Bovens 2009; Heilmann 2014; Schnellenbach 2012): nudges may push some people towards a behaviour that is not in agreement with their own true preferences.
Sunstein (2013) argue that this problem might be mitigated through personalized nudges—that is, nudges that would push agents in different directions, depending on their own preferences. However, he also acknowledges that this raises some important technical difficulties. Moreover, personalized nudging will require to accumulate enough information about the nudgees’ preferences, which raises significant issues linked to privacy: in particular, it will be impossible in many cases, through personalized nudging, to satisfy all the preferences (including the preferences about privacy) of an agent who has high privacy concerns. More fundamentally, as argued by Kapsner and Sandfuchs (2015, this issue), an agent may not even want the government to know what are his privacy preferences—so any attempt to gather information about them will defeat the libertarian paternalistic goal of respecting people’s preferences.
There is thus a general difficulty for nudges designed in a libertarian paternalistic spirit: basically any nudge will steer at least some people’s behaviour against their own true preferences, even when they are welfare-promoting for a majority of the population.
The Easy Avoidability of Nudges
A classical answer to this problem (Sunstein 2015, this issue) is that in contrast to some forms of classically paternalistic interventions, nudges are designed to be easily avoidable (Sunstein 2014; Sunstein and Thaler 2003; Thaler and Sunstein 2008) or resistible (Saghai 2013). Thus, people who want to be nudged can just “go with the flow”, and agents who have different preferences can easily avoid the nudge. However, to be able to do so, two conditions need to be fulfilled: an external condition, and an internal one.
The external condition is that to be able to avoid a nudge, one needs in a first place to know that one is being nudged, which requires a sufficient degree of transparency of the nudge. As argued by Bovens (2009), to be avoidable, nudges should not only be type-transparent (the general existence of such nudges is made transparent to the nudgee), but also token-transparent (each specific intervention is made transparent to the nudgee). Many nudges satisfy this conditions, but many others do not (see Hansen and Jespersen 2013 for a categorization, and also Lepenies and Malecka 2015, this issue).
The internal conditions have been analysed by Saghai (2013): to be able to effortlessly resist a nudge, the agent should have special cognitive capacities - namely attention-bringing and inhibitory capacities. But then appears a new issue: if people are boundedly rational, does it decrease their capacities (or their inner power to exert such capacities) to effortlessly resist a nudge’s pressure? (Hansen and Jespersen 2013). This is a question for cognitive psychology, on which neuroscientific model may have insights to bring (see Felsen and Reiner 2015, this issue, for considerations on top-down control versus automatic reasoning).
Finally, even the agents who do have such capacities will experience some cognitive or deliberative costs to escape the influence of the nudges (Heilmann 2014): they suffer, so-to-speak, of a “psychic tax” (Loewenstein and O’Donoghue 2006; Schnellenbach 2012). It seems, therefore, that in the best of all cases, libertarian paternalistic nudges will benefit some agents—hopefully a majority of them—but always at the expanse of a minority of them (Bovens 2009; Lecouteux 2015, this issue), if only because of deliberative costs.
Thus, to be avoidable (or resistible), nudges must be transparent enough, and agents need to have the right set of cognitive abilities; and even in this case, nudges benefit some population at the expanse of some costs—at least, deliberative ones—to some other population. Although this does not mean that pro-self nudges are unjustifiable (we will investigate some possible justifications in part 3.2), one cannot simply argue that they are totally innocuous because they benefit most agents and are easily avoidable.
The Unavoidability of Arranging the Choice Architecture
To make the case for nudge stronger, libertarian paternalists turn to the supposed impossibility not to nudge: as repeatedly pointed by Sunstein (e.g., Sunstein 2015, this issue), some kind of choice architecture always has to be put into place. Libertarian paternalism only suggests giving it a “good” direction, which would globally benefit the agents (Sunstein and Thaler 2003; Thaler and Sunstein 2008). There are a few caveats though.
First, although choice architecture cannot always be made neutral, there are cases where it can—see Gigerenzer’s (2015, this issue) example of framing. In fact, increasing the neutrality of the choice architecture, through proper education, training or better design, might be a policy goal in itself (Grüne-Yanoff and Hertwig 2015; Gigerenzer 2015, this issue). Therefore, one needs to be cautious, when using this argument, about what are the details of the situation at hand; for sure, it cannot be a blanket justification of all kinds of nudges.
Second, although ‘choice architecture’ refers to a decisional context and is thus indeed unavoidable, ‘nudging’, on the other hand, refers to an intentional intervention, with the goal of influencing the behavior of agents; and such an intentional intervention can sometimes be avoided (Grüne-Yanoff 2012; Hausman and Welch 2010; Rebonato 2012). Therefore, according to some non-consequentialist ethical views, the unintentional influence of a choice architecture chosen with no purpose in mind could be less ethically problematic than the intentional influence of a nudge.Footnote 4
Third, it is not obvious that, in cases where no neutral cognitive architecture can be arranged, agents should be nudged in a welfare-promoting direction - that is, that pro-self nudges should be put into place. Maybe, in some situations, it would be more justifiable to use pro-social nudges, which may not nudge in the same direction as pro-self nudges. Thus, the impossibility to choose a neutral choice architecture does not necessarily justify libertarian paternalism—at best, it would justify nudging. The direction of the nudging needs to be independently justified.
Alternative Justifications of Nudges
In the last section, we argued that although a libertarian paternalistic justification of nudging might not fail in all situations, it is weakened in a number of cases in which it pretends to apply. Yet the governmental use of nudges might be justified by other arguments. First, standard paternalistic arguments (Arneson 2005; Conly 2012, Coons and Weber 2013) might justify pro-self nudges as legitimate tools alongside more coercive measures. Second, arguments based on the harm principle,Footnote 5 or more generally on the concern about externalities (Hill 2007), might justify pro-social nudges: for example, Guala and Mittone (2015, this issue) argue that nudging agents to put enough money in their pension schemes is more justifiable as a pro-social nudge avoiding externalities than as a pro-self nudge. Third, nudges might be legitimated by democratic processes. Hagmann et al. (2015, this issue) provides insights about people’s opinions on several examples of nudges. It should be noted that ethical arguments for libertarian paternalism or pro-social nudges might have a role to play here. For example, we discussed how pro-self nudges can benefit the majority of a population at the expanse of some costs for minorities: such costs may well be judged as quite small in comparison to the benefits by the population, and thus be accepted through a democratic process.Footnote 6
Comparative Assessment of Nudges
A justification of a nudge might be absolute in the sense that it renders this nudge permissible, or even mandatory. Alternatively, one might assume a comparative perspective, and ask whether these or other justifications render nudges preferable to alternative intervention tools. Such alternative interventions include risk-savvy education (Gigerenzer 2015, this issue); boosting, which seeks to improve people’s decision-making competences by enriching their repertoire of skills and decision tools or by restructuring the environment such that existing skills and tools can be more effectively applied (Grüne-Yanoff and Hertwig 2015); and classical paternalistic measures, like economic incentives and legal commands or prohibitions.
Such a comparative perspective requires that the policy tools are distinct alternatives, which, as we discussed above, is not always the case: Sunstein (2015, this issue), for example, argues that education might be a kind of nudging. Yet even if they were distinct alternatives, it would often be possible to combine them in application. For example, many experts recommend that tobacco health warnings should be combined with taxes and other strong measures in order to limit smoking prevalence.
In some situations, however, choices between these interventions need to be made. This may be simply because of limitations of time and funds. Alternatively, one intervention may causally cancel the other. For example, when education seeks to increase the competences and skills about when and how to use which heuristics, it may conflict with heuristics-triggering nudges, which induce their unreflected use—to nudge patients into accepting a treatment by framing its benefits as relative probabilities might be less effective once we educated these patients in the skill of relating relative risk reductions to absolute frequencies.
A comparative assessment would contrast nudges with alternative intervention tools along a number of dimensions. We discuss two normative dimensions here, autonomy and transparency. Section 4 continues this perspective by discussing a number of evidential support dimensions relevant for such a comparison.
Worries about autonomy loom large in the philosophical literature on nudges. Hausman and Welch (2010) have argued that nudges infringe on the autonomy. More specifically, Wilkinson (2013) stated that a nudge is inconsistent with the nudgee’s autonomy if it is manipulative,Footnote 7 and the target has not consented to it. Kapsner and Sandfuchs (2015, this issue) argue that insofar as nudges infringe on privacy, they also infringe on autonomy. Bovens (2009) raises the worry that nudges may damage the capacities of reflection, and thus infringe the autonomy of decision-makers in the long-run. It must be noticed that even nudges who can foster the autonomy of most agents will decrease autonomy for some other individuals (White 2013). Personalised nudges could theoretically solve this problem, but are still for now largely unrealized (Felsen and Reiner 2015, this issue). These concerns suggest that nudges fare less well with respect to autonomy than educational or boosting measures.
Several authors have however claimed that some specific nudges can respect autonomy: Cohen (2013) analyses situations of informed consent (but see Blumenthal-Barby 2013), Barton (2013) examines the case of tobacco health warnings, and Trout (2005) investigates debiasing techniques. In this volume, two articles analyse in details some specific nudges and argue that they can respect autonomy. First, Nagatsu (2015, this issue) investigates pro-social nudges, specifically considering the example of the “Don’t mess with Texas” anti-littering campaign. He identifies two mechanisms by which such a campaign may work: by changing expectations that others will not litter (and expect them not to litter), and by stimulating a frame switch, as defined by Bacharach (2006), from an “I-frame” (in which the agent asks himself: “what should I do?”) to a “we-frame” (in which he reflects on the question: “what should we do?”). He argues that those two mechanisms are compatible with two important aspects of autonomy: they do not cause incoherent mental states, and they are compatible with a responsiveness to reasons. Second, Mills (2015, this issue) argues that a nudge is compatible with autonomy as long as it satisfies four conditions: being intended to facilitate the nudgee’s pursuit of her own goal, having an acceptably low opt-out cost, and satisfying conditions of publicity (see below for an analysis of the condition of publicity) and transparency. Mills explains how personalisable default rules, choice prompts and framed information provision satisfy such conditions. This holds across various definitions of autonomy, whether construed as authentic self-rule, or according to a hierarchical or relational account.
Other arguments explain more generally how nudges can actually improve autonomy. On the base of neuroscientific models, Felsen and Reiner (2015, this issue) argue that if a nudge counteracts a bias, it may be seen as increasing the autonomy of the decision, because it promotes a choice that is in line with higher-order desires. Sunstein (2013) argues that having to make all relevant choices by ourselves could make us worse off, and even less autonomous: given “limitations of time, interest and concern”, a choice architecture that guides us towards good decisions would enable us to freely concentrate on the matters that we deem as the most important for us. On a similar note, Mills (2015, this issue) holds that even if some nudges (e.g., default rules, choice prompts and framed information provision) exhibit epistemic paternalist tendencies, they do not contravene on personal autonomy, as insisting on keeping the agent in a state of full epistemic independence would prevent him from being able to pursue important goals. These arguments suggest that at least in these cases, nudges might be on a par, or even supersede, education and boosting measures with respect to autonomy.
Finally, “autonomy sceptics” raise doubts about the value of the concept of autonomy: Saghai (2013) has argued that definition of the preservation of freedom of choice in terms of autonomy (or liberty) would generate confusion, given the lack of consensus on the meaning of autonomy—and he instead proposes to focus on the conditions of choice-set preservation, and full or substantial noncontrol. On the other hand, some sceptics accept the usefulness of this concept, but suggest that the high respect in which we hold autonomy might be misguided. Sunstein and Thaler (2003) suggest that autonomy could be overriden on consequentialist grounds in some settings, and Verweij and Hoven (2012) have argued that many public health policies that restrict choice involve only minor restrictions to personal autonomy. More extremely, Sunstein (2013) suggests that autonomy concerns may actually simply function as a welfare-improving heuristic; on this view, infringement with autonomy may not be a serious problem, as long as we are guided towards welfare-promoting ends. Felsen and Reiner (2015, this issue) argue that neuroscientific evidence suggests that most everyday decisions are not free from undue external influence, which are often integrated in our decision processes in a covert way. Thus, most of our everyday decisions would not count as autonomous according to some conceptions of autonomy, and they ponder whether the high regard in which we hold autonomy may be misguided.
Publicity and Type-Transparency
Another concern about nudge policies is that because of their covert nature, some nudges are more difficult to scrutinize or monitor (Glaeser 2006). Sunstein (2013) recognizes that some nudges may lack some salience, in the sense that they do not attract the same attention as mandates and bans. We have mentioned in part 2 that if nudges go unnoticed, this constrains the opt-out option; but this also have the effect of undermining social or political resistance. To counter this worry, Thaler and Sunstein (2008) invoke a Rawlsian notion of publicity, according to which policy-makers can only use policies that it would be able and willing to defend publicly to its own citizens. This condition has been criticized in two respects. First, Hausman and Welch (2010) and Lepenies and Malecka (2015, this issue) have raised doubts about its Rawlsian credentials. Second, Lepenies and Malecka (2015, this issue) argue that this condition is insufficient. They explain that nudges are problematic for two reasons: most of them are not part of a legal system; and no nudge—by nature—does impose a requirement to behave in a particular way. Nudges thus rely on an instrumental conception of the law that is problematic on an institutional level. These worries suggest that with respect to transparency, nudges fare less well than legally enshrined incentivising or coercive measures.
As a matter of fact, the publicity condition is hypothetical: it only states that policy-makers can use policies that they would be able to publicly defend, but it does not request policy-maker to actually defend them publicly. Lepenies and Malecka (2015, this issue) argue that connecting nudges to the legal system would make nudges more visible and accessible,Footnote 8 and they suggest various tools in this respect: nudge ombudsman, legal registry of nudge, or information about the legal source of shocking health warnings. Using Bovens (2009) vocabulary, one could analyse such measures as improving the type-transparency of nudges. To summarize, whereas token-transparency is important at the individual level, in order to ensure that the opt-out clause can indeed be chosen (see part 3.1.3), type-transparency is important at the institutional level, in order to complement Thaler & Sunstein’s condition of publicity.
Normative Justifications: Conclusion
To summarize, libertarian paternalism is only one of the possible justifications of nudging policies, and not always the less questionable; in particular, as argued by Guala and Mittone (2015, this issue), some nudges that are both pro-self and pro-social might be easier to ethically justify because of their latter effect than because of the former. On the other hand, first survey results suggest that pro-self nudges are more acceptable to the public than pro-social ones (cf. Hagman et al. 2015, this issue), in which case a democratic justification might have to focus on the nudges’ pro-self effects. In any case, to be fully justified, nudges should be compared to possible alternative measures (like education, boosting or classical paternalistic measures) along several criteria that include their transparency - to ensure that the opt-out condition can indeed be chosen, and to enable the society to contest their use - as well as their general effects on autonomy.
We are now going to show that such normative criteria crucially depend on the mechanisms through which nudges operate—a question connected to the larger issue of the evidential support of nudging policies.