Punishing a person based on low unconditional credence in their deservingness to be punished is consistent with retributivist deontological principles. Negative retributivism absolutely prohibits the intentional or knowing infliction of undeserved harm on individuals identified as undeserving, not the intentional or knowing infliction of risks of undeserved harm on individuals. Meanwhile, the knowing infliction of undeserved harm on some unidentified individuals generates not overriding reasons against punishment, but pro tanto reasons against punishment that are to be weighed against other non-overriding reasons for punishment like crime prevention. The upshot is that uncertainty regarding any identified person’s deservingness to be punished does not entail that punishment is generally impermissible if negative retributivism is true. One might be misled into thinking that impossibly high levels of unconditional credence in individual desert claims is morally required by failing to distinguish our actual criminal law practices, which are extremely harsh and unjustifiable, from criminal law as it ought to be.
Adam Kolber suggests that negative retributivism requires impossibly high degrees of credence in individual desert claims for punishment to be morally permissible.1 My aim in this response is to argue otherwise. Negative retributivists can tolerate high levels of uncertainty regarding individual desert claims—claims to the effect that a particular individual deserves to be punished.
There are many different views that fly under the banner of retributivism. To avoid confusion, I will focus on one version of retributivism that I find especially plausible:
retributivism: Punishing a person who does not deserve to be punished is morally impermissible2
Retributivism, so characterized, articulates a negative constraint on permissible punishment.3 It is a deontological ethic: certain forms of conduct are absolutely forbidden. Of course, various familiar forms of consequentialist moral reasoning are in principle consistent with retributivism. For example, it is good to set punishments with an eye towards future crime prevention. The retributivist will set punishments to achieve those good ends. But unlike the pure consequentialist, the retributivist will only pursue future crime prevention subject to a strict moral side constraint: the good consequences cannot be purchased at the cost of punishing those who do not deserve it.4
Kolber thinks retributivism has the inconvenient upshot that it is never permissible to punish, given uncertainty about various matters that bear on whether persons, in general, deserve to be punished.5 For example, a person’s deservingness to be punished turns on whether they had a robust capacity to choose not to commit crimes and thereby avoid punishment—call this capacity “free will.” Given our evidence, it is highly uncertain whether any individual has the kind of free will that makes one deserving of punishment for wrongdoing. Kolber refers to the resulting uncertainty as moral uncertainty,6 but it is worth distinguishing non-moral or empirical uncertainty that turns out to be morally relevant from pure moral uncertainty. Uncertainty regarding whether we have free will is uncertainty regarding the truth of an empirical non-moral proposition, albeit one that is morally relevant. Uncertainty regarding whether retributivism is true would be a case of pure moral uncertainty.7 Our concern is solely with morally relevant non-moral uncertainty (concerning freedom of will, for example) that bears on moral questions (like whether a person deserves to be punished).
Kolber thinks that retributivism—a prohibition against inflicting undeserved punishment on persons—amounts to a prohibition against punishing a person unless one has a high degree of credence that they deserve it. We can state the relevant principle precisely (the label and choice of phrasing are mine):
certainty: Punishing a person without at least 90–95% credence that the person deserves to be punished is morally impermissible.
Retributivism does not entail certainty. Retributivism reflects what is or is not morally permissible from the objective point of view. But, it does not say what is or is not permissible given epistemic uncertainty.8 For example, it might be morally permissible to punish someone with only 60% credence in their deservingness to be punished, even if it is morally impermissible to punish them when they don’t deserve it.9 For suppose it turns out that, notwithstanding our uncertainty, Sam deserves to be punished as a matter of fact; then, punishing Sam based on 60% credence in his deservingness is strictly consistent with retributivism. Even if there is no logical connection between the two theses, it remains very tempting to suppose that the values underlying retributivism have implications for what is or is not morally permissible under conditions of uncertainty.
If there are moral reasons to embrace certainty, then free will skepticism generates a problem for the retributivist. Suppose that it is only reasonable to have 50% credence in the proposition that persons have free will (in fact, Kolber thinks a higher credence is appropriate), and that without free will no one deserves to be punished (a common assumption of retributivists). Then, a retributivist can have at most 50% credence in the proposition that a given individual deserves to be punished. Certainty, combined with facts about what is reasonable to believe about individual desert claims, entails that it is always morally impermissible to punish. This, Kolber suggests, is a hard pill to swallow.
Certainty, or the general idea that a high degree of credence in individual desert claims is required by retributivism, plays a critical role in Kolber’s argument that retributivism is inconsistent with punishment under uncertainty. But, we have reasons for doubting certainty.
The Values Implicit in BARD
Begin by considering Kolber’s own reasons for thinking that the retributivist is committed to certainty. Kolber places considerable weight on the fact that at criminal trials we require jurors to determine beyond a reasonable doubt that a defendant committed a criminalized act before punishing him. The BARD standard is routinely interpreted as requiring 90–95% credence,10 and BARD seems like a good standard to use at the guilt phase of a criminal trial. In these facts, Kolber finds reasons for the retributivist to embrace Certainty. But there are not such reasons.
First, BARD is a legal requirement, and not obviously a moral one. Whenever we set up institutions which prescribe burdens and benefits on the basis of pre-determined rules, these institutions give rise to questions of institutional desert that are not necessarily questions of moral desert.11 James Harden has a claim of institutional desert to being awarded the NBA’s most valuable player award in 2018 for his outstanding play. But he does not necessarily deserve the award in a deep moral sense. There are plenty of reasons for an institution like the criminal legal system to require high degrees of confidence in individual claims of institutional desert that aren’t equally reasons for having high credence in claims of individual moral desert. For example, legal institutions have greater legitimacy when they ensure that their rules—like the rule prescribing life imprisonment for murderers—are applied with a great deal of precision (when and only when their triggering conditions are met).
Second and more importantly, even if BARD at trial reflects an important moral value concerning individual moral desert enshrined in the legal system, the value does not support certainty. We must distinguish the conditional credence that is morally required in an individual’s deservingness to be punished from the unconditional credence that is morally required. An agent’s unconditional credence in a proposition is her degree of confidence that the proposition holds outright. An agent’s conditional credence is her degree of confidence in a proposition assuming the truth of various other propositions. BARD at trial shows, at best, that a high degree of conditional credence in the proposition that a particular defendant deserves to be punished is morally required. Jurors must find a circumscribed set of non-moral propositions beyond a reasonable doubt—that the defendant pulled the trigger, intended to kill, and so on. Jurors are not asked to find beyond a reasonable doubt that the defendant has free will or that the state has sufficient standing to punish.12 In other words, BARD shows at best that a high degree of credence in a defendant’s punishment-deservingness conditional on various non-moral propositions that the jury does not consider (like the existence of free will) is morally and legally required.
There are plenty of reasons why high conditional credence in individual desert claims might be morally required that do not support a moral requirement of high unconditional credence. For example, we want to punish people in ways that maximize various positive social consequences—like crime prevention—while minimizing various harms. Recall that this kind of consequentialist reasoning is entirely compatible with retributivism. Retributivism operates as a side constraint on consequentialist welfare maximization. High conditional credence might be necessary in order to maximize the benefits of crime prevention while minimizing the costs of imprisonment. Efficient and effective crime prevention, for example, does not require having high unconditional credence in punitive desert, but it does require high conditional credence: to send targeted messages of deterrence to would-be murderers, we must make sure we are punishing murderers, in particular.
To Kolber’s credit, he does acknowledge that the BARD standard at trial applies to a narrow range of questions that the jury is tasked with deciding, not the question of overall deservingness. But, he misconstrues the significance of this fact. Kolber imagines a retributivist arguing that high conditional credence is a way “to compensate for [overall] moral uncertainty. Precisely because there is so much uncertainty when it comes to criminal justice, we ought to be awfully certain that the accused is factually guilty.”13 But, he objects, claiming that “we cannot let high levels of certainty in some areas bolster low levels in others.”14 The imagined objection and response misconstrue the crucial point. It is not that high conditional credence at trial is a “corrective” for low unconditional credence. It is true that even if the jury is 100% certain that the defendant committed murder, it need not “bolster” or improve our unconditional certainty that the murderer deserves his punishment. What is mistaken about Kolber’s analysis is the background assumption that low unconditional credence in any individual person’s deservingness to be punished needs “bolstering” at trial in the first place. BARD does not support this background assumption.15
A Retributivist Analysis of the Problem of Uncertainty
We should set BARD aside and consider the critical moral question directly: is there pressure, internal to retributivism, to embrace certainty? In other words, supposing that it is morally impermissible to punish someone who does not deserve to be punished, is it morally impermissible to punish someone without a high degree of credence that he deserves to be punished?
A key animating principle behind retributivism is the idea that the certain forms of conduct involve objectionable instrumentalization of persons. Individuals with moral status should not be used as mere means. Certain interests and rights of persons must be respected when we act in ways that affect them. Failing to respect the relevant interests (that are typically urgent, like an interest in physical integrity and in being free from absorbing pain or emotional trauma) in order to promote the lesser interests of a large number of others involves unacceptable instrumentalization.
Whether a form of conduct involves objectionable instrumentalization turns, among other things, on the nature of the harm inflicted. There is a morally relevant difference between intentional or knowing infliction of undeserved harm on a person and the knowing infliction of a risk of undeserved harm. A person who is certain that my φ-ing (where φ is an act) will result in her being put in prison for life even though she does not deserve it has a very serious complaint against my φ-ing. Ignoring the person’s complaint risks showing inadequate concern for her interests. By contrast, a person who knows only that my φ-ing produces a risk that she will be put in prison for life even though she doesn’t deserve it does not have as severe a complaint against my φ-ing (though she may still have a complaint).
The deontological side constraint that retributivists treat as inviolable is plausibly one that prohibits intentional or knowing inflictions of undeserved punishment, not mere risks of undeserved harm.
subjective retributivism: Punishing someone while knowing or intending that the punishment is undeserved is morally impermissible16
A retributivist who has only 50% unconditional credence that a person deserves harm does not intentionally or even knowingly inflict undeserved harm for social benefit, given plausible assumptions about the high degree of credence generally required for knowledge.17 It is therefore possible for a retributivist to honor her core deontological scruples despite punishing persons when she is only 50% unconditionally certain that they deserve it. Retributivist deontology, it must be remembered, is an “agent-centered” ethic—the wrongness of certain courses of action is grounded not simply in how the putative victim is affected by the action, but the kind of attitude towards others reflected in taking the action. The distinction between knowing or intentional infliction of harm and unknowing infliction of harm matters precisely because only a limited set of objectionable attitudes towards others are strictly inconsistent with the respect that is fundamentally owed.18
One might object to this line of reasoning. Surely, given the law of large numbers, even if the retributivist is only 50% certain that any given individual deserves to be punished, she should be 100% certain that some individuals will be punished who do not deserve to be. All it takes is for five individuals to be punished and we can be 95% certain that at least one of those five is undeserving.19 So, it seems, retributivists who punish with only 50% unconditional credence in individual desert claims knowingly inflict unjustified harm on someone (in fact a great many), in clear violation of Subjective Retributivism.20
This is a problem that, in its more general form, has received extensive discussion in recent work on rights and risk. The distribution of a vaccine which prevents a non-deadly but debilitating disease in the population but has a 1% likelihood of killing someone guarantees that someone will be killed in a large enough population as a result of the vaccine’s use.21 A deontologist against objectionable instrumentalization needs to explain why it would be permissible to administer the vaccine notwithstanding the fact that the harm to the one who dies would be greater than the benefits to the rest of the vaccine users. This example serves as a reminder that even with 95% unconditional credence in individual desert claims, a retributivist can be certain that someone will be undeservingly punished in a large enough population. So, it cannot be enough, by Kolber’s own rationale, to render punishment absolutely impermissible given retributivism that some unidentified persons will be undeservingly punished.
Here we can rely on an important point made by Johann Frick. Frick, in response to cases like the vaccine case, points out that there is a morally relevant difference between identified and statistical lives.22 The implementation of the vaccine results in an unidentified statistical life lost. This is quite different from a situation where we know the identity of the person who will die if the vaccine is distributed and distribute the vaccine to her anyway. Why does this distinction morally matter? One way to motivate its moral significance is by appeal to a standard contractualist test for determining when individuals have been objectionably instrumentalized.23 Plausibly, persons are objectionably instrumentalized when the harms inflicted on them by a course of action are such that they would reasonably object.
But in considering whether individuals would have reasonable objections, we consider not statistical or hypothetical persons but actual individuals and their interests. In the case of punishing wrongdoers with only 50% unconditional confidence that they deserve to be punished, we need to consider whether such punishment would be defensible from the perspective of the actual wrongdoer. Crucially, from any actual wrongdoer’s perspective, he cannot be certain that his punishment would be undeserved (far from it). This is no artificial constraint on his epistemic situation. His epistemic situation is the same as ours: he can only be certain that someone will be undeservedly punished, not that he will be the one undeservedly punished. In other words, he is not identified as undeserving. Clearly, a wrongdoer who is identified as undeserving of punishment but is punished anyway has a stronger complaint against his treatment than a wrongdoer who is punished and knows that there is a 50% risk of his punishment being undeserved but 50% odds of his being deservedly punished (an outcome we can assume he cannot reasonably object to).
So far, we have noted only that the infliction of risks of undeserved harm on identified persons is less morally problematic than the infliction of certain undeserved harm on such persons. But this does not mean that risks of undeserved punishment inflicted on identified individuals—that is, individuals identified as bearers of the risk of undeserved punishment—can be neglected by the retributivist. The risk of undeserved harm inflicted on individuals needs to be considered by the retributivist in deciding what to do. There is serious disvalue in high numbers of unidentified individuals being undeservedly harmed. But the crucial point is that on deontological theories like negative retributivism there is an important difference between deontological reasons (reasons of basic justice) and reasons of overall welfare. Deontological side constraints—like the prohibition against intentionally or knowingly inflicting undeserved harm on individuals identified as undeserving—are inviolable. They provide decisive or overriding reasons against certain courses of action. By contrast, the reasons that bear on overall welfare provide non-decisive reasons for action. If there are many cases of unidentified individuals being undeservedly harmed in our legal system, this has disvalue for the retributivist and counts as a pro tanto reason against our punitive practices. But the disvalue does not create decisive or overriding reasons against punishment.
The disvalue of undeserved harm inflicted on unidentified individuals must be weighed against other values, like crime prevention. While knowing infliction of undeserved harm on identified individuals cannot be justified by appeal to positive social consequences, cases of unidentified undeserved harms that are knowingly inflicted can be justified by appeal to positive social consequences. Cases of undeserved suffering have some disvalue, but so does the future suffering of others. When we punish murderers, this may generate a 50% risk of undeserved harm, but failing to punish may result in a 90% risk of undeserved harm to future victims.
The precise degree of unconditional credence in any given person’s punishment-deservingness that retributivism requires will therefore be a function of the consequentialist balance: the weighing of all the reasons of wellbeing that count for or against some particular standard. We have good reasons for believing that the standard that is welfare maximizing will not be anything close to beyond a reasonable doubt (because this would result in no punishment given uncertainty). We can be sure, at the very least, that familiar retributivist side constraints on welfare maximization through punishment do not generate a reason for requiring more than 50% unconditional credence, for 50% unconditional credence falls well short of knowledge that a punished person is undeserving. And, as I have argued, the strict prohibition on standard retributivist views is best understood as a prohibition against the knowing infliction of undeserved harm on identified persons.
Kolber in conversation raises the following objection to my overall view:
Suppose that there’s a new public policy proposal that will save one million lives each year, but, unfortunately, it will require us to increase our punishment error rate a bit. Indeed, it will cause precisely 100 more innocent people to spend life in prison than would otherwise. These one hundred people weren’t targeted in anyway, and almost all people would prefer the substantial improvement in life expectancy from the plan relative to the rather tiny chance of being erroneously punished.
The negative retributivist might accept this deal on consequentialist goods. But now suppose that the policy makers ask the retributivist whether she would like to know who the 100 people would be. Kolber imagines my retributivist covering her ears, for identification would dramatically change the moral situation and prevent her from accepting the policy. Kolber is right that this reaction would be absurd. The reason it would be absurd is that in refusing to know the identities of the 100 innocents, the caricatured retributivist is imposing an artificial constraint on her epistemic situation. If members of the larger population were surveyed for their objection to the policy, they might reasonably ask: do we know who the innocents will be, and, more importantly, would I be one of the innocents? It is no response to this concern to say: we have the names written somewhere but we are not looking so we cannot tell you. The critical difference between this case and the standard case of undeserved punishment is that information concerning who the undeserving is not available, either to the policy makers or the individuals being punished. The lack of knowledge concerning the identities of the undeservingly punished is not the result of a failure of enquiry.24
A Cautionary Note Concerning Our Actual Criminal Law Practices
In conclusion, I would like to suggest a reason why one might be tempted to think any moral theory that assigns significant disvalue to cases of undeserved punishment should be pushed towards a high unconditional credence requirement. One reason is our actual criminal law practices present a serious justificatory challenge. They must be hard to justify, given the extreme severity of the sentences routinely issued to criminals. Indeed, Kolber writes as though retributivism needs to vindicate actual criminal law practices.25 Much of the force of Kolber’s argument turns, I think, on the expectation that retributivists need to show that these practices accord with retributivist principles given uncertainty. This is an unreasonable expectation.
Given the exceptional harshness of criminal sentences in the United States, retributivists should be appalled by our criminal law practices, which do not just involve using people merely as means to further social goods, but also involve cruelty, sadism, and dehumanization.26 Our practices are not a product of reasonable retributivism. Rather, our practices likely give expression to our vengeful instincts, which are not moral. In the case of extreme punishments, it becomes much more plausible to think that very high (close to certain) degrees of certainty in a defendant’s deservingness might be required to make such treatment permissible. But in a system where the poor are treated better, where persons receive better opportunities to develop their moral capacities, where vengeance has no place in criminal justice, and even the worst wrongdoers are treated with decency and compassion, it would be less plausible that near certainty is required when it comes to unconditional credences in individual desert claims. For example, if we had a system of humane quarantine that nevertheless inflicted some relatively small harms on criminal wrongdoers, retributivists would be far less worried about unidentified cases of undeserved harm because the harm would not be as high.
I prefer a retributivism that would choose humane but unpleasant quarantine over harsh sentences every time, constrained by the prohibition against knowing infliction of undeserved harm on identified lives. Kolber’s chief result—that our actual criminal law practices are inconsistent with negative retributivism given uncertainty—far from being a reductio of retributivism is a conclusion that retributivists should embrace. It does not show, however, that a system of punishment that accords with retributivists principles is in principle impossible given high levels of moral and non-moral uncertainty. The realization of such an ideal system remains the single most important criminal law objective of our time.
* Princeton University, Dept. of Philosophy, 1879 Hall, Princeton, NJ 08544.
1. Adam J. Kolber, Punishment and Moral Risk, 2018 U. Ill. L. Rev. 487 (2018). Thanks to Adam for a productive and enjoyable exchange on this important issue.
2. See, e.g., Michael S. Moore, The Moral Worth of Retribution, in Responsibility, Character, and the Emotions: New Essays in Moral Psychology 179 (1987) (“Retributivism is the view that punishment is justified by the moral culpability of those who receive it.”).
3. This is sometimes called negative retributivism. The defining feature of this view is the strict prohibition against undeserved punishment. Kolber is targeting a range of different views, including the one described. The view that Kolber treats as paradigmatically retributivist is one that treats desert as a sufficient condition for punishment, not just a necessary condition. What I say in defense of negative retributivism applies with full force in the case of this alternative position, so long as it is consistent with ends like crime prevention being treated as valuable and as pro tanto reasons to punish (subject to the desert constraint).
4. Compare Kant’s absolute prohibition on lying, even when lying would save countless lives. See Immanuel Kant, “On a Supposed Right to Lie from Altruistic Motives”, in Critique of Practical Reason and Other Writings in Moral Philosophy (Lewis White Beck ed. & trans., Univ. of Chi. Press 1949) (1797).
5. See Adam J. Kolber, Punishment and Moral Risk, 2018 U. Ill. L. Rev. 487, 491 (“Reasonable retributivists are likely to have too much uncertainty to justify punishment.”).
6. Id. at 489.
7. See T.M. Scanlon, Being Realistic About Reasons 30–40 (2014) (discussing pure vs. mixed normative claims).
8. For a helpful discussion on the difference between subjective and objective moral oughts, see Yoaav Isaacs, Duty and Knowledge, 28 Phil. Persp. 95 (2014). See also Elizabeth Harman, The Irrelevance of Moral Uncertainty, 10 Oxford Stud. Metaethics 53 (2015); Andrew Sepielli, How Moral Uncertaintism Can Be Both True and Interesting, 7 Oxford Stud. Normative Ethics 98 (2017).
9. If it turns out they deserve it, then punishment based on 60% credence is consistent with retributivism.
10. Cf., Larry Laudan, Truth, Error, and Criminal Law: An Essay in Legal Epistemology 44 (2006) (stating that 90% or 95% are “commonly cited unofficial estimates” of the standard of proof for criminal convictions; Margaret Raymond, The Problem with Innocence, 49 Clev. St. L. Rev. 449, 451 (2001) (exploring a hypothetical 99% beyond a reasonable doubt standard).
11. For a discussion of the distinction, see T.M. Scanlon, Why Does Inequality Matter 117–32 (2018).
12. I am setting aside the complex issue of jury nullification and the special context of death sentencing, about which I have written elsewhere.
13. Kolber, supra note 1, at 516.
15. Kolber, in conversation, wonders whether my view of BARD involves a reinterpretation or a rejection of the traditional Blackstonian view. I think my view might be consistent with Blackstone’s view, although what Blackstone’s view is is not entirely certain. “[T]he law,” we are told, “holds that it is better that ten guilty persons escape, than that one innocent suffer.” 4 William Blackstone, Commentaries on the Laws of England 352 (1979). Benjamin Franklin goes further: “it is better [one hundred] guilty persons should escape than that one innocent person should suffer.” Letter from Benjamin Franklin to Benjamin Vaughan (Mar. 14, 1785), in The Writings of Benjamin Franklin (1970). Note that the concepts of guilt and innocence are not obviously tracking notions of unconditional desert. It requires an argument to show that Blackstone’s observation is equivalent to the claim: “it is better that ten guilty persons escape, than that one innocent suffer.” 4 William Blackstone, Commentaries on the Laws of England 358 (1979). Absent such an argument, I think my view of the values implicit in BARD is consistent with the traditional view.
16. For a knowledge-first approach to deontology generally, see Isaacs, supra note 9.
17. Thanks to Gideon Rosen for discussion on this point.
18. The harm of instrumentalization can be conceived in “victim-centered” terms, where it is entirely a function of the effects on the victim, and in “agent-centered” terms, where it involves a particular objectionable attitude directed towards another. Kantian retributivism, for example, is an agent-centered morality. Certain ways of regarding others are absolutely prohibited, because they are inconsistent with the fundamental regard we owe others in recognizing their humanity. Thanks to Alisabeth Ayars for pushing me to make this clear.
19. This requires assuming the independence of the individual punitive outcomes, which may not be a reasonable assumption if the uncertainty concerns free will, because the outcomes are dependent, but we can set aside this complication.
20. Thanks to Adam Lerner for the objection.
21. The case is discussed in Johann Frick, Contractualism and Social Risk, 43 Phil. & Pub. Aff. 175 (2015).
23. See id. (discussing the test). The contractualist approach to instrumentalization is to be preferred to the Kantian universalization test, which is notoriously difficult to apply.
24. I recognize that here lie dragons: it is a hard question when lack of knowledge is attributable to the knower in the sense that the knower bears responsibility for the uncertainty and when it is not. We do not need a theory of this distinction for present purposes. The point is just that there is a distinction that is manifestly important from the moral point of view.
25. Kolber, supra note 1, at 516 (“And matters get worse for retributivists when we consider how a real-world legal system could possibly reflect a sufficiently high justificatory standard of proof.”).
26. See James Q. Whitman, Harsh Justice: Criminal Punishment and the Widening Divide Between America and Europe (2003).
The full text of this Article is available to download as a PDF.