Respectful Debiasing in Democratic Deliberation

ABSTRACT

Democratic deliberation about health policy requires, as one of its inputs, ethical judgments from members of the public. These judgments may, however, be biased and unreliable—they may be systematically distorted by framing and emotion effects, by the effects of self‐interest, and by a whole host of other cognitive and affective biases. Supposing we find evidence of biased ethical judgments, can we deal with these judgments in ways that promote reliability while approximating, as far as possible, the moral and political ideals of deliberation? In this paper, I explore this question by outlining and evaluating three respectful debiasing methods, which aim to mitigate bias but in a morally and politically respectful way. In brief, the methods are: (1) Improving the conditions of elicitation but accepting whatever judgments that result, (2) Suggesting to specific members of the public that they may be biased and leaving them to decide whether to debias, and (3) Surveying the public's standards for bias and applying these standards to their ethical judgments.

In democratic deliberation, members of the public come together, on an equal footing, to communicate with each other about their preferences, values, and interests over political issues and policies of common concern, and sometimes also to vote or agree on a policy decision [1]. Calls for deliberation are becoming increasingly prevalent in health and science policy, with members of the public being consulted for their judgments about issues like the allocation of healthcare resources [2] and public health measures like pandemic prevention and control [3].

Yet democratic deliberation faces issues with irrationality and bias in the deliberators' reasoning. Of particular interest to bioethics are biases affecting the public's ethical judgments, which form a significant input into deliberation about health policy. These judgments could be systematically distorted by framing and emotion effects, by the effects of self‐interest, and by a whole host of other potential biases [4]. In this paper, I focus on the cognitive and affective biases—that is, on the pervasive, systematic distortions stemming from our cognitive system and from our emotions [4]—that affect our ethical judgments. Consider two concrete examples:

• Identified Victims: Suppose we are considering how to allocate scarce medical resources, and must decide between funding Treatments A and B. These treatments vary in different dimensions, but importantly, A has beneficiaries whose plight has been featured in the news, whereas B's beneficiaries have not. In a citizen's jury, some members of the public judge that ethically speaking, we should fund Treatment A. Some argue that these judgments may be the result of bias toward identified victims, making these judgments unreliable and untrustworthy. ¹

• Framing Effects: Suppose we are deciding between two policies for preventing the spread of a deadly infectious disease, and we solicit the public's ethical judgments about them. We find that some members of the public give different verdicts under different ways of framing the choice—for example, when we frame the choice in terms of lives saved they choose policy A; and when we frame the choice in terms of lives lost, they choose policy B. Some argue that their judgments are affected by framing effects—they give inconsistent responses to seemingly equivalent ways of framing a case—and are hence unreliable and untrustworthy [6, 7].

Identified Victims and Framing Effects involve potentially biased ethical judgments, which could lead to poorer quality deliberation and worse outcomes. Other biases that might affect ethical judgments include order effects (where the order in which cases are presented affects our ethical judgments), status quo bias (where the actual, current state of affairs irrationally influences our ethical judgments), and self‐interested bias (where our potential to gain/lose from making some judgment influences our ethical judgments) [8, 9, 10]. Biases affecting ethical judgments deserve a separate analysis from those affecting non‐ethical judgments (particularly prudential judgments) which have thus far been the focus of much work in practical bioethics. For instance, see the large literature on the ethics of nudges [11, 12, 13], which could be seen as interventions which use biases in prudential judgments to help people make better decisions.

Deliberative democrats are well‐aware of the problems of bias—especially confirmation bias and motivated reasoning, selection bias in composing the deliberators, and some kinds of framing effects [14, 15, 16, 17, 18, 19]. Some argue that bias poses a problem for deliberation because it undermines our ability to revise our conceptions of value in response to new evidence and counterarguments; furthermore bias could prevent us from treating fellow deliberators as free and equal [15, 19]. Others propose that deliberation itself, when performed by deliberators with diverse opinions, can counteract biases like confirmation bias [16]. Furthermore, deliberative democrats have proposed general debiasing strategies, such as more equal recruitment of deliberators; encouraging reflexive, civil, and equal participation; providing more balanced background information as well as information about those with whom one is deliberating [19]; and specific debiasing strategies, like counter‐framing and disaggregating the claims and implications contained within a frame [15]. However, I believe the problem of bias has not been completely resolved in the sense that we have yet to work out, for each potential bias, what the most effective and ethical debiasing method is. On the point of efficacy, biases differ in their triggers and mechanisms, so a debiasing method that is effective for one bias may not work similarly well for another (compare the myriad differences between confirmation bias and identified victims bias, for instance). ² On the point of ethical appropriateness, most authors have not yet reckoned with the ethical status of debiasing methods, which (as I argue below) could raise distinctive ethical concerns by themselves.

Biased ethical judgments are importantly special because: first, it is harder to demonstrate distortion and error in ethical judgments, especially against a backdrop of reasonable ethical disagreement—depending on which background ethical theory we endorse, we may deem a particular judgment to be biased or truth‐tracking [21, 22]. This is unlike the prudential case where we may legitimately assume certain widely shared goals (like health) that then support a less controversial standard for evaluating judgments. Secondly and relatedly, we are in a poorer epistemic position when ascertaining moral as compared with non‐moral facts—because of moral disagreement, and because of the lack of agreed‐upon method for finding out about the moral facts [23, 24]. Thus we should be more cautious about diagnosing bias in ethical judgment, and should practice epistemic humility by considering whether apparent biases may in fact be justified responses. For instance, Daniels [25] considers several moral arguments for favouring identified victims—for example, because a failure to do so may lead people to feel greatly troubled and thus negatively affect welfare, or it may blunt peoples' motivation to rescue others. Gigerenzer [26], Mandel [27], Dreisbach and Guevara [28] and others point out that framing effects may be rational and stem from the exercise of useful heuristics, because frames may create justified inferences or implicatures [29] about which option is recommended, about the numerical bounds of possible outcomes, and about the moral valence of options. Thirdly, ethical judgments may have a privileged status in democratic deliberation, requiring special respect that isn't similarly required toward non‐ethical judgments. ³ There is thus a need to consider how, if we ever, we should debias the ethical judgments going into deliberation.

Two obvious ways of responding to biased ethical judgments are to: (1) Evaluate and Exclude: identify the unreliable ethical judgments and exclude them from influencing deliberation, ⁴ and (2) Accept As Is: allow these judgments to influence policy decisions despite their potential unreliability. Evaluate And Exclude promotes epistemic reliability, whereas Accept As Is may embody respect in deliberation about public policy. These two options lie on extreme ends of a spectrum, each honoring important values at stake in deliberation—moral and political ideals like respect and equality, and epistemic goals like improving knowledge or reaching better decisions [1, 31, 32, 33]. In between these options; however, lie further alternatives which are less often considered, but which nonetheless may be candidates for the best compromise between them. In this paper, I explore and evaluate these alternatives—which I'll call methods of respectful debiasing, because they aim to mitigate bias or unreliability, but in a morally and politically better way.

This investigation is important for several reasons: first, it aims to systematically chart out a landscape of options for dealing with bias, options which aren't simply Evaluate and Exclude or Accept as Is. Secondly, it seeks to understand the fine‐grained moral tradeoffs between them, which will in turn influence our stance on the more general practical question of how best we should deliberate about policy—which plausibly requires promoting reliable deliberation while approximating moral and political ideals. Third, the debiasing methods considered here have a broader range of application, beyond democratic deliberation—we might imagine adjacent debates about whether and when we should debias expert deliberation in policy discussions and clinical ethics committees too. ⁵ Fourth, debiasing methods may interact with other ethically relevant outcomes adjacent to deliberation: for example, debiasing also affects, and is affected by, polarization in policy and in research. On the one hand, effective debiasing methods hold the potential for reducing political polarization, when such polarization is created and sustained by biases in policy‐related attitudes like confirmation bias, myside bias, and various group‐based biases [34]. On the other hand, polarization in scientific research [35, 36] about biases may complicate the issue of debiasing, because researchers investigating bias might polarize and cluster around opposing camps with respect to whether some phenomenon reflects a bias or a justified response, which could violate norms of scientific research, and could generally have problematic impacts on deliberation in science and policy [35].

The rest of this paper proceeds as follows: I start by outlining three respectful debiasing methods. I then consider how each debiasing method evaluates the agents to be debiased, and whether such evaluation disrespects them. Next, I examine the effects of each debiasing method, particularly on whether they silence an agent's contribution to deliberation. Finally, I further compare two debiasing methods that were roughly tied in the previous two dimensions.

1. Three Respectful Debiasing Methods

I start by outlining three methods of respectful debiasing. ⁶ First, we might Improve Conditions—that is, design the conditions of elicitation of the public's ethical judgments to improve their reliability, but then accept the judgments made in these conditions, whatever they may be. Improve Conditions is different from Accept As Is, in that Improve Conditions explicitly and intentionally designs conditions of elicitation with the goal of improving ethical reliability—whereas Accept As Is merely lands on reliable conditions (if it does) in a purely accidental way. In the Identified Victims case, Improve Conditions might involve making the beneficiaries of different policies equally identified to members of the public—for instance by drawing attention to the plight of the non‐(or less)identified beneficiaries [5], or by damping down the identification of a currently identified beneficiary (for instance by presenting only their abstract features). In Framing Effects, we might Improve Conditions by presenting both frames side by side, or simply use a frame that describes both the number saved and number who die under each policy option [39, 40]. I will understand Improving Conditions in a process‐based way, which says that an improved condition is one that improves the quality of the process of judgment formation—and I set aside a content‐based version that says that a condition is improved if it is more likely to produce judgments with certain content (for instance, if the condition is more likely to produce judgments favouring policy A). ⁷ Concretely, we can Improve Conditions in a mini‐public (often cited as a paradigm institution for promoting democratic deliberation) by intentionally designing the informational presentations with an eye to mitigating potential bias, as well as requiring that facilitators use specific framings/phrases when discussing topics with participants. Similarly, the questions and information provided in deliberative polls may be designed in ways that best elicit unbiased ethical judgments.

Second, we might Suggest Unreliability to those members of the public who hold potentially biased judgments. ⁸ This involves detecting when someone may hold unreliable ethical judgments, pointing this out to them, but leaving them to decide whether to alter these judgments. In Identified Victims, for instance, when members of the public decide in favour of identified beneficiaries, they may be reminded that they could be biased towards them because of identification. Similarly, in Framing Effects, when someone gives inconsistent responses to potentially equivalent ways of framing a policy problem, we can point this out as evidence of potential unreliability. Importantly, however, the members of the public are then left to decide whether their ethical judgments are indeed reliable, and whether (and how) to correct in light of this evidence. Concretely, we might Suggest Unreliability in a mini‐public by having facilitators highlight when certain ethical positions may be the result of bias; furthermore, deliberative polls may be configured to ask questions that detect unreliability, and participants whose answers indicate a risk of bias may be informed privately thereafter.

Third, we can Survey Standards that the public have for bias, and apply these surveyed standards to their ethical judgments. There are two versions of this method: an aggregate version applies a standard for bias if it is sufficiently endorsed by the public as a whole (if more than 50% believe that some bias makes our ethical judgments unreliable, then we apply this standard of bias to everyone's judgments). An individualised version, by contrast, surveys each person's standards for bias, and applies a person's standards to their ethical judgments alone. I adopt the individualised version of Survey Standards because I believe it is more plausible. Applied to the Identified Victims and Framing Effects, Surveying Standards will involve presenting members of the public with vignettes or descriptions of identified victims bias and framing effects, and asking them whether the ethical judgments made therein are biased. ⁹ Having elicited standards from each member, we can then apply these to their respective judgments—for example, if a person believes that the judgments in Identified Victims and Framing Effects are biased, and if their judgments follow a similar pattern, then we exclude them from consideration. A concrete way of Surveying Standards would be to add a question about such standards in deliberative polls, and then to apply the selected standard to participants' answers to other (first‐order) questions in the poll. It is more difficult to implement Surveying Standards in a mini‐public which may not have a formal surveying mechanism, but participants may nonetheless be prompted to think and deliberate about standards of bias that they wish to apply to their judgments about the issue at hand.

In the rest of this paper, I focus on evaluating whether these debiasing methods live up to moral and political ideals of mutual respect, but a debiasing method's effectiveness at mitigating bias is also crucially ethically important. The more effective a debiasing method is, the greater its claim to ethical justification, given its ability to prevent the detrimental effects of bias. Some versions of the debiasing methods considered below have been tested in some domains: (1) Counter‐framing—a specific version of Improving Conditions—has been found to draw participants towards middle‐ground as opposed to extreme positions [43, 44], (2) Debiasing techniques that educate people about the nature and presence of biases—which partially overlap with Suggest Unreliability—have been shown to improve participants' performance on bias and cognitive reflection tests [45, 46], (3) I have, however, been unable to find empirical tests of interventions akin to Surveying Standards. Based on the empirical evidence I found around debiasing, I tentatively deem Improve Conditions and Suggest Unreliability to be more effective than Surveying Standards, which increases their relative ethical justification through the separate ethically relevant factor of effectiveness. However, I emphasise the need to perform more empirical research that compares these debiasing methods more precisely and on equal footing in the specific context of deliberation. ¹⁰

You'll likely have an intuitive feel for how different these proposals are in their ethical status: Improve Conditions tweaks the conditions of judgment prior to elicitation, Suggest Unreliability confronts a person with the suggestion of unreliability, while Survey Standards holds a member of the public against their own standards for bias. I now evaluate these methods and delineate the dimensions along which they differ morally.

1.1. How Debiasing Methods Evaluate Agents

Start with how a debiasing method evaluates those who are being debiased. One concern is that it may evaluate agents in a disrespectful way, treating them as “objects in need of repair” [47, 48] rather than as agents or persons. This could be substantiated using different moral foundations; ¹¹ I rely on Carter's [48] argument that liberal states are required to treat their citizens simply as agents, which entails that these states must treat them in a way that maintains citizens’ outward dignity—defined as the quality of having composure, of being distinct and invulnerable to destructive or corrupting influence. ¹² Carter argues that this generates a moral requirement not to evaluate the agential capacities underlying moral personality past a minimal threshold. If the relevant agential capacities are the two Rawlsian moral powers of having the capacity to form and revise a conception of the good and the capacity for a sense of justice, we get a requirement not to evaluate citizens once they meet the minimal threshold for possessing these two moral powers. Carter [48] argues that states are required to refrain from further evaluations above this threshold, as a form of respect: “we take the subject as given and ask no questions about his or her capacity to pursue the good or to understand the nature of the moral or aesthetic good life”. To engage in further evaluation would undermine citizens' outward dignity.

As an example of such undermining, Carter [48] contrasts two domains where a professor may relate to their student: in the classroom, the professor is morally permitted to assess the student's intellectual capacities and adjust their interactions based on such assessment. Yet when relating to this same student in the public sphere, they should refrain from assessing these capacities, and should not treat such assessments as relevant in their interactions as fellow citizens. Similarly, the state should refrain from assessing the capacity of citizens who pass a minimal threshold of agential capacities—put more pithily, it should treat citizens as opaque. Carter uses this requirement of opacity with respect to draw out surprising constraints on how to pursue equality: it rules out the direct evaluation of each agent's capacities—like their level of autonomy or strength of will—in order to rectify the inequalities revealed. This, he thinks, is objectionable because it evaluates particular agents' capacities to pursue their conception of the good, and must presuppose, for instance, that Jones's level of autonomy is less than Smith's, enroute to promoting equality between them. Carter's account uses the requirement of opacity respect to justify and ground a commitment to treating persons as equals in their basic political entitlements [48]. In my view, this account is also well‐suited for connecting democratic deliberation's ideal of treating deliberators as equals, with its commitment to mutual respect. ¹³

Debiasing involves evaluation of the second Rawlsian power: an agent's capacity for a sense of justice. ¹⁴ It is plausible that evaluating this sense of justice beyond a certain minimal threshold is also morally problematic. We can modify Carter's case to show this: suppose the professor evaluates the student's ability to judge the morality of policies in the public sphere, and argues on this basis that the student's contributions to deliberation should be discounted or excluded entirely. This would be the rough analogue to Evaluate and Exclude, and seems problematically disrespectful because the professor does not treat the student as an agent, a fellow supplier of moral reasons whose judgments are taken as given, but rather sees them as a thing whose moral reliability is in need of repair.

Before evaluating the debiasing methods proper, I tackle two objections to using this account of respect. The first objection draws on analogous debates about rectifying prudential biases and contends that if the state has accurately assessed someone's abilities as lacking, it is not disrespectful to adjust treatment accordingly [55]. I have two responses: first, this objection assumes the accurate diagnosis of bias—but such accuracy is less forthcoming with ethical judgments than with non‐ethical ones, given our worse epistemic position with respect to the moral facts. So even granting the conditional “if accurate diagnosis, then no disrespect”, this objection is weakened by the antecedent being less likely to obtain for biases in ethical judgment. Secondly, even if the practice isn't in fact disrespectful, some members of the public may nonetheless share Carter's intuition and feel disrespected. It is still pragmatically useful to minimize this felt disrespect—even if the feelings themselves are not justified—in order to increase the likelihood of the public accepting the outcomes of deliberation and complying with the resulting policies. ¹⁵ It is thus still useful to investigate how intuitions about respect may distinguish between different debiasing methods.

The second objection resists the intuitive plausibility of Carter's case, arguing that not all cases of evaluation are morally problematic—for instance, we seem permitted to tailor our communications to others in the public sphere, based on an assessment of their abilities. I believe that Carter's account can accommodate and explain the relatively unproblematic nature of this case, by recognizing different degrees of evaluation (see my evaluation of Improving Conditions below, which is relevantly similar). A third kind of objection contends that debiasing methods may take certain forms or use certain justifications that are not disrespectful at all—I tackle this objection after conducting my evaluation of each method.

Let's now turn to the debiasing methods themselves—how do they fare along the dimension of potentially disrespectful evaluation? I argue that they all involve evaluation of some kind—but at varying depths, and consequently with varying degrees of moral objectionability.

Improve Conditions evaluates agents in that it presupposes that agents placed in certain conditions (e.g. those who see equivalent frames side by side) tend to make better ethical judgments than in other conditions (e.g. those who see the frames separately). But such evaluation is at best only mildly objectionable, for two reasons: first, it is done of agents in general rather than of particular agents—in Improving Conditions for all, we are presupposing that everyone tends to make better ethical judgments in certain conditions, rather than, say, that Jones (in contrast to Smith) is particularly improved by these conditions. This equality of application does not wrong agents comparatively (which would involve disrespectful evaluation of one but not the other). Secondly, the evaluation is of agents' ethical reliability in hypothetical conditions, rather than of their actual judgments. For instance, in presenting equivalent frames side by side, we only presuppose that had they not been presented side by side, the member of the public would have made less reliable ethical judgments—yet we do not render any negative evaluation of their actual judgments, made in actually reliable conditions.

By contrast, Suggesting Unreliability does evaluate the ethical reliability of specific agents' actual judgments. This is because it targets only the subset of agents whose judgments raise suspicions of unreliability—for example, those who judge in favour of identified victims, and those who give inconsistent responses to potentially equivalent frames—and subjects their actual judgments to negative evaluation. This would be disrespectful. It picks out a specific agent as being morally unreliable and questions their judgments, rather than treating them as given.

Surveying Standards occupies an interesting middle ground. Surveying Standards is like Improve Conditions in that it also does not directly evaluate the agents' actual ethical judgments—rather, it is more plausible to say that agents evaluate their own ethical judgments here and find them wanting. In my view, the fact that the agent is the source of the applied standards negates much disrespect. Yet the disrespect isn't cancelled entirely. While the state hasn't performed the evaluation itself, it nonetheless enables such evaluation, by fulfilling certain necessary conditions for debiasing—namely, surveying for specific standards of bias and allowing for the possibility of applying them in the public sphere. In doing so, the state still enables agents to be treated as things, even if it doesn't directly perform the act of evaluation. Furthermore, even if the agent themselves was the source of the standards, this may not yet license using their standards to debias their own judgments. Consider an extension of the professor example: suppose that the student has mentioned before that people in the grips of a certain cognitive bias would be making a mistake in their ethical theorizing, and suppose the professor notices that the student is subject to these very biases in the public sphere. The fact that the student has committed a mistake by their own lights doesn't yet amount to an authorization of the professor's evaluating them in the public sphere. To my mind, then, while the state merely applies an agent's own standard in Surveying Standards, this nonetheless still involves some residual disrespect.

In the dimension of disrespectful evaluation, then, the three debiasing methods can be ranked as follows: Improve Conditions is mostly unproblematic because it does not evaluate particular agents' actual ethical judgments. Surveying Standards similarly does not do so, but nonetheless enables such evaluation, even if only by using the agents' own standards, and hence is slightly more morally problematic. Finally, Suggesting Unreliability is morally worst on this dimension because it targets specific agents' actual judgments in its evaluation.

It might be further objected, however, that we can greatly reduce the disrespectfulness of debiasing methods by adopting different forms or justifications for debiasing. First, we might see the debiasing methods not as saying that specific people are unreliable and that the debiaser is superior to them, but rather as saying that we are all sometimes biased—the debiasing method is justified by “shared fallibility” [55] rather than by the superiority of the debiaser. When this justification is faithfully implemented in practice, ¹⁶ I concede that it negates the comparative wrong of disrespect—of treating a fellow deliberator as having inferior status [51]. But there remains the noncomparative wrong of treating a person as if they fall below some morally significant threshold of capacities for reliable moral judgment—which is what I believe Carter is primarily concerned with. Secondly, we might see the debiasing methods not as criticizing the reliability of specific agents, but as criticizing the reliability of the arguments or reasons put forth by them. I concede that this would negate much of the disrespect, but I worry that it will also prevent the diagnosis of many biases to begin with. Such diagnosis often requires uncovering the true reasons that motivate members of the public, who may not be aware of these reasons. For instance, in Identified Victims, a member of the public who is biased by the vividness of the identified Group A may nonetheless think that their judgment is motivated by a different, morally relevant, feature of that group. Furthermore, the diagnosis of some biases requires essential reference to a specific agent: in Framing Effects, for instance, what's problematic is that the same agent gives inconsistent responses to seemingly equivalent frames, rather than that some pair of inconsistent responses can be gathered from some pair of agents. In sum, tweaking the form and justification of a debiasing method may negate some of its disrespect, and this should be further explored, but I am pessimistic that the disrespect can be negated entirely while maintaining the same degree of reliability.

1.2. The Effects of a Debiasing Method

Next, we might consider the effects of these debiasing methods, quite apart from whether/how they evaluate those to be debiased. Notice first that charges of bias have quite different effects from typical rational argumentation about policy. Typically, we trade reasons pertaining to features of the policy options: we point out, for instance, that one policy leads to less deaths but restricts freedom more, or that one policy benefits the worst off more while the other maximises total happiness, and so on. These reasons pertain to the objects of judgment—the policies themselves. By contrast, charges of bias target the judgments themselves. When we say that someone suffers from identified victims bias or from framing effects, we are not saying anything directly about the policies they judge to be appropriate—rather we are saying something about the reliability or evidential weight of their judgments. Secondly, and relatedly, charges of bias have an undermining rather than rebutting effect on a person's judgments [41]. To explain: a person's judgment in favour of some policy P is akin to placing a weight on the P side of a set of weighing scales. Rebutting this judgment—by pointing out the ill effects of that policy, or by showing that others in the relevant community judge instead that we should pursue some other policy—is akin to placing counterbalancing weights on the other side of the scale. By contrast, charges of bias aim to neutralize the evidential weight of the judgment in favour of P to begin with, which is more akin to activating a laser that destroys the initial weight placed in favour of P. This targeting of a person's judgments, and the subsequent undermining effect, are morally significant if we think that members of the public have a right to participate in the conversation (Floweree, 2023), a right to equal opportunity of political influence [57], to offer up reasons that have the potential to influence. These reasons may not be decisive, but they should have some potential to be. Successful charges of bias, however, undermine and silence a person's judgments, so that these hold no weight at all in deliberation, denying the person their right to participate and opportunity to influence. Evaluate and Exclude quite clearly does silence peoples' judgments, in that it imposes external standards—standards which have not been endorsed by members of the public themselves—which prevent some judgments from being considered at all, while Accept As Is involves no silencing at all, since it simply accepts whatever judgments have been made.

What of the respectful debiasing methods? Improve Conditions is similar to Accept As Is, in that it also does not silence judgments. It may use external standards in determining what counts as an improved condition, but whatever judgments are made in these improved conditions are not undermined.

Suggesting Unreliability also uses external standards to suggest when a member of the public might be unreliable. Yet this does not result in the silencing of their judgments about policy, because in Suggesting Unreliability, members of the public remain free to choose whether to accept the suggested standard of bias, and whether to change their judgments in response to this suggestion. Suggesting Unreliability thus seems morally unobjectionable when it comes to its potential for silencing. It might be objected, however, that Suggesting Unreliability still involves a strong kind of influence on a person's standards for bias, which may be especially worrying if the suggested standards could be reasonably contested, as they have been. Yet, so long as such influence isn't overwhelmingly strong (a matter which may come down to empirical facts about the degree to which the public would uncritically accept a suggested standard of bias), such influence doesn't amount to silencing. I thus conclude that Suggesting Unreliability is mostly morally unproblematic with regards to silencing.

Finally, Surveying Standards does silence a member of the public's judgments about policy, but only if such judgments fail that person's own standards for bias. As with the disrespect of evaluation, the wrong of silencing may be dampened because the person themselves is the source of the standards for bias. Yet, as before, the wrong is not entirely neutralised by this fact, because the member of the public hasn't authorised their standards to be applied in this way. Furthermore, the wrong of silencing is perhaps stronger than the wrong of disrespectful evaluation, given it involves a material effect on the opinions of the agent being debiased. It might be objected that Surveying Standards doesn't entirely silence a member of the public, given that it fully accepts and applies their standards for bias. In response, however, I contend that the primary goal of democratic deliberation is for a member of the public to have their reasons for and against policies to be heard, rather than for their standards of bias to be applied. Thus the fact that Surveying Standards defers to their standards of bias is cold comfort with regards to the silencing dimension. I conclude that Surveying Standards remains moderately objectionable.

Taking the twin dimensions of problematic evaluation and problematic silencing together, Improve Conditions emerges as the morally best option, as it falls short on neither dimension. Suggesting Unreliability and Surveying Standards, by contrast, are roughly tied in that each is most morally problematic on exactly one dimension (disrespectful evaluation and silencing respectively). I now compare these two debiasing methods more closely.

1.3. A Further Comparison of Suggesting Unreliability and Surveying Standards

Suggesting Unreliability and Surveying Standards are similar in that they allow some influence on the standards of bias held by members of the public. Yet they differ in an important respect: Suggesting Unreliability works through the public's own understanding of whether and how to apply a standard of bias, whereas Surveying Standards does not. This gives rise to a few issues: on the one hand, this means that Suggesting Unreliability could engage people's rational capacities in a way that Surveying Standards does not, inviting them to reconsider their judgments and to critically reflect on standards of bias (both their own and the one suggested by the state). For instance, a suggestion of unreliable framing effects may lead a deliberator to try on alternative frames and to evaluate which is the most appropriate description of a case, while a suggestion of identified victims bias may lead a deliberator to try to screen off the effects of identification while keeping in mind morally relevant features that may be revealed by identification.

On the other hand, working through a member of the public's understanding also opens up the potential for them to erroneously apply these standards—whether by overcorrecting (for instance, reversing a judgment in favor of identified beneficiaries even though they would still have picked that beneficiary in the absence of identification) or undercorrecting for bias (for instance, ignoring the caution against identified victims bias when this was in fact driving their judgments). ¹⁷ Suggesting Unreliability may even influence members of the public who don't endorse a standard of bias to take it up anyway, exerting coercive power of a different kind over the public. ¹⁸ An effect similar to silencing may ensue—but unlike with Surveying Standards, it may also silence those who do not endorse the relevant standard of bias.

These differences point to different circumstances under which each debiasing method is appropriate. To the extent that we believe members of the public will be sufficiently reflective and self‐aware to evaluate, adopt, and apply standards of bias appropriately, we may choose Suggesting Unreliability over Surveying Standards. Furthermore, to the extent that we aren't sure that the suggested standard of bias is the right standard to apply (and would instead like members of the public to deliberate about this), we should similarly choose Suggesting Unreliability, only furnishing a suggestion of unreliability for the public to deliberate about, rather than apply it outright.

2. Conclusion

Cases like Identified Victims and Framing Effects show how the ethical judgments feeding into democratic deliberation could be problematically biased—making deliberation go awry and resulting in worse decisions. One option would be to simply identify and exclude unreliable ethical judgments—this promotes reliability, but is also highly antithetical to the moral and political ideals underlying deliberation. I've instead explored other options which aim to strike a compromise between reliability on the one hand, and moral and political respect on the other. Improve Conditions changes the conditions of elicitation to best yield reliable ethical judgments—I've argued that this is the least morally problematic, on account of its neither involving negative evaluations of specific agents' actual ethical judgments, nor the silencing of their judgments in deliberation. Furthermore, the empirical evidence indicates that interventions in the category of Improving Conditions can be effective at mitigating bias, further cementing its ethical justification. The latter two debiasing methods are more morally objectionable: Suggesting Unreliability informs the member of the public that their judgments may be biased, subjecting members of the public to somewhat disrespectful evaluation; while Surveying Standards holds the member of the public to their own expressed standards for bias, but potentially silences their judgments about policy. To the extent that we can trust the public to evaluate, adopt, and apply standards of bias appropriately, and to the extent that we are unsure of the relevant standards of bias, it would be more appropriate to Suggest Unreliability rather than Survey Standards (and apply them). Independently, we might also favour Suggesting Unreliability over Surveying Standards on grounds that there is more empirical evidence for its effectiveness. The possibility and presence of moral bias detracts from democratic deliberation—but it need not compromise deliberation's lofty ideals completely, if we turn to respectful debiasing.

Conflicts of Interest

The author declares no conflicts of interest.

AI Use Declaration

The author used the Elicit tool (which uses artificial intelligence to search the academic literature based on a question prompt) to search for specific academic articles. No other artificial intelligence tools (including generative AI) were used in any other aspect of the writing process, artificial intelligence was not used for drafting, writing, proofreading etc.

Data Availability Statement

The author has nothing to report.