3.a The process of deliberation

To remedy the limitations of mere voting, deliberation must serve as a filter that removes publicly unjustified outcomes from consideration in collective choice. To see how deliberation might provide such a filter, it will benefit us to develop a brief sketch of the deliberative process itself.Footnote ⁹ Let us say that individuals begin by asserting that some rule is publicly unjustified and stating their reason for rejecting it. So, Catherine may assert that rule R is unacceptable, stating some reason A in support of that claim. Once some member of the constituency makes such an assertion, others can demand justification for the source of the reason, which creates a burden of justification for the asserter. To discharge such a burden, Catherine (or some other member sympathetic to her view) must provide an additional assertion which in some way either supports her reason A (and its status as public), or sufficiently answers the interlocutor’s objection. In response, members of the constituency can continue the deliberative process by making justificatory demands for that assertion as well (Heath Reference Heath2008, 119–22). As the process continues, individuals will likely be led to some basic reasons, such as their enduring evaluative commitments or fundamental public reasons. Whatever the appropriate class of enduring reasons might be, if the exchange of demands and reasons reaches these fundamental and enduring reasons, then the process ends, and the resulting justificatory status of the rules for which the assertion is made is determinate for that individual.

The process of giving and asking for reasons is constrained by a norm of consistency. Individuals will be required to maintain consistency among the reasons they offer.Footnote ¹⁰ It is on account of consistency that deliberation will be an effective filter for strategic or dishonest claims. In the process of giving and asking for justification, individuals make assertions of which others keep track. When some citizen makes an assertion that is inconsistent with a previous assertion, an interlocutor can require that the inconsistency be reconciled. Strategic manipulation is filtered out because others can make demands for justification and identify what other options would be acceptable to the would-be manipulator. Individuals can no longer claim without cost that certain rules are unacceptable. They have to provide justification for those rules and ensure that this justification is consistent with previous claims. While assertions that misrepresent one’s reasons are still possible, there is now a reliable means of testing the authenticity of claims which makes citizens’ reasoning accountable to their peers.

Deliberation, then, will be understood in general terms as the process of giving and asking for reasons to justify the collective choice of some proposal. Deliberation in this general sense is distinct from the medium of deliberation. The medium of deliberation refers to the particular way in which participants in deliberation exchange reasons and proposals. Media of deliberation can vary significantly. They include public debate, deliberation among representatives in parliament, or even discussion among citizens on the internet. Invariably, the medium of deliberation is sure to have some influence on how deliberation proceeds (see Stasavage Reference Stasavage2006). To maintain a high degree of generality, I shall discuss deliberation in the general sense as the exchange of reasons without committing to any particular medium of deliberation. For the following argument to hold, it need only be the case that when deliberation is either self-organized or centrally organized, the reasons present in deliberative groups will fail to represent the full diversity of reasons on offer in the population at large. While different media might vary in the degree of homogeneity they are liable to bring about, I take it as given that some degree of homogeneity will still obtain, and this will lead to the homogenizing forces I shall now discuss.Footnote ¹¹

3.b Homogenization of deliberative groups

The above account of deliberation is subject to a significant limitation on account of what I call the problems of polarization and inconclusive justification. If deliberative groups are either self-organized or organized through some centralized process,Footnote ¹² it is almost inevitable that the groups will not accurately represent the diversity of the population. In being relatively homogenous, they are likely to be subject to the polarizing forces of deliberation among relatively homogenous individuals. Hence without some further modification, deliberation will not reliably track publicly justified outcomes on account of this robust propensity for homogenization.

To begin, note that in any modern political system the population is likely to be so large that simultaneous deliberation among all citizens is impossible. Nations, federal subunits, and even most cities are comprised of too many people for a productive exchange of reasons to take place among them all. If deliberation is to be productive, citizens must be organized into various deliberative groups—proper subsets of the population small enough that individuals can gather and engage in an exchange of reasons for or against the justification of proposals under consideration. Given the size of contemporary political communities, we can expect that such subunits will be relatively small compared to the complete constituency.

Within current electoral institutions, deliberative groups might be organized in one of two ways: first, we might allow citizens to self-organize and choose the group with whom they deliberate; or second, we might have some centralized agent (nonrandomly) assign the citizens their deliberative group. On either of these approaches, the deliberative groups produced will, with overwhelming likelihood, be relatively homogenous compared to the general population, and this relative homogeneity will obstruct the use of deliberation as a mechanism for identifying publicly justified outcomes. I begin by defending this relative homogenization claim.

Consider first the self-organizing approach. An array of social scientific literature suggests that when individuals organize themselves, the resulting groups will be relatively homogenous compared to the population at large. First, various biases for association with particular individuals (such as in-group bias) will drive individuals to organize into groups with others who are relatively similar to them (Bernhard, Fehr, and Fischbacher, Reference Bernhard, Fehr and Fischbacher2006a; Bernhard, Fischbacher, and Fehr, Reference Bernhard, Fischbacher and Fehr2006b; Bowles and Gintis Reference Bowles and Gintis2013, 142–46; Hertel and Kerr Reference Hertel and Kerr2001). Second, if the distribution of reasons is neither uniform nor random (as when characteristics such as geography correlate with reasons), then the underlying correlation will result in self-organized groups being relatively homogenous. Lastly, individuals simply do not have access to the information necessary to make informed judgments regarding the reasons held by other potential codeliberators. Even if a citizen is sincerely interested in joining a deliberative group so as to contribute to the diversity of that group, the citizen must have reliable knowledge about the reasons members already present in that group have in order to do so. But, citizens will at best have imperfect and limited knowledge regarding the reasons that members of a given group have, and so are liable to make errors in their assessment of diversity. The degree of this error is likely to be quite significant, and so individuals will unreliably track the diversity of groups. Thus, in all likelihood, people will choose to join deliberative groups which they then make relatively homogenous.

Suppose, then, that well-meaning citizens who seek to self-organize into diverse deliberative groups will be subject to various sources of error, such as biases, contingent correlations in the distributions of reasons, and lack of information regarding the reasons of others. The rate of error need not be particularly high in order to generate radically segregated deliberative groups. This can be viewed as a generalization of Thomas Schelling’s seminal discussion of self-segregation among groups (Reference Schelling1971).Footnote ¹³ Schelling showed that even when individuals have a very moderate tendency to self-segregate despite having preferences for diverse groups, these individuals will produce highly segregated social groups.Footnote ¹⁴ Schelling’s result applies here. The sources of individual error in self-organization introduce a tendency toward homogenization. If this tendency is widespread, it will lead to highly segregated social groups, even if the tendency is rather weak. Self-organization, then, will lead to relatively homogenous groups, especially when compared to the population at large.

Centralized (nonrandom) mechanisms of organizing deliberative groups will not be exempt from the Schelling segregation dynamics. A centralized system, in contrast to the self-organizing approach, has some official entity assign citizens to deliberative groups. A centralized process of organizing deliberative groups may confer various benefits. For example, the central organizer could be insulated from various biases, or she could be in a position to correct for underlying correlations (e.g., they can match individuals from diverse geographic areas). Notice, however, that we may reinterpret the Schelling results to apply to such a process of centralized organization. A central organizer, even if they are exempt from a number of biases and can correct for underlying correlations between reasons and exogenous factors, will almost certainly have extremely limited access to the reasons held by the citizens. Indeed, a central organizer will have less evidence about the reasons that citizens have than the citizens themselves, since citizens at the very least have local knowledge about their own reasons. The organizer’s reliance on unreliable and often radically incomplete information regarding the public reasons that citizens have means that she will likewise be subject to error in the assignment of individuals to deliberative groups. Reinterpret the Schelling model, then, to be such that individuals do not choose what group to join, but rather that they are assigned their place by the central organizer. Minor errors still obtain in the assignment of individuals to groups. Now, the error is made by the organizer rather than by the citizens themselves. The core elements of the Schelling segregation model still obtain—all that has changed is our interpretation of them.Footnote ¹⁵

Thus, regardless of whether individuals assign themselves to deliberative groups voluntarily or if they are assigned to groups by a central organizer, we can anticipate that the resulting deliberative groups will be relatively homogenous compared to the complete population. This homogenization is the result of error in the process of assigning individuals to groups—individuals will not reliably be placed into deliberative groups in such a way that the groups replicate the diversity of the population. Let us now consider how relative homogeneity will frustrate the reliability of public justification on the present deliberative proposal.

3.c Polarization and inconclusive justification

Relatively homogenous deliberative groups face the problems of polarization and inconclusive justification. When members of groups are too similar, deliberation tends only to reinforce extant views, and deliberation becomes a positive feedback loop which entrenches the dominant view rather than being a constructive exchange of reasons. This is the problem of polarization.Footnote ¹⁶ Furthermore, it is unlikely that deliberative groups are relatively homogenous in the same way. In all likelihood, there will be variation in the constitution of different deliberative groups. This variation may lead to groups coming to hold different sets of outcomes as publicly justified, leading to indeterminacy as to what rules do and do not constitute publicly justified outcomes. This is the problem of inconclusive justification.

The polarizing tendency of relatively homogenous groups is, at this point, among the most well-studied social phenomena. If the starting pool of reasons is too similar, the deliberative process will lead individuals to become overly confident in the extant views (Mercier and Landemore Reference Mercier and Landemore2012, 252–53). Returning to our present model: when individuals raise locally dominant reason P, given the relative homogeneity of the deliberative group, there is a low likelihood that any other member of the deliberative group will challenge P. Individuals will thus come to be overconfident in P, which leads them to fail to adequately consider other public reasons and the rules they support.

Empirical observations regularly confirm this polarizing tendency of homogenous groups. Relatively homogenous deliberative groups have a strong tendency to suppress dissent, which reinforces conformity to the dominant view (Sunstein Reference Sunstein2002, 181). Individuals also typically experience a significant conformity bias—when they observe some prevailing (apparent) consensus, they become likely to revise their beliefs in accordance with that consensus (Bicchieri Reference Bicchieri2016, chap. 2; Sunstein Reference Sunstein2002, 176, 179).

These polarizing effects not only make members of that group overconfident, but also increase the likelihood of costly confrontation with out-group members. If the polarization is sufficiently strong, different polarized groups may come to view one another in confrontational terms (Sunstein Reference Sunstein2002, 186). Individuals may become so overconfident in their prior commitments that they become social optimizers (Gaus Reference Gaus2016, 215–20), holding that their most preferred rule, supported by the entrenched reasons that dominate their group, is exclusively acceptable. Clearly, it would take only two groups bent on social optimization for two different outcomes to ensure that public justification becomes impossible. Under such conditions, one group getting their way means that the other must lose—politics becomes zero-sum, and the citizenry will have reasoned themselves into undermining any possibility of political cooperation.Footnote ¹⁷ With the possibility of public justification foregone, citizens will find only instability and distrust.Footnote ¹⁸ Should the public arrive at such an outcome, deliberation will have been eminently self-defeating.

Polarization that leads to the breakdown of political cooperation is but one extreme outcome to which deliberation among homogenous groups may lead. Even if this rather depressing result is not reached, it remains the case that more moderate (but no less problematic) results may obtain which render deliberation ineffective as a mechanism of public justification. First, note that it is extremely unlikely that all deliberative groups will be relatively homogenous in the same way. Whatever public reasons are overrepresented or absent within some deliberative group, other groups will have different reasons that are overrepresented and absent. When some degree of polarization obtains, even if it does not lead citizens to become full-blown social optimizers, the resulting sets of rules which deliberative groups will hold to be publicly justified will, in all likelihood, be different between the groups. The process of public deliberation under relatively homogenous conditions is to some degree path-dependent—the outcomes that the deliberative groups will come to believe are publicly justified will depend on what public reasons are overrepresented and what reasons are ignored, which in turn depend on the initial distribution of public reasons within the deliberative group and the way in which it is homogenous.

The following case serves as a relatively simple illustration of inconclusive justification. Suppose that two different deliberative groups have different relatively homogenous starting sets of public reasons, which lead them to believe that two different sets of outcomes, O and P, are publicly justified (with all rules outside O and P being regarded as publicly unjustified according to the respective groups). As we have just established, O and P are not identical. While some rules may be in the intersection of O and P, there will be a number of rules which are in one set but not the other. These rules have indeterminate justificatory statuses; well-meaning citizens after public deliberation disagree as to whether the rule outside the intersection of O and P is publicly justified or not. By the above assumptions, the set of rules which the deliberative groups jointly regard as publicly justified (the intersection of O and P) is a proper subset of both O and P.

Consider now what happens with the inclusion of further relatively homogenous groups which are constituted differently compared to either of the groups that yielded O or P. This new group will come to hold that the set of rules Q consists of all (and only) publicly justifiable rules, and by the path-dependence of relatively homogenous deliberation, Q will be different from both O and P. The resulting set of rules that could be publicly justified to all members of the constituency in this case is constituted by the intersection of O, P, and Q, which, by the conditions assumed here, will be a proper subset of all three sets, which in all likelihood is also strictly smaller in size than the intersection of just O and P. The introduction of further deliberative groups with different starting conditions thus restricts the set of publicly justifiable rules even further. The result generalizes to the problem of inconclusive justification: when groups undertaking political deliberation are subject to some homogenizing tendencies, the set of rules with determinate justificatory status is inversely related to the number and diversity of the groups. In modern political units, which are both rather large and rather diverse, we should expect the set of conclusively justified rules to be extremely limited.

That some number of rules are inconclusively justified can be problematic for a number of reasons. For one, it means that a number of options which could have been publicly justified are mistakenly removed. At the very least, this creates an opportunity cost of foregone alternatives, some of which may have been desirable. Second, should the case obtain that no rules are in the intersection of the rules that all deliberative groups regard as publicly justified, then possibility of public justification will have been foreclosed—there is no option that every member of the population lacks reason to reject. If this occurs, deliberation as a mechanism of public justification has once again become self-defeating, resulting in conditions under which public justification of rules is unattainable.

Hence, where deliberative groups are either self-organized or centrally (and nonrandomly) organized, there is a high likelihood that deliberation will be self-defeating as a mechanism of public justification. The problem here is that the procedures considered for organizing deliberation are unreliable means of reproducing the diversity of the population—at best, deliberative groups will be highly imperfect, relatively homogenous reflections of the population. This problem also invites a promising solution: if we were to find a means of organizing deliberation such that the diversity of the population is reliably reproduced, then the problems facing deliberation would be resolved.

4.a Lotteries and representation

Recall that deliberation has been proposed as a sort of filter mechanism meant to ensure that the results of voting more reliably track the justificatory status of proposals. The limitations of deliberation, we have found, stem from the fact that deliberative groups will be relatively homogenous. Were we to make deliberative groups more representative of the population at large, then the problems of polarization and inconclusive justification would not obtain, or would at least obtain to a lesser degree.

The selection of deliberators by lottery is a mechanism for generating more representative deliberative groups. If a sufficiently large random sample is taken from the complete population of citizens then, following the law of large numbers, we should expect the sample to be a perfect small-scale replica of the population at large (Stone Reference Stone2009, 390). Such random selection applies the techniques of social sciences for developing representative samples to democratic ends— pure chance, in being unbiased and error free (if it is properly random) will reliably reproduce the diversity of the population. Hence, deliberative groups constructed by random selection are descriptively representative in that they replicate the reasons present in the complete population with a high degree of accuracy (387–89).

Since diversity is adequately preserved within deliberative groups, the problems of polarization and inconclusive justification will no longer obtain. These problems require that deliberative groups are relatively homogenous compared to the population at large. When some degree of homogeneity obtains, positive feedback or pressures for social conformity lead individuals to be overconfident about some reasons and dissmissive of others. By accurately replicating the diversity of the population, most reasons should be present within the deliberative group, which undercuts positive feedback. And, given the diversity of the group, pressure for social conformity should be minimal because there will be no widely shared position to which individuals are pressured to conform. The forces which undercut the exchange of reasons are thus limited in the context of diversity, which means that the deliberative dynamics identified in the preceding section can now reliably filter out nonpublic reasons.

Deliberation among randomly selected individuals has been found to have numerous appealing properties in theory and practice (for a recent review, see Smith and Setälä Reference Smith, Setälä, Bächetiger, Dryzek, Mansbridge and Warren2018). Let us start with its theoretical virtues. First and perhaps foremost, lotteries are egalitarian. Each citizen has an equal opportunity to influence the outcome of collective decision-making by virtue of having an equal chance at being chosen to act as a representative (Lopez-Guerra Reference Lopez-Guerra2011; Saunders Reference Saunders2010).Footnote ¹⁹ Second, by virtue of accurately representing the public, assemblies constructed by lottery take into consideration all interests of citizens, which satisfies the classical democratic condition of equal consideration of interests.Footnote ²⁰ Third, certain formal results suggest that the problem-solving abilities of randomly constructed assemblies as a whole will be at least as good as those of elected assemblies, if not better (Landemore Reference Landemore2013). And fourth, randomly constructed assemblies are less susceptible to capture and manipulation by outside parties than alternative arrangement of social decision-making (Guerrero Reference Guerrero2014).

In addition to these appealing properties which have been identified in theory, the available empirical evidence on the performance of randomly assembled deliberative groups has often (though not exclusively) been favorable. It is important to note that much of the evidence I discuss now has been drawn from experiments using mini-publics. Although they are similar in spirit to randomly constructed assemblies, mini-publics are of insufficient size to replicate the complete diversity of the population.Footnote ²¹ Considerations regarding how to organize relatively larger assemblies so that deliberation might take place are addressed in the following subsection. With that said, empirical observation of the deliberation and choice of mini-publics, while rather recent as far as results in the social sciences go, provide pro tanto grounds in favor of their use. First, deliberation within mini-publics has been observed to be efficacious, leading to changes in beliefs among participants (Himmelroos and Christensen Reference Himmelroos and Christensen2013; Fishkin Reference Fishkin2011; Niemeyer Reference Niemeyer2011). Deliberation among randomly selected individuals thus has the desired effect of having them reflect upon and revise their beliefs in light of new consideration. Such belief revision following deliberation was also observed in the case of the British Columbian Citizens’ Assembly, which is perhaps the most well documented case of political decision-making done by a randomly constructed assembly (Blais, Carty, and Fournier Reference Blais, Carty, Fournier, Warren and Pearse2008). Remarkably, deliberation in the British Columbian case also led to a consensus amongst a supermajority. Second, when individuals uninvolved in deliberation of a mini-public are told of the decision reached by the mini-public, they tend to revise their opinions in favor of the result (Boulianne Reference Boulianne2018; Ingham and Levin Reference Ingham and Levin2018). The general public in democratic societies thus seems receptive to the judgments of randomly constructed deliberative bodies. Lastly, a number of appealing features have been observed from a range of empirical cases of mini-publics: mini-publics are capable of creating policy, legitimizing policy choices, informing public debate, and building novel constituencies for cooperation in the political sphere (Gooden and Dryzek Reference Gooden and Dryzek2006).

This review of theoretical and empirical considerations is meant to speak in favor of using lotteries as mechanisms of organizing deliberative groups prior to social decision-making. The principle motivation for introducing lotteries in the present context, however, is that they are conducive to social decision-making which tracks the justificatory status of proposals better than alternatives. And while such favorable qualities may obtain, it remains to be seen whether there is a tractable way of organizing institutions in such a way that these favorable qualities might be realized. More conventional conceptions of voting and deliberation also possess appealing properties, but, as we have seen, they are unreliable means of bringing about publicly justified outcomes. In order to show that randomly constructed assemblies are candidate means of achieving publicly justified outcomes, I shall now turn and consider some details of the institutional design of such assemblies.

4.b Lotteries and institutional designFootnote ²²

I shall consider three features pertinent to the design of randomly constructed assemblies: the scope of decision-making of such assemblies, the mechanism and organization of decision-making within the assembly, and the construction of assemblies. To begin, let us consider the role and scope of randomly constructed assemblies. Recall that the principle concern in this article has been to identify a democratic institutional arrangement which might reliably lead to publicly justified outcomes. The introduction of randomly constructed assemblies is meant to improve the likelihood that the social choice process leads to such outcomes. Given our present focus on publicly justified outcomes, then, randomly constructed assemblies should serve that role which best improves the reliability of reaching publicly justified outcomes.

Public justification applies to certain features of our social and political world with the aim of introducing desirable publicly justified features and removing or revising existing unjustified features. Within constitutional democracies, this role naturally falls upon the legislative branch of government. So, lotteries will need to be incorporated into that branch in some capacity. The pertinent question is to what degree we should incorporate lotteries within the legislative branch.

We might first consider the radical proposal that we replace elected legislative chambers with randomly constructed assemblies altogether (Hennig Reference Hennig2017; Van Reybrouck Reference Van Reybrouck2016). Since elections are less likely to produce publicly justified outcomes than lotteries, replacing them will improve the reliability of achieving publicly justification. Considered in the context of pure theory, the radical proposal is motivated well enough. When it comes to questions of reform, however, such radical change may bring along with it costs which make moderate suggestions more appealing. In the context of reform, other values are sure to be salient, such as the stability and perceived legitimacy of institutions. This leads to an empirical question: To what degree, and in what direction will replacing elections affect the perceived legitimacy and stability of political institutions? While the empirical evidence reviewed above speaks favorably to the question of legitimacy, it nevertheless relies on evidence generated by controlled laboratory experiments and not reform of fundamental political institutions, which may generate side effects that are not anticipated in the lab.Footnote ²³

More moderate introductions of lotteries into the social decision-making process may still contribute to improving the reliability of achieving publicly justified outcomes (Gastil and Wright Reference Gastil and Wright2019). First, we might create a separate randomly constructed assembly which is given veto power over proposed legislation. Giving the assembly the ability to review legislation and veto that which is deemed publicly unjustified will improve the reliability of public justification by significantly reducing the likelihood of unjustified legislation being passed. Another moderate proposal is to allow the randomly constructed assembly power only to recommend and approve legislation. Such recommendations might serve only as a sort of formal evaluation of legislation without further effect. The relative weakness of this proposal is likely to improve reliability of public justification less than a veto, but insofar as approval of the randomly constructed assembly is regarded as a beneficial prospect for legislation (and its absence a detriment), then the resulting incentives for political actors should still improve the reliability of public justification.Footnote ²⁴ An even more modest means of incorporating lotteries is to have mini-publics structure public deliberation before elections. In this case, deliberation and decision-making by randomly constructed groups has neither formal role nor power, but instead hopes to improve the prospect of public justification by making salient for voters the justificatory status of proposals under consideration.

Which proposal to take is not a question that I can answer here. Too much depends on empirical questions of feasibility, perceived legitimacy, and stability of reformed institutions. Answering this question requires balancing the value of public justification against these other practical values. However, by sketching a range of ways of introducing lotteries into social decision-making at different levels of modesty, I hope to have shown that there is some way of introducing lotteries into extant social decision-making procedures. And, insofar as we incorporate lotteries into decision-making to some degree, then we should improve the reliability of reaching publicly justified outcomes relative to alternatives where lotteries are not used at all.

I now turn to examining decision-making within randomly constructed assemblies. Recall that deliberation is initially introduced as a means of supplementing voting. Since the mere casting of ballots is not a reliable mechanism for tracking the justificatory status of proposals, deliberation is introduced as a kind of filter mechanism. On the lottery proposal under consideration, our aim is to use random selection in order to generate a descriptively representative assembly such that every reason present in the population is also on offer in the course of deliberation. A pertinent question, then, is whether deliberation can be organized within the assembly such that it serves as an effective filter mechanism within the assembly such that a vote after the fact can reliably identify publicly justified outcomes.

In order to make a genuinely representative assembly, the sample taken by lottery must be relatively large, ranging from hundreds to perhaps over a thousand representatives.Footnote ²⁵ Whatever the exact number of representatives, it is clear that the minimal number needed is so large as to make effective simultaneous deliberation among all representatives prohibitive. This raises a conflict between accurate representation and tractability of deliberation. Thus far, experimentation has favored quality of deliberation above accuracy of representation, but such favoring comes with the obvious cost of making lotteries less reliable at avoiding relative homogeneity, which makes the problems of polarization and inconclusive justification pertinent once again.

Is there any way of reconciling quality of deliberation with accuracy of representation? Lotteries were meant to accurately recreate diversity such that individuals were reliably exposed to reasons they themselves would not have considered. This reliable exposure to diversity is the mechanism by which the problems of polarization and inconclusive justification are stymied. If we can organize the assembly in such a way that individuals may deliberate in smaller groups while also reliably being exposed to other views, then we might reconcile representation and quality of deliberation.

One way to accomplish this is to organize smaller deliberative groups within the assembly for tractable deliberation, but then have individuals regularly meet in different deliberative groups. If these smaller deliberative groups were reorganized such that at each iteration individuals deliberate with different peers each time, after enough iterations each individual will have deliberated with every other individual in the assembly, and so will have been exposed to the full diverse array of reasons pertinent to the proposal under consideration. Such a suggestion does, however, come at the expense of taking a significant amount of time to accomplish. One must thus balance time of decision-making against range of exposure to diverse reasons. Compromising in favor of time, however, need not come at too great an expense of representing diversity. As representatives deliberate with other diverse persons, they can come to hold the reasons of their peers themselves. When they deliberate with others, they may then spread those reasons even further. Acquisition of a reason, in the sense that one becomes aware of the reason and the proposals it supports or opposes, need not require exposure to the original reason holder, and so for each representative to be exposed to all pertinent reasons, it need not be the case that they actually deliberate with all other representatives. Since the aim is to expose representatives to all pertinent reasons, such indirect exposure is adequate.

Iterated deliberation within smaller groups may thus expose individuals to a sufficient degree of novel and diverse reasons such that the problems of polarization and inconclusive justification are avoided, all while ensuring that deliberation of sufficient quality can take place. Depending on how quickly reasons can diffuse within the population, the number of iterations can be rather low. Note also that the success of this argument ultimately depends not on the deliberation process among randomly selected representatives guaranteeing full exposure to all reasons, but only on the reliability with which it leads to an exposure of more reasons than representatives would receive if the assembly were constructed by other means. Given that lotteries will produce deliberative groups which are significantly more representative than groups constructed by other means, this will almost surely be the case. Hence, even if randomly constructed assemblies still face the problems of polarization and inconclusive justification, they face these problems to a lesser degree than assemblies constructed by other means, and so are to be favored in that regard.

The last feature of institutional design which I shall consider here are some challenges regarding effective representation within randomly constructed assemblies. Two worries arise. First, many of the above examples of citizens’ assemblies and mini-publics incorporate oversampling of minorities, which can lead to the assembly being imperfectly representative. Second, the accuracy of the representative sample depends on selecting individuals from the complete population, and so if individuals are permitted to opt out of selection then the resulting assembly may not be representative. Beginning with oversampling, representing minorities at a relatively higher degree does not itself appear problematic. Incorporating fewer members of the majority (or plurality) into the assembly should pose a minimal cost, since their reasons are still almost certainly going to be represented within the assembly. That said, one might have concerns for the mechanism of assigning overrepresentation since it might make the system liable to manipulation or abuse. The discretion accorded in the selection of which groups to oversample may both be liable to manipulation, but also may turn on controversial judgments of what determines group membership and group interests. In this case, we might have better reason to favor pure random selection of a sufficiently large sample without any interference—even in the worst case, every reason will be represented within the assembly, and so regardless of what we take to be the relevant groups we are able to represent all of them, and we do so in a way insulated from manipulation. Hence, while oversampling might be common practice in current experiments of mini-publics, it is not necessary in order to generate an accurately representative assembly, and insofar as the assembly is so representative we will have attained the desired conditions of public justification.

The second worry concerns how to handle individuals who opt out of the lottery procedure. Presumably, we have a desire to maintain individual independence in the selection of what occupation they take. Freedom of occupation is generally taken as an uncontroversial individual right, and such a right entails the permissibility of individuals choosing to abstain from the lottery for representatives. But it is almost certain that opting out behavior will not be uniform, in which case sampling will not be from the complete population, but rather a skewed subset, and so the sample generated is less likely to be representative. This presents a challenge for the use of lotteries: if the sample is not representative, then randomly constructed assemblies may be no better at dealing with the problems of polarization and inconclusive justification than electoral alternatives. The use of lotteries thus seems to be at odds with individual liberty.

Iterated sampling, I believe, can reconcile the conflict between accuracy of representation and individual liberty. To begin, we allow individuals the opportunity to opt out of the selection. However, we do not thereby remove their name from selection, rather we only mark the lot as one which has opted out. Every name is thus left in for selection, ensuring we sample from the entire population. Then, we select names at random until a sample of sufficient size is generated. With the sample in hand, we check whether any member has opted out. If no individual opted out, then we have a representative assembly where every participant enters voluntarily. If any individual did opt out, then we replace all the names that have been drawn and conduct another drawing. We may thus repeat the process drawing from the complete population until we generate a representative sample where no individual opted out.Footnote ²⁶ Since each drawing is from the complete population, we face no problems arising from drawing from the incomplete population while nevertheless maintaining a commitment to individual liberty.

This concludes the discussion of the more technical institutional details of incorporating lotteries into democratic social decision-making. This discussion is hardly exhaustive, but I hope to have addressed the most pressing questions regarding the details of how lotteries would function within democratic institutions. Moreover, as has been noted at several occasions, for the argument for the use of lotteries to be successful, it need not be the case that the construction and decision-making of randomly constructed assemblies proceeds perfectly. Rather, the success of the argument turns on the comparative claim that the organization of deliberative groups by lottery improves the prospects of public justification relative to organization of these groups by other means. Hence, even if lotteries still tend toward some degree of relative homogeneity, and so still face the problems of polarization and inconclusive justification to some degree, this need not count against them insofar as they generate relative homogeneity to a lesser degree than alternatives. The above institutional sketches hope to show that there are tractable means of achieving relatively more accurate representation with the use of lotteries, thus making them preferred to more conventional alternatives in the context of public justification.