1 Introduction

The epistemic value of a political procedure—such as democracy or a civil trial system—depends on how well it performs in arriving at decisions that are correct by some independent standard.Footnote 1 A core assumption in the literature on epistemic democracy is that boosting the epistemic value of such a procedure makes it better overall. This assumption often operates in the background. But the fact that many theorists take it for granted can be discerned given how common it has become to argue for a particular governance architecture based on its anticipated epistemic value (Blum & Zuber, 2016; Brennan, 2016; Guerrero, 2014; Landemore, 2013). Equally telling is the emergence of a whole literature that is mainly concerned with the question of how epistemic value can be boosted without giving much consideration to the prior question of whether doing so is generally desirable.Footnote 2 Whereas the assumption remains tacit in most contemporary treatments, it is salient in David Estlund’s influential work on epistemic democracy. He writes: “Democratic legitimacy requires that the procedure can be held, in terms acceptable to all qualified points of view, to be epistemically the best (or close to it) among those that are better than random” (Estlund, 2008, p. 98). Estlund’s claim that democratic legitimacy requires choosing on the basis of epistemic value seems to entail that boosting it makes an eligible process more choice-worthy overall.

The goal of this paper is to put forward and defend a modest claim. We will argue that it is not obvious that boosting the epistemic value of an eligible political procedure makes it better overall. We will specifically show that increases in epistemic value can give rise to egalitarian objections. An analogy with wealth is helpful to convey the general trajectory of our argument. It is reasonable to be concerned not only with increasing wealth in a society but also with the distribution of that wealth. This is because it is possible to increase total wealth in absolute terms to the benefit of some and to the detriment of others. Accordingly, sometimes egalitarians prefer less overall wealth, if this is the only way to generate a more even distribution of it. The idea here is that the maximization of wealth, even if a value in itself, cannot be simply undertaken without regard for fairness considerations that ensue.

Epistemic democrats fail to note that increases in epistemic value, just like increases in wealth, can produce winners and losers. This gives rise to questions about the distribution of the benefits that result from increases in epistemic value. In particular, fixing only for epistemic errors that are to the detriment of a particular group could increase the procedure’s epistemic value in absolute terms, but it would also bias it overall in that group’s favor. Thus, it is not obvious that the intervention, though increasing the procedure’s epistemic value, renders it more choice-worthy overall. Importantly, egalitarians may prefer the procedure prior to the intervention because it distributes the benefits (and detriments) arising from systemic error more evenly. If this is right, then reasonable people will disagree, at least in some cases, about whether increasing the epistemic value of a political decision-making procedure would be desirable. Thus, we should not merely concern ourselves with boosting epistemic value, but also think about whether doing so treats the involved parties fairly.

We shall focus on Estlund’s epistemic proceduralism because he is one of the few theorists, if not the only one, that explicitly discusses the moral complexities involved in boosting the epistemic value of legal and political procedures. For instance, Estlund’s argument for court authority (2008, pp. 156–8) entails the claim that within the set of demographically neutral juries, authority goes to the one that maximizes epistemic value.Footnote 3 Elsewhere, Estlund permits some departure from the equality of political input in order to boost the epistemic value of the political process, writing: “unequal opportunity for input should be allowed, to some extent, if by doing so the overall amount of input is increased and as a result the expected epistemic value of the overall arrangement is improved” (Estlund, 2000, p. 139). Like other epistemic democrats, Estlund overlooks the fact that the distribution of epistemic benefits matters. As we shall argue, this omission leads him to overstate the desirability of boosting epistemic value in such contexts.

Inspired by Estlund’s analogy between judicial procedures and the political process, in Sect. 2, we start by discussing how the epistemic value of a court could increase in ways that are systematically to the detriment of one of the parties. To this end, we consider a scenario in which one party gains an advantage by preventing the court from erring to their detriment, i.e., precisely through boosting the overall epistemic value of the system. We show that although in such scenarios the demands of substantive justice are better met at the level of individual decisions, this comes at the price of biasing the process to the other party’s detriment at the systemic level. The prevention of error makes the court better in an epistemic respect, but the ensuing unevenness in the distribution of errors makes it worse in an egalitarian respect.

In Sect. 3, we abstract away from court decisions and extend our argument to the political process. We will argue that it is not obvious either that increasing the epistemic value of the political process makes it better overall. For it is possible to boost the epistemic value of the political process yet make it worse on egalitarian terms. Our aim is not to resolve such tradeoffs but argue that epistemic democrats should take them into account.

2 Increasing epistemic value by decreasing fairness in trials

Estlund understands the epistemic value of a procedure in terms of its accuracy, that is, its probability “to produce decisions that are correct by the appropriate independent moral standards.”Footnote 4 Accordingly, a political procedure would have greater accuracy if it yields a greater number of correct decisions than an alternative.Footnote 5 Now suppose that a civil trial system’s accuracy rate for cases in which one of the parties is richer than the other is eighty percent, that is, in eighty percent of these cases, the system reaches the substantively just outcome. A fair assumption is that rich and poor parties each deserve to win fifty percent of all cases. Thus, absent any disparity, the rich and the poor have each a fifty percent chance of winning every case, though of the fifty percent of the cases they win, they each only deserve to win forty percent. The other ten percent of the cases that they win, they win due to the system’s errors that happen to be to the detriment of their opponents. Thus, absent any disparity, trial statistics for both the rich and the poor for 100 cases will be as follows:

  • Total number of cases deserved to win: 50.

  • Total number of cases won: 50

    • Cases won in error: 10

    • Cases justly won: 40

  • Total number of cases lost: 50

    • Cases lost in error: 10

    • Cases justly lost: 40

Now suppose that the richer half of the parties decided to hire a prestigious law firm that can prevent the ten cases that they normally expect to lose due to the system’s error. If all else is equal, this will raise the trial system’s accuracy from eighty to ninety percent. In doing so, however, this change will also make the rich a whooping fifty percent more likely to win cases against poor adversaries. Consider the trial statistics for each party after this change in turn. For the rich, the new statistics will be as follows:

  • Total number of cases deserved to win: 50

  • Total number of cases won: 60

    • Cases won in error: 10

    • Cases justly won: 50

  • Total number of cases lost: 40

    • Cases lost in error: 0

    • Cases justly lost: 40

Compare these statistics now to the new trial statistics of the poor:

  • Total number of cases deserved to win: 50

  • Total number of cases won: 40

    • Cases won in error: 0

    • Cases justly won: 40

  • Total number of cases lost: 60

    • Cases lost in error: 10

    • Cases justly lost: 50

The ten additional cases that the rich win serve substantive justice at the level of individual decisions.Footnote 6 Yet, with this increase in the number of individually just decisions, the poor lose while the rich benefit. It is not obvious therefore that this change makes the system better overall. On the contrary, the system might now be open to an egalitarian objection.

One might worry whether a reasonable egalitarian objection can be articulated here, given that the benefits that the poor previously enjoyed as a result of epistemic errors were, well, in error. It was wrong for them to have enjoyed those benefits. They did not deserve them. Substantive justice requires that those wrongful benefits should be taken away from them. The worry would be then that any egalitarian objection on the basis of the poor’s losses would amount to the claim that the poor will have a right in distributive justice to benefit from epistemic error. But we have a different egalitarian objection in mind.

The poor have a claim to be treated just like the rich in substantive terms. By correcting the errors to the detriment of the rich, the new system protects their rights to win all the 50 cases they deserve to win. The poor on the other hand is left to win only 40 cases of the 50 they deserve to win. The egalitarian objection will be that in such a system, not everyone enjoys the same fair value of rights. On that basis, some egalitarians may prefer the system prior to the intervention, which has a smaller number of correct results but in which parties have the same chances of winning the cases they should win.

The worry may persists on the basis that the egalitarian’s preference for the system prior to the intervention endorses leveling down, given both parties start at wining 40 of the 50 cases they deserve to win. One may object that endorsing leveling down in such a case is beyond the pale of reason and maintain that the egalitarian has failed to put forward a valid objection to increasing epistemic value.

We can remain agnostic about whether endorsing leveling down in this context is reasonable or not.Footnote 7 For even if leveling down proves to be never reasonable, the leveling down objection only applies to a small subset of interventions to boost accuracy, namely those that result in Pareto improvements. Consider any number of possible interventions after which the rich in addition wins some more of the cases that the poor deserves to win. In such scenarios, the leveling down objection does not apply, because the poor will be made worse off compared to the initial setting. Thus, suppose that the prestigious law firm helps the rich to win the ten cases that they were losing due to error as well as unjustly wining one more of the cases that the poor should have won. Compared to the initial scenario, this change increases the overall accuracy from 80 to 89% with the trial statistics for the rich as follows:

  • Total number of cases deserved to win: 50

  • Total number of cases won: 61

    • Cases won in error: 11

    • Cases justly won: 50

  • Total number of cases lost: 39

    • Cases lost in error: 0

    • Cases justly lost: 39

Compare these statistics now to the new trial statistics of the poor:

  • Total number of cases deserved to win: 50

  • Total number of cases won: 39

    • Cases won in error: 0

    • Cases justly won: 39

  • Total number of cases lost: 61

    • Cases lost in error: 11

    • Cases justly lost: 50

In this revised scenario, the poor is worse off compared to the initial setting, given they start with winning 40 of the 50 cases they deserve to win but end up winning only 39 of them. Now, even a prioritarian will object to these sorts of interventions on the basis that they fail to protect everyone’s fair value of rights though they increase the system’s aggregate accuracy. The poor can object that their chances of winning the cases they deserve to win decreases as a result of an aggregate increase in the system’s accuracy.

Once again, the idea isn’t that the poor has a right to benefit from epistemic errors. The intervention merely increases aggregate accuracy by correcting the errors to the detriment of the rich while permitting further errors to the detriment of the poor. Rich and poor both have a right (in substantive justice) to win the cases they deserve. But the new system increases the value of the rich’s right (from 40 to 50) while decreasing the value of the poor’s (from 40 to 39). Thus, in the new system, though the parties continue to enjoy the same rights, they no longer enjoy an equal value of said rights. This is the basis of the egalitarian objection we have in mind, which in the revised sorts of cases, is not open to leveling down objections.

Finally, one might worry that our example features deliberate engineering to benefit the rich (by the rich themselves). The idea here would be that in the absence of that kind of deliberate intervention, the egalitarian will have no reasonable grounds for complaint. But once again, this mischaracterizes the egalitarian objection we have in mind. The egalitarian’s concern isn’t about deliberate wrongdoing. The egalitarian is rather concerned about which criteria should be deployed for choosing competing institutions. More specifically it is about whether aggregate accuracy is an uncontroversial criterion.

As discussed above, Estlund is committed to the view that within the set of demographically neutral juries, authority is to be granted to the one that maximizes epistemic value, that is, the system that produces more just results in the aggregate. When comparing the system before and after the intervention, Estlund does not consider the potential egalitarian objection we articulated, thus permitting the rich to increase the system’s accuracy and thereby the value of their rights. Our analysis shows that such an increase of epistemic value will not secure uptake from egalitarians. Our claim isn’t that the system would be better overall without such increase but that a tradeoff must be made between two competing values.

Some might suggest that when evaluating any two trial systems, they should first be evaluated according to how reliably they track the truth, and if—and only if—they are equally good by that criterion, then considerations of fairness ought to break the tie. Call this the “epistocratic lexicographic principle.” This lexicographic principle implies that the intervention makes our thought experiment’s civil trial system better overall, the poor’s detriment notwithstanding. A reasonable competing view may be that, when evaluating any two trial systems, they should first be evaluated according to how well they fare with respect to protecting the fair value of rights, and iff they are equally good by that criterion, then considerations of epistemic superiority ought to break the tie. Call this the “egalitarian lexicographic principle.” This lexicographic principle implies that the intervention renders our civil trial system worse overall, the system’s higher aggregate accuracy notwithstanding.

The competing lexicographic principles prioritize different values. Yet, it is not obvious which principle is right. Estlund’s epistemic argument for court authority seems to presuppose the rightness of something like the epistocratic lexicographic principle, without considering the drawbacks we’ve discussed.

Moreover, even if the epistocratic lexicographic principle turns out to be right, our analysis supports a weaker though still interesting conclusion, namely that there is a moral reason to attend to the distribution of accuracy, and not only its volume. That is to say that even among systems of equal epistemic value, we must pick the one with the most even distribution of error.Footnote 8 In short, our analysis at the very least reveals that the distribution of error is an important dimension of epistemic value, which the literature has overlooked.

For Estlund (2008, pp. 11–15), the importance of serving substantive justice in trials reveals something about the importance of serving substantive justice in the political process. Equally, the egalitarian objection we specified in the context of trials shows potential egalitarian concerns about boosting epistemic value in the political process, to which we shall now turn.

3 Increasing epistemic value by decreasing fairness in politics

As discussed in the introduction, Estlund (2008, p. 98) maintains that given some qualifications, democratic legitimacy requires that the political decision-making procedure can be held to be epistemically the best (or close to it). This seems to entail that increases in aggregate accuracy of a political system do not alter it in any other morally significant way that could render it worse overall. This controversial assumption also permeates Estlund’s discussion of political input (2000, p. 127). As we saw in the introduction, he permits some deviation from equal input if this boosts the epistemic value of the political process. Both positions can be challenged by extending our earlier analysis to the political context, showing that increases in accuracy may elicit an egalitarian objection.

Consider a democratic vote on how to legislate on 10 issues of roughly equal moral importance. Suppose 6 laws are just and the other 4 are unjust. Of the unjust laws the first pair unjustly benefits group A to the detriment of group B and the other pair unjustly benefits B to the detriment of A. In a system that mandates equal input, all 10 measures are adopted. Here, A and B are equally treated under the system’s shortcomings.

Let’s start with Estlund’s second position, i.e., the one on political input first. According to Estlund, unequal opportunity for input is justified if it (a) increases the epistemic value of the arrangement and (b) the increase of input is regulated by the “epistemic difference principle” (Estlund, 2000, p. 147). The epistemic difference principle allows for unequal input if everyone’s possibility for input is increased.

Allowing unequal input according to the “epistemic difference principle” in our thought experiment can result in outcomes that can increase the epistemic value of the political process but admit egalitarian objections. Suppose that both A and B spend more input but that A significantly outspends B to secure an outcome that is as before except that the two laws that unjustly benefit B to the detriment of A are repealed or simply recast in a fully just fashion (scenario 1). Alternatively, suppose that by outspending B, A secures an outcome that is as in scenario 1, except that in addition, one of the just laws is turned into an unjust one benefitting A to the detriment of B (scenario 2).

Both scenarios meet all of Estlund’s criteria. In the original case the procedure generated 6 just and 4 unjust results. In scenario 1, it generates 8 just and 2 unjust results, whereas in scenario 2, it generates 7 just and 3 unjust results. Imagine further that this is not an incidental outcome but reflective of a long-term trend of legislation under the new input rule. According to Estlund’s argument, we are morally permitted to move to an unequal input regiment, even if its outcomes are systematically along the lines of either scenario 1 or scenario 2. Our prior analysis demonstrates that either of the latter sorts of moves will be open to egalitarian objections, given that they are to B’s detriment (the objection to scenario 1 entails endorsing leveling down while the objection to scenario 2 doesn’t). In both scenarios, B suffers more from injustice than A. Therefore, it is not obvious that these scenarios will be morally justified. This challenges Estlund’s “epistemic difference principle.” The justification of deviations from egalitarian opportunities of input are not only sensitive to the fair distribution of input opportunities and the accuracy of outputs but also to the fair value of the parties’ rights to not suffer from injustice.

The same logic can now be applied to Estlund’s first position, which holds that increases in aggregate accuracy of a political system couldn’t make it morally worse. As should be clear from our discussion thus far, it is contentious whether there are overriding reasons to move from the original case of our thought experiment to either scenario 1 or 2, despite their greater aggregate accuracy. The move entails a tradeoff, and reasonable people may disagree about how it ought to be settled. And at least some reasonable agents do not only care about substantively just decisions alone, but also about being treated equally.

Let us conclude by considering the nature of the moral disagreement that our discussion has unearthed. Hélène Landemore (2013, p. 191) distinguishes between two kinds of political problems. On the one hand, there are political problems for which “some common ground can be found between citizens and one can expect that they will go beyond their own self-interest and toward the good of the community as a whole.” On the other hand, there are political problems for which we cannot draw on such shared resources. These are problems that are best characterized by irreducible conflicts of interest. Landemore argues that epistemic arguments are only applicable to the former but not to the latter kind of problems. What is interesting about the type of moral disagreement that we have unearthed is that it applies to situations in which we share a common ground. In the cases at hand, there is common ground about the independent criterion that the political procedure should track in its individual decisions. What the parties disagree about is what value the political procedure should prioritize at a systemic level, accuracy in tracking said independent criterion or protecting the fair value of rights.