2. SET UP: EPISTEMIC DEMOCRACY AND THE CHALLENGE OF EPISTOCRACY

Landemore's book is a major contribution to the literature on “epistemic democracy.”Footnote ² Epistemic democrats contend that, contrary to common skepticism about the epistemic merits of democracy, epistemic considerations actually weigh in favor of democracy over at least some non-democratic rivals. There are many ways of supporting this contention, but the most typical approach proceeds by positing normative standards for the correctness of political decisions that are independent of the procedures by which such decisions are made, and then arguing that, relative to such standards, democratic procedures tend to make correct decisions more reliably than non-democratic alternatives.Footnote ³

Epistemic democrats face a challenge from what David Estlund calls epistocracy (Reference Estlund2008a: 30). Whereas democracy is rule by the many, epistocracy is a form of rule by the few, specifically by the few who know best (or at least better than the rest). In an epistocracy, ultimate political power rests with a group whose membership is defined by satisfaction of certain epistemic criteria. Many different systems could count as epistocratic. For example, rule by a small council of highly skilled and knowledgeable experts is epistocratic, but so is rule by all and only those who have attained a high-school diploma.

Epistocracy poses a challenge to epistemic arguments for democracy because it seems, to many people at least, that some form of epistocracy is likely to make correct decisions more reliably than democracy. “It is certain,” says Estlund, “that there are subsets of citizens who are wiser than the group as a whole” and who “know better than the rest of us what should be done” (Reference Estlund2008a: 40). Why shouldn't these wiser and more knowledgeable citizens rule over the rest of us? This is a question that any defense of democracy must answer, but it poses an especially acute challenge for epistemic democrats who must explain why their own emphasis on epistemic considerations favors democracy rather than epistocracy.

In his own influential account of epistemic democracy, Estlund concedes that the challenge of epistocracy cannot be met by any appeal to epistemic considerations. To answer the challenge, says Estlund, epistemic democrats must “appeal to some principle” that rules out epistocracy on non-epistemic grounds (Reference Estlund2008a: 40). Estlund invokes just such a principle – the qualified acceptability principle – which requires that, “any legitimate exercise of political power … be justifiable in terms acceptable to all qualified points of view (where “qualified” will be filled in by “reasonable” or some such thing)” (Reference Estlund2008a: 41). This principle rules out epistocracy because “any particular person or group who might be put forward as such an expert would be subject to controversy, and qualified controversy in particular” (Reference Estlund2008a: 36).

By invoking a non-epistemic principle to answer the challenge of epistocracy, Estlund significantly shrinks the epistemic aspect of his account. For it turns out that democracy is preferable to most of the alternatives people actually worry about – epistocracy, benevolent dictatorship, oligarchy, and so on – not because democracy is epistemically better, but because those alternatives fail to meet the qualified acceptability requirement. Estlund's epistemic defense of democracy thus turns out to be mostly non-epistemic. As Elizabeth Anderson notes in her review, “the qualified acceptability requirement … leaves democracy as the only authoritative form of public decision making left standing, provided there is some reason to believe that it has some modest tendency to produce just decisions” (Reference Anderson2008: 132).

3. LANDEMORE'S DEFENSE OF EPISTEMIC DEMOCRACY

Landemore contends that Estlund's epistemic minimalism “concedes too much, and is, furthermore, unwarranted” because the claim that epistocracy is likely to epistemically outperform democracy is false (p. 52). On her view, “epistocracy is not a tempting option, not because we can never agree on who the knowers are … but because there are reasons to think that the more reliable knower is actually the group as a whole, as opposed to any particular individual or group of individuals within it” (p. 52). Landemore argues that democracy is likely to epistemically outperform any non-democratic rival, including epistocracy. As she puts it, “the rule of the many is epistemically superior to any version of the rule of the few” (p. 166). In this section I set out Landemore's argument for this claim and briefly summarize the findings upon which it depends.

Landemore's argument rests heavily on work on collective intelligence by Lu Hong and Scott Page. Using computer models consisting of simulated agents, Hong and Page compared the performance of two types of groups on collective problem-solving tasks: groups drawn at random from the population as a whole, and groups composed of the best individual problem solvers. Hong and Page found that the former type of group reliably outperformed the latter: “a random collection of agents drawn from a large set of limited-ability agents typically outperforms a collection of the very best agents from that same set” (Hong and Page Reference Hong and Page2004: 16386).

How could a randomly selected group consistently beat a group composed of the best individual problem solvers? Hong and Page answer that the latter group, “while its members have more ability, is less diverse” (Hong and Page Reference Hong and Page2004: 16385). The sort of diversity Hong and Page refer to is what they call functional or cognitive diversity: “differences in how people encode problems and attempt to solve them” (Hong and Page Reference Hong and Page2004: 16385).Footnote ⁴ The best individual problem solvers all have similar ways – the most successful ways – of encoding and solving problems. This enables them to outperform everyone else on individual problem-solving tasks, but when working as a group their relative cognitive homogeneity limits them. Because they are so cognitively similar “a collection of the best problem solvers performs little better than any of them individually” (Page Reference Page2007: 137). The randomly selected group is not limited in this way. Their diverse approaches to problem solving, while individually inferior to the approaches of the best individual problem solvers, interact in group settings in ways that enable the randomly selected group to, on average, outperform the group of the best individual problem solvers. In short, when it comes to group problem solving, “diversity trumps ability” (Hong and Page Reference Hong and Page2004: 16386).

Landemore aims to “exploit the potential of these findings for the epistemic case for democracy by drawing a connection between the inclusiveness of the democratic procedure and the presence of cognitive diversity” (p. 90). For Landemore, the defining feature of democracy is “the fact that it is inclusive, more or less directly, of all the members of the group for whom decisions need to be made” (p. 10). It is this fact that distinguishes democracy as “the rule of the many” from non-democratic systems such as “the rule of the one and the rule of the few” (p. 11). In Landemore's view, democracy “always includes … two complementary procedures”: inclusive deliberation and majority rule (p. 11).

Landemore's case for the epistemic superiority of democracy hinges on her argument for the epistemic superiority inclusive deliberation. This is because Hong and Page's finding that diversity trumps ability holds only for deliberative problem solving, not for voting. In fact, “what [Hong and Page's] results suggest is that all things being equal otherwise, the aggregated prediction of a large group of average citizens will not necessarily be more accurate than the aggregated prediction of a small group of experts” (p. 147). And, on Landemore's view, majority rule is simply a means of aggregating predictions about what the best course of action will be, who will do the best job in office, and so on (p. 11). So, if democracy were just majority rule, democracy would be no better, and perhaps even worse, than epistocracy. However, “assuming democracy combines majority rule and inclusive deliberation … overall, the rule of the many is epistemically superior to any version of the rule of the few” (p. 166).

Unlike in the case of aggregation (i.e. voting) where “cognitive differences among voters matter just as much as individual ability” (p. 161), Hong and Page's results “establish that, in a problem-solving context, cognitive diversity actually matters more to the production of smart collective decisions than individual ability does” (p. 90). The upshot of these results, Landemore argues, is that “all other things being equal otherwise, deliberation among more-inclusive groups is likely to produce better results than deliberation among less-inclusive groups, even if those less inclusive groups include smarter people” (p. 90). This is because a more-inclusive group will include a larger number of people, and “larger deliberating groups are simply more likely to be cognitively diverse” (p. 104). Therefore, “to the extent that cognitive diversity is a key ingredient in collective intelligence, and specifically that it matters more than average individual ability, the more inclusive the deliberation process, the smarter the solutions resulting from it should be, overall” (p. 104). So, inclusive deliberation that involves everyone is epistemically better than any form of non-democratic deliberation such as epistocratic deliberation that involves only groups of experts. This makes democracy – the conjunction of inclusive deliberation and majority rule – epistemically better than any alternative.Footnote ⁵

4. THE EPISTEMIC COSTS OF DIVERSITY

“Of course, difference does not magically translate into benefits,” says Page, “my claim that diversity produces benefits rests on conditions” (Page Reference Page2007: xxix). Landemore acknowledges these conditions, but insists that they “are not utterly demanding” (p. 102). The four conditions are: (1) “the problem be difficult enough, since we do not need a group to solve easy problems”; (2) “all problem solvers are relatively smart (or not too dumb)”; (3) “the participants think very differently, even though the best solution must be obvious to all of them when they are made to think of it”; and (4) “the initial population from which the problem solvers are picked must be large and the collection of problem solvers working together must contain more than a handful of problem solvers” (p. 102).

I believe that the second and third conditions cause more trouble for Landemore's argument than she realizes. In this section I focus on condition (3). I argue that the condition is unlikely to be satisfied by any liberal democracy and that the non-satisfaction of this condition undermines Landemore's argument.

4.3 The Costs of Cognitive Diversity

The fact that moral disagreement has adverse effects on the epistemic quality of deliberation is especially troubling for Landemore's argument given that moral diversity is likely to be positively correlated with cognitive diversity. As Page points out, cognitive diversity and moral diversity are causally intertwined: each is a cause of the other. People value different things, and this leads them to pursue different life paths in the course of which they acquire different cognitive tools. And using different cognitive tools can lead people to make different judgments about what is morally valuable, or about how to specify or prioritize various values. Moreover, both value diversity and cognitive diversity are a result of a third sort of diversity, diversity of experience. People's lives take different twists and turns, and the twists and turns they take affect both their moral outlook and which cognitive tools they develop. Cognitive diversity and moral diversity thus tend to go hand in hand. Because of this Page insists that, “we cannot consider these types of diversity in isolation” (Page Reference Page2007: 296). Maximizing cognitive diversity in practice will tend to also maximize moral diversity, and at least relative to the performance of groups in Hong and Page's models, moral diversity is detrimental to the epistemic quality of deliberation.

The more general lesson to take from this section is that diversity has epistemic costs. Landemore fixates on the epistemic benefits of cognitive diversity while ignoring these costs. Factoring the costs back in undermines her claims about the epistemic benefits of democracy. Here I have focused on the epistemic costs associated with moral diversity, but moral diversity is not the only sort of diversity that comes with epistemic costs. Cognitive diversity, by itself, has costs. For example, cognitive diversity can make communication and mutual understanding more difficult. As Hong and Page note, “problem solvers with diverse perspectives may have trouble understanding solutions identified by other agents” (Hong and Page Reference Hong and Page2004: 16389).

In calling attention to the epistemic costs of diversity I am not denying that diversity has epistemic benefits, or even that the epistemic benefits never outweigh the epistemic costs. Page provides a simple, but helpful, equation (Page Reference Page2007: 325):

$$\eqalign{& \hbox{Net Benefits of Diversity} = \cr & \quad \hbox{Gross Benefits of Diverse Tools} - \hbox{Costs of Diversity}} $$

My claim is not that diversity cannot have a net benefit. My claim is just that the net benefits are likely to be smaller and less certain than Landemore suggests because there are attendant costs to cognitive diversity that Landemore ignores. An accurate assessment of the epistemic merits of democratic deliberation must take these costs into account. In the absence of such an assessment, Landemore's claim that democracy is epistemically second to none is unfounded.

5. THE RELEVANCE OF DIVERSITY

In the forgoing section I argued that Hong and Page's third condition is not likely to be met by liberal democracies, and that the reasons why it is not likely to be met point to costs of diversity that Landemore ignores. This section shifts focus to Hong and Page's second condition requiring that “all problem solvers are relatively smart (or not too dumb)” (p. 102). The word “smart” is potentially misleading here since the condition requires not just that the problem solvers are generally intelligent, but also that they possess knowledge and skills that are relevant to the task at hand. As Page puts it, “we cannot expect that adding a poet to a medical research team would enable them to find a cure for the common cold” (Page Reference Page2007: xxix).

Landemore argues that inclusive deliberation is epistemically better than non-inclusive deliberation because including more people increases cognitive diversity and cognitive diversity improves epistemic performance. But given that diversity must be relevant in order to produce epistemic benefits, increasing cognitive diversity may not actually improve epistemic performance. In fact, given that cognitive diversity has attendant epistemic costs – a key point from the preceding section – the net result of adding irrelevant diversity might actually be a decrease in epistemic performance. So even if, as Landemore argues, the group as a whole is more cognitively diverse than any subset of the group, this greater cognitive diversity may not provide any epistemic benefits, especially once the concomitant increase in costs is factored in. The group as a whole may be epistemically less reliable than some subset of the group that contains only cognitive diversity that is relevant to the task at hand.

What is the task at hand in politics? Landemore says that, “problem solving aptly describes a central activity of democratic assemblies, whether they are trying to fix the national health care system, come up with a way to regulate bankers’ compensation levels, or find environment-friendly ways to develop the economy” (p. 97). Consider just one of these examples: fixing the national health-care system. What sort of cognitive diversity is relevant to this task? The diverse skills of doctors, health-care economists, legal analysts, and insurance experts immediately come to mind. Suppose we have a group of such individuals. Will adding a park ranger increase the epistemic reliability of the group? Adding the park ranger will probably increase the cognitive diversity of the group, but it is not clear that it increases the diversity of the group in a relevant way. As Page points out, “just because someone brings a different perspective doesn't mean that it will lead to a better solution” (Page Reference Page2007: 35). If the park ranger has little understanding of epidemiology, the economics of health care, or how various health-care systems work, the diversity gained by adding the park ranger might not improve the epistemic reliability of the group at all. It might even diminish it.

Landemore resists this sort of move claiming that, “in most cases, we cannot predict the relevant dimension of cognitive diversity in advance, because we simply cannot predict the future” (p. 111). To this she adds that, “we simply can't tell in advance from which part of the demos the right kind of ideas are going to come. It therefore does not make much sense to try to engineer cognitive diversity ex ante” (p. 112). There is a sense in which what Landemore says is true; we cannot predict in advance exactly which sorts of expertise will be useful for which political problems. Insights can come from surprising places. But to say that we cannot make any reliable predictions at all is to overstate the facts. We do have a pretty good idea of the sorts of expertise that are likely to be more useful in the long run. For example, we can predict that, in general, the cognitive skills of an actuary or a climate scientist will be more relevant to most political problems than the cognitive skills of a landscape painter or a firefighter. That is not to say that landscape painters and firefighters have nothing to add to political discussions; only that their specific cognitive skills sets will, on average, be less relevant for solving the sorts of political problems we typically face than the skill sets of actuaries and climate scientists.

Given that some cognitive skill sets are more applicable to political problems than others, the epistemically best strategy may not be the maximally inclusive one (i.e. democracy) as Landemore contends. If the epistemic costs of including individuals with less relevant skill sets exceeds the epistemic benefits, then the epistemically best strategy may be an exclusive, i.e. non-democratic, strategy that prioritizes the sort of diversity most relevant to political problems. In other words, once the requirement that diversity be relevant is adequately taken into account, some form of epistocracy might emerge as the epistemically best option. In the next section, I will argue that the limits of direct democracy, together with Hong and Page's findings, suggest just such a result.

6. DEMOCRACY OR AN EPISTOCRACY OF THE COGNITIVELY SKILLED?

Landemore argues that inclusive deliberation, which involves everyone, is epistemically superior to non-inclusive deliberation, such as deliberation among a small group of experts. But genuine inclusive deliberation is only possible at fairly small scales. As we scale up the size of a group it becomes impractical, counter-productive, and, eventually, impossible, to involve everyone in deliberation. Landemore recognizes this. “There is a point,” she says, “beyond which … including more people will simply worsen the quality of the deliberative outcomes even if we assume that it still increases the cognitive diversity of the group” (p. 105). This necessitates a shift from direct democracy, wherein the entire populace participates in institutionalized political deliberations, to representative democracy, wherein institutionalized political deliberation is entrusted to a small subset of the population. Landemore claims that this necessary shift does not undermine her argument: “the epistemic argument for deliberation among the many presented earlier translates from direct to representative democracy, provided representation reproduces at least some of the cognitive diversity of the larger group in the smaller one” (p. 107). I am not convinced that the translation proceeds so smoothly.

Suppose that we are selecting a deliberative assembly of 300 people. How should we select its members? In modern democracies the membership of such assemblies is typically decided by elections. But, says, Landemore “elections tend to bring to power socially and economically homogenous people, suggesting that the assembly is not likely to be as cognitively diverse as it should” (p. 108). To this she adds that:

In place of elections, Landemore suggests lotteries:

In short, Landemore contends that a deliberative assembly consisting of representatives selected at random from the (presumably, voting age) population will be epistemically better – it will make correct decisions more reliably – than a deliberative assembly of equal size the members of which are selected (at least in part) on the basis of their individual talents and abilities.

Landemore's argument for this claim depends on an erroneous assumption about the application of Hong and Page's results to deliberative assemblies. Landemore says:

This is mistaken for two reasons. First, as I argued in the previous sections, diversity has costs, and in some cases these costs will outweigh the benefits, especially if not all of the diversity is relevant to the task at hand. So maximizing diversity may not be the epistemically best strategy. And second, if, as Page says, “collective ability equals individual ability plus diversity” (Page Reference Page2007: xiv) then what we should aim to maximize is both cognitive diversity and individual ability.

In some cases simultaneously maximizing both cognitive diversity and individual ability might prove impossible. Some increases in cognitive diversity might necessarily come at the cost of a decrease in average individual ability, and vice versa. But it does not follow from the fact that we cannot maximize both that we ought to aim at maximizing only one. For it to be true that we should focus solely on maximizing cognitive diversity, it would have to be true that any gain in cognitive diversity is, epistemically speaking, worth any loss of average individual ability. But this is not the case. Given that diversity has costs, and that, as Page says, “any gains from diversity will be small,” (Page Reference Page2007: 320) we should expect that there will be some gains in average individual ability that epistemically outweigh the losses of cognitive diversity needed to achieve them.

There is also no reason to assume that every increase in average individual ability must come with an attendant loss of cognitive diversity. In fact, the nature of cognitive diversity suggests that increasing average individual ability will often concomitantly increase cognitive diversity. A cognitively diverse group is one that has many different cognitive tools at its disposal. “A tool,” Page says, “might be a perspective, such as the ability to think in another language. It might be a heuristic or even a set of heuristics, such as knowledge of calculus. It might be an interpretation. It might be a predictive model, or even a class of such models” (Page Reference Page2007: 103). Such tools are not evenly distributed across the population; some people have more tools than others.

This suggests two different ways of increasing the cognitive diversity of a group. We could increase the number of people in the group on the assumption that a larger group is more likely to contain a greater number of different cognitive tools. Landemore's argument that inclusive deliberation is epistemically better than non-inclusive deliberation is an argument for adopting this strategy. It rests on the assumption that “larger deliberating groups are simply more likely to be cognitively diverse” (p. 104). But a group does not need to be large in order to be cognitively diverse.

We can increase the cognitive diversity of a group by increasing the average number of cognitive tools possessed by its members. Suppose, for example, that each member of a group gains one additional cognitive tool. This will increase the cognitive diversity of the group as long as at least some of the newly acquired tools were not already possessed by other members of the group. It will also increase the average individual ability of the group given that, “all else being equal, the more tools a person possesses, the better she can solve problems, construct theories, and complete other cognitive tasks” (Page Reference Page2007: 105). Increasing cognitive diversity and increasing average individual ability here go hand in hand.

Returning to our deliberative assembly of 300 people, what selection procedure should we adopt to create the most cognitively diverse group possible? Landemore contends that we should select randomly from the population as a whole because the population as a whole is more cognitively diverse than any subset of it. But this is unlikely to maximize the cognitive diversity of an assembly of fixed size. A group of 300 people drawn at random from the population as a whole will contain some individuals that possess a below-average number of cognitive tools. Such a group will tend to be less cognitively diverse than a group of equal size that contains only individuals that possess a significantly above-average number of cognitive tools, provided, that is, that their sets of tools are sufficiently different from one another. So a procedure that selects individuals with large but different cognitive toolsets will tend to produce a more cognitively diverse assembly than one that selects individuals at random from the population as a whole. Selecting individuals with larger cognitive toolsets will also tend to maximize the average individual ability of the group given that individual ability is positively correlated with possession of more cognitive tools.

How might we identify and select such individuals with large but different cognitive toolsets? There are several possibilities. One is to select individuals from a diverse range of professions that require large cognitive toolsets containing many different tools that are relevant to solving political problems. Think, for example, of the cognitive skills necessary to be a successful behavioral economist, climate scientist, sociologist, doctor, journalist, technology entrepreneur, judge, or geologist. Many of these professions require a grasp of sophisticated statistical and modeling tools useful for understanding complex social and economic phenomena. Others require the ability to represent a problem in multiple ways, or to “think outside the box.” And all of them require many cognitive tools for integrating and weighing information from diverse and sometimes conflicting sources. It can plausibly be assumed that the members of such professions possess larger than average cognitive toolsets, and that many of their tools will be relevant to solving political problems. Alternatively, we could select on the basis of educational attainment. On the plausible assumption that education and training tend to increase the size of one's cognitive toolset, we could decide to randomly select our deliberative assembly from the set of all university-educated individuals.

Of course any means of identifying and selecting individuals with large cognitive toolboxes is bound to smack of elitism. And there are good moral grounds – concerns about political equality, autonomy, and so on – upon which to oppose a selection procedure that privileges such individuals who are, in most cases, already quite privileged. Such moral reasons are likely to play a decisive role in our political theorizing, but these are moral reasons, not epistemic ones. The fact that a selection procedure is morally objectionable does not preclude it from being epistemically best. Conversely, that some selection procedure is epistemically best does not entail that the procedure ought to be adopted. There are further questions about authority and legitimacy that must be answered first.Footnote ⁷ And once we move from ideal theories of democracy to the real world, we must also contend with questions about the political empowerment of historically oppressed groups.

For present purposes, both the details of the selection procedure and any moral objections to it are irrelevant. It is sufficient for my argument that some procedure that selects individuals with large but different cognitive toolsets is likely to produce an assembly that has both more cognitive diversity and more individual ability at its disposal than an assembly randomly selected from the population as a whole. If, as Page says, “collective ability equals individual ability plus diversity” (Page Reference Page2007: xiv), then such an assembly is likely to be epistemically superior to any democratically selected assembly. This is true whether the assembly is drawn by lottery as Landemore proposes, or selected by elections which, as Landemore says, tend to stifle cognitive diversity via selection biases and the monetary requirements of mounting a campaign.Footnote ⁸

In sum, the evidence Landemore musters in support of her epistemic argument for democracy turns out, on closer examination, to suggest that some form of epistocracy of the cognitively skilled is likely to be epistemically better than democracy. Once we factor in the costs of diversity, the requirement that diversity be relevant, and the limits of direct democracy, Hong and Page's findings actually seem to undermine, rather than support, Landemore's thesis that democracy is epistemically second to none.

7. CLOSING REMARKS: DIVERSITY, EPISTEMOLOGY, AND DEMOCRACY

I have argued that Landemore's account falls short of showing that democracy is epistemically better than epistocracy. I have also suggested that the work that Landemore appeals to in order to make her case actually supports the opposite conclusion, that some form of epistocracy is likely to be epistemically better than democracy. This reinforces Estlund's claim that an appeal to non-epistemic considerations is necessary to rule out epistocracy (Estlund Reference Estlund2008a: 40). However, it would be a mistake to conclude from this that there is nothing of value in Landemore's account of epistemic democracy. I will conclude by briefly outlining two positive points.

First, even if epistemic considerations cannot show that democracy is better than all non-democratic alternatives, they may still show that liberal societies – societies wherein liberal freedoms are protected such as freedom of expression, freedom of religion, and freedom of the press – are generally epistemically better than illiberal societies. The argument here would be that liberal societies are better at fostering cognitive diversity, and that cognitively diverse societies will likely be more innovative, and ultimately more successful, than less diverse societies. Such an argument sometimes seems to lurk in the background of Landemore's account. She says, for example, that, “democracy is likely to be smart only if it is also liberal” (p. 234).

Of course the argument from the value of diversity to the value of liberalism is not new. John Stuart Mill offered a particularly influential version of it in On Liberty. But Mill did not have results to back up his assertions. An analysis like Landemore's, drawing as it does on results like Hong and Page's, may thus buttress old arguments for liberal institutions. It may also provide an indirect argument for democracy, insofar as democracy is the only form of government compatible with liberalism.Footnote ⁹

Second, Landemore provides a framework and tools that may be used to justify particular institutional arrangements and interventions within democracies. Landemore's analysis highlights the epistemic value of diversity, which has often been overlooked by those focused on the political competence of individual voters and their elected officials. Even if, as I have argued, the epistemic value of diversity is not sufficient to justify democracy all on its own, we still have good reasons to believe that existing legislatures are not as cognitively diverse as they should be, and that the epistemic quality of such legislatures would be enhanced by increasing the diversity of their membership. Epistemic considerations may thus provide grounds for democratic reforms. This, I suggest, is where the real value of an epistemic approach to democracy lies; not in vindicating democracy against non-democratic rivals, but rather in revealing avenues by which we can build better democracies.Footnote ¹⁰