1. Overview

What is the relationship between the epistemic status of group-level attitudes and member-level attitudes? It is natural to assume that when we talk of “collective justification” or “group knowledge” we are merely making generalizations about the epistemic status of the attitudes of the group’s members. For example, we might think that a group’s belief that p is justified iff all (or a significant percentage) of the group’s members justifiedly believe that p. This is a version of epistemic summativism: the view that, roughly, to ascribe an epistemic status to a group is to indirectly ascribe it to a sufficient portion of its relevant members (where debate exists about what proportion is sufficient and which members are relevant).

Many find summativism intuitive, and it is certainly both theoretically and ontologically conservative. Consequently, some of the most promising theories in collective epistemology are explicitly summative (Goldman, Reference Goldman and Lackey2014) or at least retain summativist elements (Lackey, Reference Lackey2016, Reference Lackey2021; Tuomela, Reference Tuomela, Schmid, Sirtes and Weber2011).Footnote ¹ Yet, as I will argue, all of those accounts face a serious problem: they are incompatible with permissivism about epistemic rationality. According to permissivism, for some bodies of evidence, there is more than one rational doxastic attitude epistemic agents are permitted to hold. Analogously, permissivism allows a single body of evidence possessed by a group and its members to rationalize different doxastic states for different members of the group and the group itself. Even if all members have the same evidence, other non-evidential factors, such as the members’ cognitive goals (Kelly, Reference Kelly, Matthias, Turri and Sosa2013), or belief-formation mechanisms they take to be truth-conducive (Schoenfield, Reference Schoenfield2014), could rationally require them to form different doxastic attitudes. In such cases, the doxastic attitude that is rational for the group cannot be determined summatively as a function of the attitudes rationalized by its members, since the group could have different epistemic standards rationally requiring a divergent doxastic attitude. This might be the case if a group’s judgments, but not those of its members, play institutional functions, have diverging normative requirements, or are governed by a specific form of charter. We call this class of cases permissive divergence cases.

These permissivism-based cases can be used to support so-called divergence arguments, a type of argument that purports to show that there can be a conflict between member-level and group-level attitudes (Lackey, Reference Lackey2016, Reference Lackey2021, 56–58). Divergence arguments have been used to motivate and support non-summativist accounts of group belief (Gilbert, Reference Gilbert1987, Reference Gilbert1989), group assertion (Lackey, Reference Lackey2021, 158–163; Ludwig, Reference Ludwig2014), collective virtue ascriptions (Fricker, Reference Fricker, Tamar and John2010), collective justification (Mathiesen, Reference Mathiesen, Schmid, Sirtes and Weber2011; Schmitt, Reference Schmitt and Schmitt1994), or different kinds of collective knowledge, such as knowledge-how and knowledge-why (Bird, Reference Bird2010, Reference Bird and Lackey2014; Habgood-Coote, Reference Habgood-Coote2019; Hutchins, Reference Hutchins1995a, Reference Hutchins1995b). While the number of cases considered has accumulated, divergence arguments often rely on individual case judgments of differing intuitive appeal. Permissive divergence cases, on the other hand, are directly built on a position that, whilst controversial, has a great deal of antecedent support independent of the collective epistemology debate. Using epistemic permissivism as a starting point, therefore, gives us a general recipe for generating epistemic divergence cases that do not rely on intuitive case-by-case judgments but on an explicit theoretical commitment.

With the general idea in place, the remainder of my discussion focuses on working out the specifics. First, we need to clarify the general structure of divergence arguments and the summativist assumptions they are rejecting (Section 2). Afterward, I will introduce a very general version of epistemic permissivism and highlight some of the normative implications we can derive from it (Section 3). This, together with the assumption that groups and their members may have different epistemic standards, enables me to provide a template for constructing epistemic divergence cases. These cases can then be used to undermine summative understandings of collective rationality, group justification, and group knowledge. This will also render summative approaches to group belief implausible (Section 4). Afterward, I will address some possible objections and draw some general lessons about the nature of collective epistemic attitudes from my responses to them (Section 5). In particular, I argue that any account of the epistemic status of group attitudes needs to account for the fact that groups have member-independent epistemic requirements. In the final section (Section 6), I will give a quick overview of other types of divergence arguments found in the literature and show that permissive divergence is immune to various shortcomings of these arguments.

2. Summativism and divergence

The term “summativism” subsumes a diverse set of theories about group-level attitudes that rest on the assumption that ascribing an attitude to a group is to indirectly ascribe it to (some of) its members. To figure out, for example, whether a group believes that p summativists think that we need to sum up or aggregate the beliefs of the respective members. Similar to belief-summativism, there are also summative views about the epistemic status of doxastic attitudes such as justification or knowledge, which are the main target of the argument presented in Section 4.

Summativist views are a subclass of deflationary views within social philosophy that think of group attitudes as being dependent on, or reducible to, rather than “over and above” their member attitudes.Footnote ² While summativism makes a very specific reductionist claim, namely that the group attitudes are reducible to the respective member attitudes, there are also deflationary but non-summative views.Footnote ³ For example, group-level beliefs may be reducible to member-level behavior more broadly, and therefore depend on other member-level attitudes such as member-level acceptance.

Following Lackey (Reference Lackey2021, 20–30), we can differentiate between conservative and liberal versions of summativism. The conservative summativist claims that a group G has the attitude D iff all the members of G have the attitude D (Lackey, Reference Lackey2021, 21). On the other hand, liberal summativists only require some relevant subset of G to have the attitude D. The subset sufficient to ascribe the respective attitude to the group may be defined in general, for example, the majority as in majoritarian summativism, or it might differ based on the type and organizational structure of the group. Sometimes the summativist needs every member to have a specific attitude and sometimes it is only the majority, a significant part, or some particularly influential or important member. These liberal, more flexible versions of summativism are especially promising because they can make sense of the fact that groups can be hierarchically structured in manifold ways. Groups may have operative and non-operative members, whereas only the attitudes of the former constitute the formation of the group’s attitudes.Footnote ⁴ For example, the fact that the president or the CEO of a group has the attitude D toward p could in some cases constitute or ground the fact that the group has attitude D toward p (Lackey, Reference Lackey2021, 21).

In light of these discrepancies between different versions of summativism, it is useful to define a minimal summativist commitment <MSC> (Lackey, Reference Lackey2021, 21; Faria, Reference Faria2021, 85):

<MSC> is accepted by all summative (Cohen, Reference Cohen1989; Faria, Reference Faria2021; Quinton, Reference Quinton1976), as well as partially summative (Lackey, Reference Lackey2021, 48–53) understandings of group belief and other doxastic attitudes.Footnote ⁵ Similar to <MSC>, we can define a minimal commitment for epistemic summativism as follows:

Minimal epistemic summativist commitment <MESC> is central to fully summative views of collective justification (Goldman, Reference Goldman and Lackey2014), group knowledge (Mokyr, Reference Mokyr2002; Tuomela, Reference Tuomela, Schmid, Sirtes and Weber2011), as well as various other hybrid approaches to collective justification, which retain summativist elements (Lackey, Reference Lackey2016, Reference Lackey2021).Footnote ⁶

Consequently, any argument that renders <MSC> or <MESC> implausible or incompatible with other well-established views has a large impact on our theorizing in social philosophy. Divergence arguments do exactly that: they illustrate possible cases in which the respective MSCs are violated. The general structure of psychological divergence arguments against summative views of doxastic attitudes can be summarized as follows (Faria, Reference Faria2021, 85):

The main job of the proponent of psychological divergence arguments is to provide examples to support (2) and, therefore, demonstrate the possibility of divergence of group doxastic attitudes and member doxastic attitudes. In other words, psychological divergence arguments are usually accompanied by illustrative divergence cases that aim to serve as counterexamples to <MSC>.

Here, the distinction between deflationism and summativism becomes important. By supporting (2) divergence arguments demonstrate that group-level attitudes are not solely dependent on the respective member attitudes, but they do not show that group-level attitudes are independent from other things happening on the member level. Consequently, attempts to support (2) via divergence arguments have not only been used to support inflationary but also deflationary non-summative accounts within social epistemology.Footnote ⁷

Accordingly, we can capture the structure of epistemic divergence arguments as follows:

Interestingly, while epistemic divergence arguments as psychological divergence arguments are built on examples in support of (2′), these examples can work in two different ways. First, they can show that a group can have an attitude D with the epistemic status S while no member has the relevant attitude D (this would refute both <MSC> and <MESC>). Second, the divergence can arise because the epistemic status of the relevant member-level attitude diverges from the epistemic status of the group-level attitude (this would refute only <MESC>).

In what follows, I will introduce a new type of epistemic divergence argument of the second kind, called permissive divergence. This argument reveals that <MESC>, and thereby summativism about the epistemic status of doxastic attitudes, such as justification or knowledge, is incompatible with epistemic permissivism. While the main target of permissive divergence is summativism about epistemic states, I will also argue that retaining belief-summativism in light of my findings comes at a high theoretical cost (Section 4). In contrast to other epistemic divergence arguments, permissive divergence will turn out, not only to be more appealing, by being built on a widely defended epistemological thesis (Section 3), but also to be less vulnerable to potential objections (Sections 5 and 6). However, before I can move on to the proposed argument, some preliminaries on epistemic permissivism are needed (Section 3).

3. Types of permissivism

There are two opposing views about the relationship between evidence and doxastic attitudes central to much recent work in epistemology. While impermissivists defend Uniqueness, the thesis that there is just one rationally permissible doxastic attitude given one particular body of evidence, permissivists hold the view that one body of evidence might rationalize multiple doxastic attitudes.Footnote ⁸

One way to understand the debate is as, both permissivists, as well as impermissivists, making statements about the doxastic attitude(s) agents rationally ought to take given one body of evidence.Footnote ⁹ While impermissivists and permissivists disagree about the precision of rational requirements, both usually agree that rationality requires us to be in exactly, or at most, one of the permissible states. Here is one way to formulate permissivism:

This particular form of permissivism is called interpersonal because it relies on a specific reading of Uniqueness. As originally formulated, the quantifier within Uniqueness is ambiguous, it leaves open “whether different attitudes are permitted (or required) for different individuals on that same body of evidence” (Kopec & Titelbaum, Reference Kopec and Titelbaum2016, 191).Footnote ¹⁰ Based on two different readings of Uniqueness, we can distinguish two kinds of permissivism. While intrapersonal permissivism allows for situations in which more than one doxastic state is fully rational for the very same epistemic agent, interpersonal permissivism is “only permissive across individuals” (Kelly, Reference Kelly, Matthias, Turri and Sosa2013, 304).

Since this distinction has first been introduced, <InterP> has received a lot of attention in the literature. This is largely because <InterP> is less theoretically demanding and therefore less vulnerable to worries raised by impermissivists than its intrapersonal competitor. While any systematic comparison of different types of permissivism would exceed the scope of this paper, there is one feature of <InterP> that helps us to construct divergence arguments based on permissivism: <InterP> can be understood as explaining the permissibility of multiple doxastic attitudes by referring to the different epistemic standards had by different epistemic agents. In other words, <InterP> understands evidential support as a three-place relation between the evidence E, some epistemic standards S, and a prescribed doxastic attitude D. When criticizing White (Reference White2005), whose defense of Uniqueness is explicitly built on a two-place relational understanding of evidential support, Decker points out that “evidence supports a proposition, not simpliciter, but rather relative to an interpretation” (2012, 780).Footnote ¹¹ Decker, therefore, concludes that a more reasonable way to understand Uniqueness would be, that “given any admissible total interpretation of one’s total evidence, there is a unique rational doxastic attitude that one can take to any proposition” (2012, 782). In that sense, most interpersonal permissivists defend intrapersonal Uniqueness:Footnote ¹²

<IntraU> is compatible with <InterP> and with it, the possibility of situations in which two epistemic agents possess the same evidence but differ in their epistemic standards and, therefore, are rationally permitted to have diverging doxastic attitudes, that is, <InterP> rests on two assumptions:

1. (i) Epistemic support is a three-place relation between the evidence E, epistemic standards S and doxastic attitudes D.Footnote ¹³

2. (ii) Epistemic standards are variable.

Some impermissivists, such as White (Reference White2005, Reference White, Steup, Turri and Sosa2013) deny (i). However, there are versions of impermissivism that are compatible with accepting (i) but rejecting (ii). Take, for instance, the objective Bayesian understanding of Uniqueness defended by Hedden (Reference Hedden2015). For Hedden, the doxastic attitudes of epistemic agents should be the result of taking the “uniquely rational prior probability function and conditionalizing it on your total evidence” (2015, 470). This could be read as understanding the evidential support relation to be a three-place relation between the evidence, the unique rational prior and the doxastic attitude, whereby it is denied that there are multiple rationally permitted priors as claimed by subjective Bayesians. Hence, to end up with permissivism, we need both assumptions. Here is a very general version of permissivism that results from combining (i) and (ii):

We can understand <StandardP> as subscribing to a very weak form of uniqueness. While for some bodies of evidence there are multiple permitted doxastic attitudes, given <StandardP>, any epistemic situation (any particular body of evidence and set of epistemic standards) determines a uniquely rational doxastic attitude. Furthermore, <StandardP> does not make any assumptions about the nature of the variability: standards may vary across agents, time, contexts, or in any other evidence-independent way. As such, <InterP> is a version of <StandardP> that restricts the variability of epistemic standards to the interpersonal dimension.

To illustrate how widespread <StandardP> is, I will spend the rest of this section providing examples of epistemological frameworks that imply <StandardP>.

Let me start by elucidating further how various versions of <InterP> rely both on (i) and (ii). As illustrated above, Decker (Reference Decker2012) thinks that we can only make sense of a certain body of evidence, given a certain (permissible) interpretation. While not all interpretations are born equal, for Decker, there is often a wide range of equally coherent, and similarly fruitful, simple, or elegant interpretations available—a fact that is well-known to philosophers of science.Footnote ¹⁴ Similarly, Kelly (Reference Kelly, Matthias, Turri and Sosa2013) contemplates a version of interpersonal permissivism built on the rivaling Jamesian cognitive goals of “attaining truth” on the one hand and “avoiding error” on the other hand. He thinks that “subtly different ways of responding to the same body of evidence seem equally reasonable, given corresponding differences in the weights that we give to our shared cognitive goals” (Kelly, Reference Kelly, Matthias, Turri and Sosa2013, 302). Other proposals for defending interpersonal permissivism that are compatible with (i) and (ii), are found in Titelbaum (Reference Titelbaum2010), who thinks of evidential support as being relative to a preferred set of predicates, or Meacham (Reference Meacham2013), who points out that subjective Bayesianism implies that evidential support is relative to an agent’s prior probability functions. Schoenfield (Reference Schoenfield2014) is defending an intrapersonal permissivism that arises from a divergence in epistemic standards that are taken to be truth-conducive by the epistemic agent.

Apart from these interpersonal understandings, other theoretical frameworks that can be utilized to generate versions of <StandardP>. Take, for example, inductive risk. It is a long-known conundrum in the philosophy of science that any belief or disbelief of an empirical hypothesis H involves an inductive leap that risks accepting “H while H is in fact false, and, conversely, […] rejecting H when H is in fact true” (Hempel, Reference Hempel1965, 91–92). Since there is no general trade-off rule for inductive risk, we may argue that there are different rationally permissible ways to handle the evidence given different inductive standards (Wilholt, Reference Wilholt2009, 94). Similar things can also be said about stances (Van Fraassen, Reference Van Fraassen2002), cognitive resources (Jackson, Reference Jackson2021), or stakes (Stanley, Reference Stanley2005); all of which not only, (i) sometimes rationalize different doxastic attitudes with respect to a single body of evidence, but also, (ii) are variable in the above-considered way.

While this illustrative catalogue of epistemological frameworks that accept both (i) and (ii), is not intended to be exhaustive, it gives a good impression of the wide range of epistemic positions that could and have been used to support <StandardP>. This is an important observation that gives the upcoming argument against epistemic summativism (Section 4) additional theoretical appeal since it is directly built on <StandardP>.

4. Permissive divergence

What does permissivism tell us about the relation between the epistemic status of member-level and group-level attitudes? The two assumptions identified above that <StandardP> is based on, namely that (i) evidential support is a three-place relation and (ii) epistemic standards are rationally variable, establish that there could be cases in which the group-level and member-level attitudes rationally ought to diverge. Given one body of evidence but diverging epistemic standards, different doxastic attitudes might be rationalized for the group and its members. This provides us with a general template for constructing epistemic divergence cases:

<PDTemplate> is reliant on understanding evidential support as a three-place relation. It acknowledges that evidence simpliciter cannot be understood as a signpost that designates a particular doxastic attitude without any appeal to some epistemic standards (Decker, Reference Decker2012; Kelly, Reference Kelly, Matthias, Turri and Sosa2013). Accordingly, if we think of groups as being able to rationally acquire and possess doxastic attitudes, we need to think that they possess, or are governed by some epistemic standard(s) which make(s) sense of the evidence possessed by the group. Otherwise, no doxastic attitude can ever be rationalized for G.

In real-life scenarios, the group’s epistemic standards S will often be represented by an idiosyncratic modus operandi of the group. Groups, such as juries or investigative panels, have charters or rules of conduct. Scientific groups often have explicit or implicit norms and are shaped by their inner-disciplinary and wider scientific culture, defining proper and improper inquiry, methodology, and research questions within a paradigmatic framework. While some of these modi operandi govern non-epistemic aspects of the group’s functioning, others will influence the group’s belief formation processes in a way that is analogous to how epistemic standards govern the belief formation processes of individuals. While what exactly those processes are will depend on the exact nature of group beliefs, there are many plausible candidates compatible with a wide range of understandings of group-level attitudes. For example, in scientific research groups, there are often norms influencing how much weight is given to research articles published in journals of varying qualities or different test results from specific types of experiments. Other examples include norms governing representative samples, sample sizes, or statistical thresholds for labeling something a discovery, such as the famous “5 sigma rule” in particle physics.

These observations, together with the provided template <PDTemplate> give us an instruction on how we can cook up scenarios in which member-level and group-level attitudes rationally ought to diverge. The only thing left to argue for is that there actually are such cases. This should be an easy task given the long and diverse list of possible epistemic frameworks that could be used to construct such cases (Section 4). Take, for instance, this illustrative example:

Consider, for example, Ambitious Amy, one of the theoretical physicists of the scientific collaboration G. While Ambitious Amy desires to avoid errors, her primary epistemic goal is to attain true beliefs. As such, she has a low but not unreasonable liberal epistemic threshold for believing empirical propositions about prospective climate developments. This allows her to mistrust the scientific methodology underlying the group’s suspension of judgment about p. Simultaneously, however, Amy understands that due to the inherently complex nature of the research, her conclusion that p would be regarded as being overzealous given G’s epistemic standards S* (in this case their epistemic goal of avoiding error).Footnote ¹⁶ In other words, Amy is perfectly capable of processing the data and inferring certain results in accordance with S*. However, while she believes that the collaboration would not be justified to believe that p, given S*, she is simply not convinced that G is right in being so cautious with respect to p. Given that being epistemically ambitious in Amy’s way is an epistemically permissible endeavor, she is rationally permitted to believe that p is based on the body of evidence E possessed by the group. Since equally all other members of G could be in the same situation, this directly refutes <MESC>.

As constructed, DIVERGING STANDARDS directly violates any summative understanding of rational group belief: while G rationally suspends judgment about p, all of G’s members rationally believe that p. Equally, any summative understanding of group justification is rendered implausible. The fact that the doxastic attitudes of both G and all members m₁–m_n are based on E, as well as S₁–S_n and S*, respectively, ensures that these attitudes are rational. <StandardP>, as a thesis about how doxastic attitudes are rationalized by a body of evidence E makes restrictions on propositional justification, the kind of justification that arises from the mere fact of having sufficient evidence for a proposition p (Mathiesen, Reference Mathiesen, Schmid, Sirtes and Weber2011, 360–361). Given <StandardP>, multiple doxastic attitudes toward p may be propositionally justified for different agents given one body of evidence E, but at most one attitude is propositionally justified for one agent having a distinct set of epistemic standards S. That is, the suspension of judgment on p is propositionally justified for G independent of the members’ doxastic attitudes and their justificatory status. Furthermore, as described in DIVERGING STANDARDS, m₁–m_n properly base their beliefs that p on E, while G also properly bases its suspension of judgment on E, which makes their attitudes not only propositionally but also doxastically justified (Turri, Reference Turri2010, 312–314).

Given the traditional analysis of knowledge as (non-gettierized) justified true belief, this similarly proves summative understandings of group knowledge to be wrong. If both the group G and all members m₁–m_n can have diverging but nonetheless justified beliefs, at most one, either G or m₁–m_n, could have a true belief. In other words, by stipulating either that p is true or that p is false, we can generate a divergence case for knowledge.Footnote ¹⁷

Interestingly, this also has consequences for how we think about group beliefs, despite permissive divergence being about the epistemic status of doxastic attitudes rather than doxastic attitudes per se. When confronted with permissive divergence cases, the belief-summativist is in a dilemma: if they insist that the group cannot have a doxastic attitude that diverges from all its member’s attitudes, they are forced to accept that there are cases in which the epistemic status of the member-level beliefs and group-level beliefs diverge. So presented with DIVERGING STANDARDS, the summativist needs to say that the group irrationally believes that p (since G’s standards S* require suspending judgment about p) in virtue of all members m₁–m_n rationally believing that p is based on E and S. For the summativist, the only way to ensure that both the group and the members are rational is to ensure that the group and individual standards align (which leads to all kinds of problems which I will discuss in the next section [Section 5]). The non-summativist about belief, by contrast, by allowing diverging belief states as well as diverging justificatory statuses, makes room for both the group and its members to hold rational beliefs in DIVERGING STANDARDS. Thus, the best way to understand cases such as DIVERGING STANDARDS is as cases in which both the doxastic attitudes and the epistemic standards of the members and the group diverge.

In sum, <StandardP>, along with the observation that groups and their members may have divergent epistemic standards, gives us a recipe for crafting counterexamples to <MESC> and consequently summativism in general. This recipe is encapsulated in the template <PDTemplate>, which inspired the concrete example DIVERGING STANDARDS. While this illustrative example enabled us to demonstrate a direct conflict between epistemic permissivism and summativism about the epistemic status of doxastic attitudes, such as justification or knowledge, it also rendered summative understandings of group belief implausible.

This simple method makes permissive divergence theoretically appealing since it is directly built on (a very general version of) permissivism, a thesis that has a great deal of antecedent support independent of the collective epistemology debate. In the remainder of the paper, I will demonstrate that in contrast to other divergence arguments found in the literature, permissive divergence is not only theoretically more appealing but also less vulnerable to potential objections. I will do so by first introducing and discussing potential summativist defense strategies (Section 5). Afterward, I will compare permissive divergence to other divergence arguments (Section 6) and illustrate that it is immune to multiple concerns raised against these competitors.

5. Permissive divergence: Discussion

The most straightforward objection to the above-presented argument is to deny that groups can have epistemic standards in the same sense as individuals do. Based on the discussion above (Section 4), I can resist such kind of skepticism in two ways. First, if we think that epistemic support is a three-place relation (as implied by <StandardP>), we need to think that groups have epistemic standards that tell us which attitudes are propositionally justified given the group’s evidence. Otherwise, we cannot make sense of the epistemic status of group attitudes at all. That is, any skepticism against the idea of group-level epistemic standards would lead to general skepticism with respect to collective epistemology. Second, if we take epistemic standards to be mechanisms that have a systematic influence on the formation of doxastic attitudes, given a body of evidence E, there are numerous plausible examples of group-level epistemic standards. In particular, I have mentioned various scientific standards governing the weighing of different types of evidence, the ways experiments are designed and conducted, and how scientific findings are reported. Since these mechanisms have a direct influence on the overall behavior of the group, given minimal assumptions about the nature of group beliefs, I think these mechanisms are best understood as group-level epistemic standards.Footnote ¹⁸

Another way to try to resist permissive divergence arguments is to challenge the assumption that epistemic standards can diverge among group members and the group itself. Let us call the view that group-level standards cannot diverge from member-level standards, standard summativism. As with other types of summativism, we can define a minimal commitment as follows:

Unfortunately, accepting minimal standard summativism <MSS> does not protect the summativist from permissive divergence arguments. First, even if <MSS> is true, to avoid permissive divergence cases, the relevant epistemic standards would need to be aligned with the group’s doxastic attitudes in the right way. Otherwise, we can still have situations in which G has the doxastic attitude D with the epistemic status E, while no member m₁–m_n of G has an attitude D with the epistemic status E. Namely, in cases in which G’s attitude D is rationalized by standard S, which G has due to m₁–m_n having S, while m₁–m_n do not have the doxastic attitude D (while any member that has D does not have the relevant epistemic standards).Footnote ¹⁹

Second, the fact that groups are composed of independent epistemic agents, each with their own independent epistemic life makes standard summativism very implausible. So for <MSS> to be viable the summativist not only needs to deny that the norms identified above (Section 4) are group-level epistemic standards, but also that there is some mechanism that ensures that member-level and group-level epistemic standards do not diverge.

In so doing, standard summativists might claim that epistemic agents can perhaps endorse one set of beliefs privately based on S but need to temporarily adopt diverging epistemic standards S* (together with the relevant doxastic attitude) when acting as a member of G. For example, in DIVERGING STANDARDS, Ambitious Amy adopts S* when she is in “scientist mode” but exchanges S* with S when she is at home in “ambitious mode.” Assuming that, we then treat Amy_[scientist] and Amy_[ambitious] as distinct epistemic agents, who can permissibly believe different things.

The problem with this framing of the case, however, is that Amy seems to be able to voluntarily switch her epistemic standards from S to S*. This would transform Amy’s doxastic attitudes in a way that they provide an unstable basis for action and practical deliberation. This concern is discussed as the so-called arbitrariness worry in the literature on epistemic permissivism (Kelly, Reference Kelly, Matthias, Turri and Sosa2013; White, Reference White2005, Reference White, Steup, Turri and Sosa2013). The basic idea behind this worry is described by White as follows (2005, 455):

Accordingly, the most prominent versions of permissivism defend themselves against the arbitrariness worry by denying that voluntary switching of epistemic standards is allowed. Take, for example, Schoenfield’s (Reference Schoenfield2014) interpersonal permissivism. While for Schoenfield, there are potentially many different sets of permissible standards rationalizing different attitudes regarding the given evidence E, epistemic agents cannot switch them voluntarily. From their individual perspective, rational agents must be immodest, that is, they believe that their way of forming doxastic attitudes is truth-conductive, which, in turn, rationally demands them to hold onto their standards (Schoenfield, Reference Schoenfield2014, 201–202).Footnote ²⁰ In other words, treating Amy_[scientist] and Amy_[ambitious] as different epistemic agents conflicts with <InterP> (as well as most intrapersonal versions of <StandardP>), as traditionally conceived.

However, regardless of whether we think that rational agents ought to be immodest or not, the burden to prove that members switch their epistemic standards in permissive divergence cases is, after all, on the summativist. In other words, the summativist not only needs to show that Amy and all other scientists can and ought to switch back and forth between standards (from S₁–S_n to S* and vice versa), but also that this always and inevitably happens. Otherwise, we would still have permissive divergence cases.

One way to do this would be to insist that when acting as a member of a group, individuals necessarily form attitudes qua group member and not qua (independent) individual.Footnote ²¹ Hence, the <MESC> could be updated accordingly:

While <MESC*> illustrates a way in which epistemic agents could and do voluntarily, but not arbitrarily, switch between different sets of epistemic standards, it is already a significant deviation from the initial summativist position. First, it admits that groups have an epistemic life of their own that forces certain attitudes and “modes of thinking” onto their members. Second, even if we think that this voluntary switching of standards is unproblematic, it is far from clear that the attitudes adopted qua group members are genuine doxastic attitudes. As a group member, you can be perfectly aware of the epistemic standards of the group and how you ought to act according to them without temporarily forming a diverging doxastic attitude toward p. In other words, none of the scientists in DIVERGING STANDARDS needs to actually suspend judgment toward p to display the described behavior and to understand that suspension is the rationally required attitude relative to G’s standards. (The scientists can even be immodest in not acknowledging that G’s and their own standards are rationally on a par.)

This becomes especially clear when we think of instances in which agents are members of multiple groups simultaneously. If m is a member of two groups with different epistemic standards (but the same evidence), one that mandates believing that p, and another that mandates suspending on p, m would be committed to suspending judgment toward p and believing that p at the same time. The only way to make sense of these cases, without requiring m to have an incoherent set of doxastic attitudes, is then, to understand qua-group member attitudes not as genuine doxastic attitudes.

In sum, the best explanation of permissive divergence cases is that there are divergent standards in place which rationalize the group’s attitudes. This demonstrates that <MSS> is false. While it is an open question how group-level epistemic standards relate to the member’s behavior and the member’s qua attitudes, there is no reason to think that individuals form novel doxastic attitudes when acting as members of a group. And, for (epistemic) summativism to be true, the attitudes must be genuinely doxastic.

These considerations provide us with some general lessons about the epistemic status of collective attitudes, as well as a recipe for how to modify existing views to account for these insights. First, every understanding of the epistemic status of group-level attitudes needs some specification of the relevant agent for whom the evidence provides sufficient reason for belief. Here, the presented argument shows that we cannot take the relevant agent to be (some) member(s) of the group—rather, the relevant agent must be the group itself. Second, we need to tell a substantial non-summative story about what it means that something is rational for the group to believe; that is how the group’s epistemic standards come about.

Take, for example, Silva’s (Reference Silva2019) Evidentialist Responsibilism for Groups <ERG>, which I take to be one of the most promising extant accounts of collective justification. For Silva, a group G justifiedly believes that p on the basis of evidence E iff (2019, 263):

First, <ERG> needs to be modified by making it clear that E needs to be a sufficient reason to believe P for G. This is done relatively easily by changing clause (1) accordingly.Footnote ²² Second, <ERG> needs to be supplemented with an additional (non-summative) story of how the group’s epistemic standards come about. While there are manifold options, each of them comes with further theoretical commitments that need to be argued for independently. Maybe group epistemic standards are chartered into the group via legislation, or maybe they supervene on some non-doxastic qua member attitudes, or maybe they need to be read off the most rationalizing interpretation of the group’s behavior.

While discussing any of these suggestions would exceed the scope of the paper, I think that there is another general lesson that can be drawn from the discussion above. Whatever the nature of collective epistemic standards is, the version of epistemic permissivism we subscribed to gives us important insights into answering that question. For example, if we think that different ways of weighting the cognitive goals of “attaining truth” and “avoiding error” rationalize different doxastic attitudes (given one body of evidence), we need to have an underlying understanding of what it means for the group to weigh these cognitive goals in a particular manner. While this is not to say that extant views in collective epistemology (such as <ERG>) cannot be modified to implement these insights, it illustrates that significant work still needs to be done to do so.

6. Permissive divergence: Comparison

Having demonstrated the resilience of permissive divergence arguments against potential objections enables me to systematically compare them to other epistemic divergence arguments found in the literature. One major virtue of permissive divergence is that it is immune to multiple concerns raised against its competitors. I will reconstruct some of the worries here and demonstrate how and why permissive divergence does not succumb to them.

The first set of problems faced by other epistemic divergence arguments, particularly those proposed by the so-called distributed cognition accounts (Bird, Reference Bird2010, Reference Bird and Lackey2014; Hutchins, Reference Hutchins1995a, Reference Hutchins1995b) has to do with the connection of group-level and member-level cognition. According to distributed cognition accounts, socio-epistemic entities can form doxastic attitudes by directly processing and deliberating evidence, which is spread among the group’s members or integrated into the group’s structure. Since this can happen without the awareness of most or all of the group’s members, the doxastic states of the group produced in this way may differ from any of the doxastic member states. So, distributed cognition accounts allow “an organic group to know that p without individual belief that p, joint acceptance that p, commitment that p; indeed, without a single individual person—group member or not—even being aware of that p” (Lackey, Reference Lackey2021, 113).

This raises various problems. First, for Lackey, this severs the vital connection between group knowledge, group action, and collective responsibility (Reference Lackey2021, 115–123). Second, it is not entirely clear how to understand the collective evidence possessed by the group in these cases. On the one hand, distributed cognition accounts have been criticized for leading to unnecessary and implausible inflation of the group’s evidential base: any evidence possessed by any group member that could play a functional role within the group would count as part of the group’s evidence (Habgood-Coote, Reference Habgood-Coote2019, 948). On the other hand, these accounts have problems with non-integrated evidence possessed by individual members that could serve as a defeater (Lackey Reference Lackey2021, 123–126). Accordingly, other promising accounts of group justification often demand the group to partly disclose their evidence (Silva, Reference Silva2019), or at least that disclosure would not undermine the group attitude (Lackey, Reference Lackey2016, Reference Lackey2021).

Without judging the legitimacy of these worries, we can immediately demonstrate that permissive divergence is immune to them. First, situations of permissive divergence can arise even if the relevant body of evidence is fully disclosed, for example, via deliberation within the group. With the group and all of its members sharing the exact same body of evidence, any worries about defeaters or willfully excluded evidence are effectively ruled out.

Second, the permissive divergence case described above, Section 4 assumes mutual awareness not only of the evidential disclosure but also of the inner cognitive workings and the modus operandi of the group. This means that permissive divergence does not threaten the connection between justified group belief/knowledge and group action as distributed cognition does. In permissive divergence cases, members are aware of what is rational for the group to believe, and this awareness makes it rational for the members to act on the group’s behalf even though they do not themselves (rationally) believe that p (Section 5).

Another worry with extant epistemic divergence arguments is that even if they do presuppose evidential disclosure and mutual awareness, they may indirectly manipulate the evidential base of the group by having different notions of what counts as evidence on the individual and collective levels. One often-discussed case, concerning a jury deliberating the guilt of a defendant, is found in Schmitt (Reference Schmitt and Schmitt1994).Footnote ²³ In this case, all jury members possess an epistemically reliable piece of hearsay evidence supporting the guilt of the defendant, which is ruled out as being inadmissible by the judge. Schmitt thinks that while the jury as a chartered group justifiedly believes that the defendant is innocent, none of the jurors justifiedly believe this proposition because their justification is defeated by the relevant reliable hearsay evidence. This is possible since the group’s charter prohibits it from considering the hearsay evidence and, therefore, excludes it from its evidential base. In its “legal capacity, the court rightly excludes hearsay, and its legal capacity is the only capacity in which it operates” (Schmitt, Reference Schmitt and Schmitt1994, 274).

However, if the evidential base can be changed in this way, we face similar worries to the ones concerned with distributed cognition. Namely, epistemically relevant pieces of evidence or defeaters possessed by the members may be left unconsidered. After all, as pointed out by Lackey, the exclusion of hearsay is based on “practical or procedural concerns but not necessarily epistemic ones” (2016, 355). That is, the court deeming the evidence as being inadmissible is not based on its epistemic properties, such as its reliability or truth-conduciveness (Lackey, Reference Lackey2021, 68).Footnote ²⁴ Treating these cases otherwise would open the door for instances in which groups willfully manipulate their evidential base (Lackey, Reference Lackey2021, 68–69). In other words, if declaring the hearsay evidence inadmissible would be an epistemic rather than a juristic-pragmatic decision, we could also imagine groups being chartered in a way that allows them to manipulate their evidential base inappropriately by, for example, cherry-picking evidence that fits their practical goals.

While I agree that any cherry-picking of evidence is unacceptable, permissive divergence only relies on different evaluations of the same body of evidence via different rationally permitted methods based on different epistemic standards. We do not need to allow for any cases in which there are defeaters possessed by any of the group members––the evidential base itself is not manipulated or changed in any way (as suggested by Schmitt).

This also shows that permissive divergence arguments are compatible with a wide range of views about group evidence, including various kinds of summativism about group evidence, which understand the group’s evidence as being the evidence possessed by some (or all) group members. After all, permissive divergence cases are possible even if the group’s evidential base is the same as the evidence possessed by all of the group’s members.Footnote ²⁵

Furthermore, any divergence arising from diverging epistemic standards is still epistemically rational. While the juristic standards governing a court may sometimes lead to epistemically irrational group attitudes, in cases like DIVERGING STANDARDS the divergence is purely epistemic.Footnote ²⁶

In sum, permissive divergence distinguishes itself from its competitors in allowing for divergence even under very stringent conditions, namely: (a) full disclosure of the evidence within the group, (b) mutual awareness of the disclosure and the modus operandi of the group, even if, (c) the group G itself, as well as all members m₁–m_n are fully rational.