When writing about the metaphysics of social institutions, philosophers rightly emphasize that institutions’ existence and nature are “up to us” in a deep and important way. Objects such as insects and stars could exist with the same basic nature in the absence of human activity; corporations, governments, and religious institutions could not. This feature of institutions makes it natural to think that their existence and nature depend on our thoughts in some way, and accounts of institutions that appeal to mental phenomena dominate the literature.1 But there are also important ways in which institutions’ existence or nature can be less “up to us” than we might initially expect, which tend not to be emphasized. I’d like to try starting with these kinds of cases—that is, with cases that highlight ways in which institutional reality can get away from us. In my view, these cases push us toward a behavioral approach to social institutions. According to this approach, institutions are made up of behavior that clusters into copied roles, and the individuals who engage in that behavior, where all of the roles together promote some particular result(s).2 I’ll motivate this behavioral theory of institutions, discuss and develop its features, and finally, compare it to other accounts.

It is important to note that by ‘social institution’ I mean a complex, structured social entity such as the Colombian government, the Catholic Church, the U.S. National Science Foundation, Yamaha Corporation, or Major League Baseball. I will offer a behavioral account of just this specific kind of social entity, and not of social entities, tout court. For instance, social norms, laws, and social groups fall outside of the present paper’s scope.3

We’ll begin with three cases in which institutions seem to be less “up to us” than we might expect. Though they are inspired by real-life examples, they are described in hypothetical terms, to allow for the stipulation of certain details.

Institutions, then, are in various ways less “up to us” than we often assume. We might all share beliefs about the existence and nature of an institution and it can either fail to have the nature we think it does (as in Racist Courts) or fail to exist at all (as in Watchdog). These two cases highlight ways in which we have less power than we might think to determine the existence and nature of institutions. On the other hand, an institution can exist with a determinate nature even though we disagree about what that nature is (as in Disputed Government). This last case highlights a way in which institutions and their natures are less dependent on us than we might suppose.5

To begin to construct a theory on the basis of these cases, we need to find features that unite Racist Courts, Disputed Government, and paradigmatic examples of institutions, and which are absent from Watchdog. My initial suggestion is the following: in all cases in which an institution exists, there is behavior that clusters into roles, working together to promote some particular result(s). In Disputed Government, although there is disagreement about the electoral process and thus about who is the leader, only one person is actually able to enact the leader role in ways that promote the government’s results (such as peace and order within the society it governs). And in Racist Courts, there is behavior that clusters into roles such as attorney and defendant, of which everyone is aware, but what everyone is not aware of is one of the results that the behavior promotes: over-incarceration of a certain racialized group. Moreover, in cases that most theorists would treat as paradigmatic, such as a typical university, there is agreement about the institution’s nature and purpose, but there is also actual behavior that clusters into roles, together promoting certain results (specifically, knowledge generation and transmission). On the other hand, in Watchdog, we can see that although there is agreement about the nature and purpose of the purported institution, no institution is actually created because the requisite behavior is absent.6

However, the picture of social institutions just suggested cannot differentiate institutional reality from mere social reality.7 Imagine a neighborhood in which one person, Juana, is so kind and witty that her company is unusually enjoyable. Juana’s neighbors visit her more frequently than they visit anyone else, to seek social interaction with her. Imagine too that Juana tends to refrain from seeking social interaction because her visitors meet her socialization needs. Here we can distinguish one behavioral role occupied by Juana, and another occupied by the other neighbors, with all of the roles together promoting the fulfillment of Juana’s socialization needs. On the picture we’ve been considering, this would be a Juana-centered social institution. But it’s clear that all we have here is the merely social phenomenon of Juana being popular. Similarly, consider social phenomena among nonhumans. Ants in a colony evince behavior that clusters into roles (such as queen, nurse, and forager), together promoting the ants’ survival and reproduction (Gullan and Cranston, 2014, p. 337). But we wouldn’t want to call an ant colony a social institution.

We need to differentiate institutional reality from mere social reality while still properly categorizing Racist Courts, Watchdog, and Disputed Government. There are three features that I’ll propose distinguish social institutions from mere social reality, falling into two main categories: momentum, and optionality. I’ll motivate these features and then argue that all three are needed to differentiate institutions from mere social reality.

First, momentum: social institutions involve one particular way of doing something getting some momentum behind it and carrying itself forward. One aspect of this momentum, in my view, is that some of the behavior within an institution is copied from past behavior: things are done a certain way because that is how they were done in the past. I follow Millikan (2005, pp. 3, 5, 31) in seeing copying as a matter of causal connections among features of particular instances of behavior, which does not require any particular intentions or even awareness from the copying individual (cf. Stotts, 2023, p. 2123). If I flick my wrist in a particular way when cooking pancakes because my father did so, then I am copying him, though I may never realize I am copying him. As I see it, individuals enacting institutional roles are not making completely fresh decisions about how to act within their roles; rather, they are causally influenced by what has been done in the past.

But there is more to institutional momentum than copying, leading to a second differentiating feature for social institutions: the vacating and passing on of institutional roles. In my view, institutional momentum is strong enough to go beyond any particular set of individuals. Once something has become an institution, as opposed to just some particular individuals’ way of doing things, it goes on to influence the behavior of other individuals as well, leading roles to be vacated and passed on. In this way, the institution takes on a life of its own (cf. Millikan, 2017, p. 20).

In addition to institutional momentum, institutions have a distinctive kind of optionality: when people form a new institution, they are instituting one particular way, among others, of doing something. Their way of promoting a certain result was not the only option, and thus it needed to be instituted in order to become their way of doing things. This motivates the third key feature of social institutions: equally accessible alternative ways to promote the institution’s result(s).8 Crucially, the alternatives need not be equally good. If there is one objectively best form of government, governments taking that form would still be social institutions. There would still be the right kind of optionality to the process of instituting that particular form of government, as opposed to others that might have equally easily been instituted instead.

So, there are three features that I take to be distinctive of social institutions: copying, the vacating and passing on of roles, and alternatives. Having provided some initial motivation for these features, I’ll show how they differentiate social institutions from mere social reality.

First, the ant colony. Ants do vacate and pass on roles as they age and as the colony’s circumstances change. But ants do not copy each other. They do not learn how to be foragers or nurses from other ants; rather, they are just responding to changes in their environment, in ways presumably shaped by their genes (Gordon, 2010, pp. 30–44). They also do not have equally accessible alternatives to living in colonies (p. 15).

Second, the popularity case. That case does involve equally accessible alternatives: there are other ways of promoting fulfillment of Juana’s socialization needs, such as offering her free transportation to visit friends in other neighborhoods. But, the popularity case does not involve copying. People enact the roles of popular person, friend of popular person, or seeker of popular person’s friendship, but without copying each other in doing so. They are just (we’re supposing) independently motivated to pursue Juana’s company. The popularity case also does not involve the vacating and passing on of roles. It has all of the optionality of an institution, but none of the momentum.

At this stage, there may be some doubt about whether all three features are necessary. Neither the ant colony nor the popularity case involves copying, so couldn’t we require just copying and omit the other two features? Or, given that the ant colony does not involve equally accessible alternatives and the popularity case does not involve roles that are vacated, couldn’t we require just those two features and eschew copying? To assuage this worry, I’ll argue that no two of the three features can rule out all cases of mere social reality, without the addition of the third.

First, if we required both equally accessible alternatives and the vacating and passing on of roles, but did not require copying, a series of transient social arrangements could count as an institution. Imagine that at some particular beach on Saturday, every two hours a different group of four people happen to play Bocce on the same stretch of sand. So, the players’ roles get vacated and passed on, and there is optionality: there are other options for leisurely outdoor competition. But, there are no causal connections between the successive groups of people. This mere coincidence would amount to a social institution, just as much as if the players had organized a weekend Bocce club. Requiring copying rules out this kind of case.

Second, consider what would happen if we required both the passing on of roles and copying, but not equally accessible alternatives. If we were then to discover that ants do, in fact, learn from each other how to forage or build nests, we’d have to count ant colonies as social institutions. I’m not averse to countenancing social institutions among non-humans, but the mere addition of copying to ant colonies seems insufficient. If their options are to learn from older ants how to live together in an ant colony, or to die, that seems fundamentally different from the way humans establish social institutions in the face of multiple options.

And finally, consider what would happen if we required equally accessible alternatives and copying, but not the passing on of roles. This last requirement is perhaps the most controversial of the three. But consider, for instance, a group of people stranded on an island who organize themselves into leader, judge, and citizen roles. Over time, all participants form habits in enacting their roles, thus being causally influenced by their own past behavior as role-holders (which does count as copying). They, of course, had other options for organizing themselves, and all of the roles together promote the result of peace among them. In my view, this arrangement falls short of institutionality. They have certainly created a social structure, but whether it will have the staying power to become an institution has yet to be seen. If they all die before anyone has passed on a role, it seems to me that we can say that they organized themselves effectively while they were alive, but they did not establish a new institution. On the other hand, if some of them have children to whom they pass on their roles, or even if just one role shifts from one person to another, then we have a social structure with the right kind of momentum behind it to count as an institution. It has some amount of staying power, beyond the habits of any particular set of individuals. Institutions may vary in how much of this kind of staying power they have, but they all have it to some degree.9

Here, then, is the account of social institutions I propose, which I will call the Behavioral Theory of Institutions (BTI):

The BTI captures the key elements that our cases other than Watchdog shared (namely, behavior that clusters into roles, working together to promote some particular result(s)). It also has the three key features that differentiate social institutions: copying (in condition 1), the passing on of roles (condition 2), and alternatives (condition 3). I’ll offer some clarificatory comments on various aspects of the account.

As a brief comment about condition 1, I want to emphasize that in some institutions, the only past behavior a role-holder copies may be their own. This can happen in institutions where only some roles have thus far been passed on. We can be causally influenced by our own past behavior, and thereby copy ourselves in the required sense.

Regarding condition 2, when I refer to the “institution’s history,” I am talking about its history in a broad sense, where things that happened within a social structure that was not yet an institution but would later become one can be considered part of that eventual institution’s history. So, there will have been a moment when there was a social structure that satisfied conditions 1 and 3, but not yet condition 2, and then a role was vacated. This was not yet a social institution. But then, when a new individual took on that role, it became an institution. That first instance of the role being vacated and passed on is part of the institution’s history, even though it occurred before the social structure met all criteria for institutionality.

Condition 3 requires more extensive discussion. First, a few brief clarifications. The result(s) which the behavior within an institution promotes must be specified in non-institutional terms. Otherwise, we will not have an account of where institutional reality comes from. Moreover, results cannot be disjunctive (so, we cannot say that the National Science Foundation promotes the result of “scientific discovery or artistic achievement”). Individual results must be considered individually, to determine whether each is in fact promoted by the institution.

More should also be said about the notion of “promoting” a result. To promote a result is just to make that result likelier to obtain. The results an institution promotes (in the way condition 3 requires) are a central part of its nature. They are what the institution does—its contributions to the world. Participants in an institution may be unaware of the results it promotes; in fact, institutions often have unintended results. For instance, participants in a vast number of institutions, from the Catholic Church to Major League Baseball to the American Philosophical Association, causally inter-enable each other to promote pollution (among other results, of course). This, according to the BTI, is as much a part of their nature as the results that were intended.

Also related to condition 3, I’d like to say a bit more about the notion of alternatives being equally accessible. When an institution is well established, its ways of doing things often become much easier to use than the alternatives, sometimes to the point at which there no longer seem to be real alternatives at all. This feature is part of what makes institutions so powerful, for good or ill. So, when I say that an institution must have equally accessible alternatives, what I mean is equally accessible once we have factored out the influence of the copying that has happened within the institution, or in other words, once we have factored out institutional momentum. This still captures the optionality that characterizes institutions, at least at the time of their founding.10

The final aspect of condition 3 that we’ll discuss is its requirement that promotion of the institution’s result(s) be done by “all roles in the institution in a way that is causally inter-enabling.” We’ll simultaneously discuss the nature of roles, as the BTI posits a tight relationship between institutional roles and institutional results. To simplify the discussion, we’ll leave conditions 1 and 2 aside for now. So, we’ll be considering a pre-institutional social structure that satisfies condition 3, where copying has not occurred and no roles have been passed on.

A pre-institutional social structure, and an initial set of roles within it, emerges when it comes to be the case that there is a set of individuals engaging in behavior that is interconnected in two ways, which we’ll discuss in turn. First, the behavior must be interconnected by a relation I will call ‘reciprocal exchange of behavior’ (REB), which is central to my understanding of roles. When one individual engages in behavior of kind A and refrains from behavior of kind B, and another individual engages in B and refrains from A, those two individuals stand in the REB relation (Stotts, 2024). Each individual who is part of the social structure is such that, for each other individual, they either stand in the REB relation to that second individual, or they stand in the REB relation with a third individual who is part of a chain of REBs that ultimately connect to that second individual. This is what is meant by the BTI’s statement that the behavior in an institution “clusters into roles,” where each role is defined by its REBs (that is, by various behaviors in which holders of that role engage while others abstain, and the behaviors from which its holders abstain while holders of other roles engage in them). For example, imagine a professor (at an emerging university, where roles have not been copied or passed on) who generates syllabi, gives out assignments, and submits grades while others (i.e., the students) refrain from those activities but do attend classes, take notes, and write essays, while the professor refrains from their activities.

The second way in which the behavior in such a pre-institutional social structure must be interconnected is by causally inter-enabling each other to promote some result(s), in line with condition 3. This is my way of spelling out the intuitive idea, used earlier in the paper, of roles promoting a result “together.” It means that there is at least one result that all roles promote, where each role’s ability to promote that result has as a partial cause some of the behavior associated with some other role.11 So, for instance, students’ behavior of attending classes is a partial cause of the professor’s success in promoting knowledge transmission, as is maintenance staff’s behavior of removing debris from the classroom. Only roles that contribute to the pre-institutional social structure’s result(s) in this way are roles within that structure.

Now, let’s bring conditions 1 and 2 back in, returning to full-fledged institutions. Condition 1 has bearing on the nature of institutional roles, because copying is what individuates institutional roles across time. Specifically, a presently enacted role is the same role as the past role with which it shares the greatest number of causal links formed via copying.12 This applies even if some behaviors within the role have changed (cf. Harré, 1979, p. 99). For example, consider a corporation in which several job duties shift from administrative staff to communications staff. If there is a tie with respect to this causal criterion between two (or more) distinct past roles, the current role counts as a merger of those past roles. This causal way of identifying the same role across time means that different institutions cannot have any of the same roles within them. Roles here are institution-specific.13

The causal individuation of institutional roles also means that some behavior that does not promote the institution’s result(s) can still be part of an institutional role, provided that it stands in the REB relation to other role-holders in the institution and that it was copied from the past behavior of occupants of that role. This allows roles within an institution to have some aspects that do not contribute to the institution’s result(s), perhaps due to poor planning. For instance, imagine a certain worker at a corporation is required to write an annual report, which contributes nothing to the institution’s results. Writing this report can still be part of their role.

One final remark I’d like to make about roles at this stage is that the roles that actually exist within institutions may be decidedly less fine-grained than everyday talk suggests. Consider empty promotions within a corporation. A corporation may have, for instance, Analyst 1 and Analyst 2 positions on the books, where the job duties and compensation do not actually change when one is “promoted” from Analyst 1 to Analyst 2. The BTI implies that there really aren’t two different roles in that scenario; just one, for which participants in the institution mistakenly (or perhaps, deceptively) have two different names.

The BTI specifies the conditions individually necessary and jointly sufficient for an institution to exist, but there are other features that many institutions have. First, there can be behavior by role-holders that is not part of their role, but which is still part of the institution. All behavior performed by role-holders that promotes the institution’s results in a way partially causally enabled by the behavior of other role-holders is part of the institution, even if that behavior is not copied or even repeated. In other words, the behavior that does cluster into roles determines what the institution’s results are, and then other behavior by role-holders that promotes the same results in a way partially enabled by others in the institution is also part of the institution. For instance, if a bank’s CEO takes an unusual, one-time action promoting the bank’s interests, that behavior may still be part of the institution, even though it is not part of the CEO role.

Many institutions also include what I will call secondary roles. The primary roles are the ones we have already discussed, defined by role-holders’ behavior. In addition to this, entities (singular objects, relations, kinds of objects, and kinds of relations) can come to have secondary roles in an institution by being systematically impacted by the behavior that enacts the primary roles, without themselves engaging in behavior that would make them a holder of a primary role. For example, a particular building takes on the role of county courthouse for a particular county because of the court activities that occur within it. Similarly, a kind of object, such as balls manufactured at a particular factory, might have the role of official game ball within a particular athletic institution. For an example involving relations rather than objects, consider the secondary role of foul in an athletic institution, where several different relations (such as grabbing someone’s clothing, tripping them, or forcefully bumping them) are all treated as members of a kind. Institutions are not required to include secondary roles, so within the BTI itself, ‘role’ means just primary roles.

With the BTI’s main features clarified, I now want to say more about the metaphysical details. There are several important questions here: What is the precise metaphysical relation that generates social institutions? What is the metaphysical status of roles? How do institutions persist through time? Where are institutions located? And, how do institutional facts fit into the picture?

The next step is to confirm that the BTI deals adequately with Racist Courts, Disputed Government, and Watchdog. In Racist Courts, behavior clusters into roles, such as attorney and judge. Occupants of those roles copy each other, and the roles have been vacated and passed on. All of the roles together inter-enable each other to promote certain results, including over-incarceration of a certain racialized group, that could have been promoted in other ways. The injustice against the over-incarcerated group is one of the institution’s results—a defining aspect of its nature—regardless of anyone’s intentions or beliefs. So far, so good. In Disputed Government, there is behavior clustering into roles that have been copied and passed on, leading to inter-enabled promotion of certain results that could have been promoted in other ways, and on the basis of that behavior, Candidate A has the leader role and Candidate B does not. The government has one leader, despite all of the disagreement. And in Watchdog, there is no watchdog institution because the requisite behavior is missing, regardless of people’s beliefs.

Let’s also consider how the BTI applies to a straightforward, paradigmatic example of an institution: Harvard University. Harvard promotes the results of knowledge generation and transmission. At Harvard, there is the role of student, which involves attending classes and submitting assignments, among other things. Occupants of the student role abstain from other behavior that promotes knowledge transmission, such as creating and grading assignments. The instructor role, on the other hand, involves engaging in some of the result-promoting behavior from which students abstain, and abstaining from some of the result-promoting behavior in which students engage. There are many other roles, such as administrative roles, alumni roles, the president role, various food service staff roles, and various facilities staff roles. Importantly (for condition 1), these roles are all copied: students are influenced by other current or past students, instructors influence each other, and so on. And, as condition 2 requires, these roles have certainly been vacated and passed on in Harvard’s history. Finally, in line with condition 3, there are equally accessible alternatives (such as an entirely different leadership structure for the university), and the roles promote Harvard’s results in an inter-enabling way. That is, the actions of faculty, students, and staff enhance each other’s ability to promote knowledge generation and transmission.19

I’ll discuss two main ways in which the BTI differs from other accounts of social institutions: it eschews mental phenomena, and it precludes inconsistency.

First, the avoidance of mental phenomena. It should be plain at this point how deeply the BTI differs from collective acceptance accounts such as Searle’s (1995, 2009), Tuomela’s (2002, 2007, 2013), Thomasson’s (2003), and Ludwig’s (2017).20 It differs just as starkly from accounts that appeal to mental states other than collective acceptance, such as Miller’s (2001, 2003), Guala’s (2016), Torrengo’s (2017), and Hindriks’s (2023).21 It is worth mentioning that Guala (2014, p. 65) and Miller (2003, pp. 241–242) agree with my emphasis on what actually happens in institutions, as opposed to what we collectively accept as true about our institutions. But they still go on to offer accounts of social institutions that, unlike the BTI, appeal to mental phenomena. In eschewing mental states, the BTI also differs importantly from Epstein’s (2014) pluralist approach to social institutions. According to Epstein’s approach, institutional facts can obtain in ways not metaphysically dependent on mental phenomena, but mental phenomena of various kinds can give rise to institutional facts.

By eschewing mental states in the BTI, I am not eschewing mental states in the broader explanatory story of the behavior that gives rise to social institutions. Intentions (conscious or otherwise) are often present, guiding the behavior of primary role-holders in institutions. My point is just that these intentions and other mental states do not metaphysically determine the institution’s existence and nature.22 They are part of the causal story of why the institution exists, and not part of the metaphysical story of what generates the institution. This means that, in principle, non-conscious machines could have a social institution, provided that they were sophisticated enough to satisfy the BTI (including having the ability to copy each other). But I also want to acknowledge that for the purposes of certain conversations within political philosophy or the social sciences, we might be concerned only with the kinds of institutions that do have intentions and other mental states underlying the behavior (a criterion that, I assume, most actual institutions satisfy). The BTI is not intended to undermine these conversations, but just to show that when it comes to the project of doing the metaphysics of social institutions—that is, the project of understanding what makes it the case that they exist—those mental phenomena are not relevant.

The second feature of the BTI that I’d like to highlight is its preclusion of inconsistency in institutional reality. Brouwer (2022, pp. 24, 30) argues persuasively that Epstein’s approach to social ontology, as well as Searle’s and Tuomela’s, implies the possibility of inconsistent institutional facts—that is, of situations in which some fact A and another fact not-A obtain simultaneously, within a single institution. I won’t delve into the details of how inconsistency can arise for other theorists, but one key feature that allows it to arise for each of them is that they hold that institutional facts can be generated one by one (p. 38).

On the other hand, according to the BTI, institutional facts about which individuals hold which roles are determined by all of the behavior within an institution across a given span of time, taken together. That behavior as a whole will either cluster into roles that work together to promote some result(s), or it will not. And any given individual will either satisfy the criteria for being a primary role-holder within a given institution, or they will not. That is: they will either behave in ways that link into the web of behavior by role-holders in the institution that is interconnected by REB relations and inter-enabled promotion of the institution’s result(s), with the right history of copying, or they will not. It is not possible for someone to simultaneously satisfy, and not satisfy, these requirements with respect to the same role in a given institution. Similarly, a given entity that does not hold a primary role in a given institution either will or will not be systematically impacted by the behavior of primary role-holders, and thus either will or will not have a secondary role.

Now, there may be genuinely murky cases, such as when an entity is just beginning to manifest a passive behavioral profile and take on a secondary role. These may be cases of indeterminacy, but they certainly aren’t cases in which someone or something both has, and does not have, a given role.23

One further question that may arise about how the BTI compares to other accounts of institutions is the following: what kind of normativity, if any, do institutions have, according to the BTI? It is common for authors writing on social institutions to build some kind of inherent normativity into their account (Miller, 2003, p. 243; Searle, 2009, p. 91; Tuomela, 2013, pp. 227, 43; Guala, 2016, Ch. 6). The BTI does not explicitly build in normativity, but the question of whether subtler normativity may nonetheless be present is a complex one that I will leave for another occasion. Whether institutions have inherent normativity or not, I want to emphasize that we can (and should) normatively evaluate them. Just as a powerful machine someone creates can be normatively evaluated despite not itself being inherently normative, so could normatively inert institutions be evaluated with respect to independent moral considerations.

We began with the idea that social institutions’ existence and nature are “up to us” in a deep, important way. Institutions really are “up to us” in that their existence and nature are determined by and dependent on our behavior, and also in that each of us can control our behavior. But still, there are two important ways in which institutions are not “up to us.” First, their existence is not determined by or dependent on what we believe, decide, or agree upon. Second, their nature is determined by the results that our behavior actually promotes and not by results at which we intentionally aim. Social institutions, though deeply dependent on our behavior, can slip out of our grip. The right account of social institutions must, as I hope the BTI does, reckon fully with this reality.