3.1. Level 1: Proto-institutional systems

Much research into the evolution of social behavior suggests that social coordination problems can be readily solved through what we will call proto-institutional systems: emergent systems of organization, such as kinship relations and biologically determined divisions of labor, in which different individuals carry out different tasks. Importantly, while this might look role-like from the outside, it can occur without anyone in the system representing it in terms of roles (Figure 2a). In these systems, each member knows a set of rules and regularities that determine how to behave within a group and how to interact with others. These simple social structures can solve social interaction problems that would otherwise require mentalizing. For instance, a simple linear dominance hierarchy might encode a social affordance as basic as: in the presence of a conspecific, the higher-ranking member can do as they please while ignoring subordinates; subordinates must get out of the way and not interfere with the dominant conspecific. Such social structures enable creatures to co-exist in a group without access to each other’s minds as long as they know how to behave within the social system. These representations are consistent with the notion of “minimal cooperation” (Maibom, Reference Maibom2007; Ritchie, Reference Ritchie and Fiebich2020b), in which individuals playing roles in organized structures can, by simply doing their part, achieve relatively complex coordination as long as others do the same.

Figure 2.

(a) A fixed proto-institutional structure where a group of agents act in coordination, showing here a dominance hierarchy. This type of implementation does not require users to differentiate roles from role-occupiers, or even to represent positions in the structure as roles at all. (b) Role-based structure where a group has an abstract representation of the social structure, and a mapping from agents to roles. The ability to represent social roles as detached from individuals enables members to consider alternate arrangements such as role switches. (c) Group with multiple role-based representations to structure a more complex social life. Here, the dominance representation is a linear structure that can be expanded or compressed based on group size; the caregiver/kin structure represents a system which includes expectations for how agents in the same role act towards one another (e.g. mutual expectations among siblings or among caregivers) as well as how agents in different roles act towards one another (e.g. expectations of care and deference). The hunting structure represents the prey (Rp) as a role in the system, despite the prey not being aware of this, as well as relationships between different hunter roles. As social complexity increases, conspecifics must represent a larger number of role-based systems, and consider how they might combine productively (e.g., how caregiver/kin relations affect conflict in or have critical entailments within dominance hierarchies). The increased number of institutional structures and the need to reason about them compositionally may have led to the generative institutional grammar in humans, which enables us to rapidly synthesize and infer institutional structures.

Figure 2. Long description

Panel A: This panel shows three characters each represented as a circle with two black dots as their eyes. The three characters have a single thought bubble above them representing a shared thought. Inside the thought bubble are the same three characters connected by arrows, indicating a linear hierarchy. Panel B: This panel shows the same three characters from the first panel with a shared throught bubble divided vertically. The left side of the thought bubble shows a linear hierarchy represented as a sequence of three rectangles labeled R1, R2, and R3, each representing an abstract role. Next to each role rectangle is one of the characters with an arrow pointed towards the role. This represents a distinction between the roles, and the characters occupying the roles. The right side of the thought bubble shows the same linear hierarchy with three rectangles labeled R1, R2, and R3. The same characters appear to the right of the roles, but each arrow connects each character to a different role, representing the idea that different characters can occupy different positions in the hierarchy. Panel C: This panel shows the same three characters with a shared thought bubble. The thought bubble is divided into three sections labeled Dominance, Hunting, and Caregivers/kin. The Dominance section shows a horizontal sequence of rectangles labeled R1 to Rn, indicating an abstract linear hierarchy. The Hunting section shows three R1, R2, and R3 connected by arrows, which represent the character roles, and a fourth gray rectangle labeled Rp, representing the role of prey, with all other role rectangles pointing at it. The Caregivers/kin section shows a bidirectional arrow between rectangles R1 and R2.

These basic forms of social organization appear to be recurrent solutions to coordination problems (Sumpter, Reference Sumpter2010). They may have emerged many times independently across multiple social species, such as kinship relations, where conspecifics develop expectations about who will react and care for an infant in need (Dasser, Reference Dasser1988; Cheney & Seyfarth, Reference Cheney and Seyfarth1980); primitive forms of in-group/out-group representations (Lorenz, Reference Lorenz2021); or forms of coordinated hunting (such as with Aplomado falcons, where males and females perform different but fixed tasks; Hector, Reference Hector1986). One of the best-known examples is the division of labor between workers and soldiers in ants (Hölldobler & Wilson, Reference Hölldobler and Wilson1990), who act within biologically fixed social structures with clear roles that provide different social affordances. This is particularly apparent in ants in the case of dulosis – the process where parasitic ant species capture worker ants from a different species, which then continue to carry out their duties, now in the service of the parasitic colonizers (D’Ettorre & Heinze, Reference D’Ettorre and Heinze2001).

We call these systems proto-institutional because members of the social group do not, or at least need not, have any representation of roles or of the social structure itself. That is, the role and role-occupier are not differentiated: associating different agents with different social affordances is enough (known as a lack of role-filler independence; Figure 2a). Despite this, knowing what behaviors are associated with each agent still supports some forms of inference. For instance, the types of transitive inferences in a dominance hierarchy (e.g., those seen among pinyon jays; Paz-y-Miño et al., Reference Paz-y-Miño C, Bond, Kamil and Balda2004) can be instantiated by simply representing each conspecific as having a particular amount of power. The cognitive prerequisites of a proto-institutional system are also not trivial; an agent must know what social affordances each individual has, including themselves. This can be achieved in a variety of ways, including individual identification (e.g., recognizing your offspring) and consistent visual markers (e.g., in European paper wasps; Tibbetts & Lindsay, Reference Tibbetts and Lindsay2008). These behaviors might also require some basic form of action understanding (at a minimum, a capacity to process body movements and make simple predictions such as direction of motion, but they could also be grounded in richer representations such as a teleological stance; Gergely & Csibra, Reference Gergely and Csibra2003). This suggests diversity in how proto-institutional systems work, but the critical point is that these systems do not yet require a flexible or complex representational understanding of roles or the social structures they are embedded within.

3.2. Level 2: Fixed institutional representations with role-filler independence

As social organization becomes more complex, groups benefit from the ability of conspecifics to take on different jobs depending on the context and the current group’s needs. This is the origin of role-based systems that allow members of a group to swap roles (Figure 2b)Footnote ¹ . The increased flexibility of such systems also imposes new representational and inferential demands via role-filler independence: Members must have an abstract representation of roles that is independent of the members occupying them; they must be able to infer which role a conspecific is occupying based on their appearance or behavior; they must track and remember role assignments and role changes throughout time; and they must know how to adjust their behavior as a function of the role that they themselves, as well as their interaction partners, occupy.

A candidate example is male dominance hierarchies in monkeys, which are dynamic and continuously renegotiated. Indeed, some primates, like baboons, are thought to have specialized intelligence for tracking and remembering rapid and complex role changes (Cheney & Seyfarth, Reference Cheney and Seyfarth2019; Bergman et al., Reference Bergman, Beehner, Cheney and Seyfarth2003). The variety of social behaviors non-human primates engage in also increases the demands for context-specific systems of organization, and it is possible that they have additional institutional representations such as troop patrols or even multi-role hunting structures (Boesch, Reference Boesch1994, Reference Boesch2005; although such interpretation is disputed; see, e.g., Moll & Tomasello, Reference Moll and Tomasello2007).

These types of abstract institutional representations may be present beyond primates, suggesting they evolved independently multiple times or emerged a long time ago in our evolutionary history. In desert lions, for instance, female lions hunt through a coordinated structure that consists of two center hunters and two wings (Stander, Reference Stander1992). While individual lions have preferred roles that correlate with successful hunts, they can nonetheless switch between roles, suggesting they represent the different roles and understand that all roles must be filled for the hunt to succeed. Naked mole rats also show role specialization (e.g., between defense, working to sustain the burrow, and caring for young; Faulkes & Bennett, Reference Faulkes, Bennett, Buffenstein, Park and Holmes2021), and they switch to less preferred roles both when a preferred role is occupied (Gilbert, Rossiter, & Faulkes Reference Gilbert, Rossiter and Faulkes2020) or when the animals previously fulfilling a role are removed (Mooney et al., Reference Mooney, Filice, Douglas and Holmes2015). Another striking example comes from honeybees, which flexibly shift from foraging to receiving nectar. When returning foragers can’t find a bee to receive nectar, they perform a tremble dance (distinct from the more famous waggle dance that indicates the source of nectar); observing foragers sometimes then switch roles to become receivers, maintaining an optimal balance between these two activities (Seeley, Reference Seeley1997).

These representations reflect the most basic capacity to build a mental model of an institutional structure – i.e., a set of interconnected roles occupied by and constraining the behavior of a set of agents. But they do not yet constitute a full institutional stance, for two reasons. First, conspecifics only apply role representations to the actual roles that exist, rather than employing them to interpret any form of collective behavior, where roles can be a useful fiction through which to make sense of what is happening. Second, the set of institutional structures that conspecifics can represent is predetermined (Figure 2c). There is no evidence of variability or cultural innovation across groups, even when they live in ecologically very different circumstances (di Fiore & Rendall, Reference Di Fiore and Rendall1994). However, within these bounds, these representations are already minimally generative, in that the exact number of roles is mutable, e.g., a dominance hierarchy can expand and contract based on group size.

3.3. Level 3: Generative institutional representations

Species that evolved fixed institutional representations have internally generative models of the social organizations they are part of (e.g., representing hierarchies of arbitrary size or multiple friend cliques). But as social life becomes more complex, social species have a growing repertoire of increasingly complex social organizations that govern behavior in different circumstances. Chimpanzees, for example, have separate male and female dominance hierarchies, friendship coalitions, caring relations, and troop patrols. At this point, we hypothesize that three pressures come together to create a fully-realized institutional stance. First, the growing repertoire of social organizations creates a pressure to efficiently place them in a common representational space. Second, the complexity of social life creates a pressure to reason about how different social systems get combined (e.g., how to predict the behavior of a conspecific who is simultaneously dominant in the hierarchy but also a family member; Bergman et al., Reference Bergman, Beehner, Cheney and Seyfarth2003). Finally, the separation of role and role occupier allows increasing recognition that some social arrangements are more beneficial than others, entailing new advantages if one has the capacity to change those arrangements.

This leads to the full-fledged institutional stance, which we propose is uniquely human. There are two key differences in this final implementation. The first is that the representations over institutions are fully generative: they no longer rely on a fixed set of evolutionarily ancient role-based systems. Instead, we have a generative representational system to synthesize novel institutional structures, leading to an ability to understand any kind of role-based system, including ones we have never encountered in evolutionary history, from reading clubs and sports teams to university committees and governments. And, as we explain in Section 6.4, this interpretative framework becomes a system that is itself reused for creating new institutional structures, leading to an explosion of role-based systems in human society. This implies that we are equipped with a grammar for generating institutional structures, which can be built over complex representations of roles. In other words, we have an intuitive Theory of Institutions, akin to our Theory of Mind. The second key difference is that this ability to create arbitrary social structures allows us to apply this form of reasoning to any social system, making it a full-blown stance. This means that we can use it not only to reason about established systems of social organization, but also as a way to interpret social collectives even when we don’t necessarily think the members think of themselves as implementing an institutional structure with roles (e.g., interpreting sibling dynamics or friend groups in terms of roles).

This is the institutional stance, which enables us to (1) understand complex social behavior in terms of structured systems of interconnected roles, (2) compositionally construct new such systems, and (3) use that knowledge to interact with others. Indeed, the ethologists’ ability to watch animal behavior and infer the complex latent institutional representations at play is itself the institutional stance in use. Even if the cases we reviewed above turned out not to be precursors to the institutional stance, the propensity of humans to interpret animal behavior in this way illustrates our tendency to read complex social behavior through latent representations of roles and role relations. We now formalize this proposal, review supporting evidence, and discuss its implications.

4.1. What is an institution?

When deploying the institutional stance, an institution is a type of cognitive representation consisting of a set of interconnected roles with a level of persistence and continuity. These representations don’t need to be explicitly codified or mapped onto formal institutions, and they can also represent evolutionarily ancient social interactions like caring relations, as well as informal relations like friendships. Specifically, an institution can be thought of as a representation of a graph, where each node represents a role and each edge represents a relation.

Representations of roles are defined by the content of the role (i.e., how does one behave in that role?) and the conditions through which an agent can occupy or leave it. While the specifications of role content are likely flexible and can be culturally modified, most roles have three properties. First, they encode tasks or functions (i.e., the everyday meaning of role) that the role occupier prototypically enacts. Second, they encode socially granted powers that allow particular affordances specific to the role. Third, they impose limitations on how one can wield the socially granted power and how they complete their tasks or function. These components typically reference other roles (i.e., they are graph edges linking the role to other roles). For example, a barista has the task of fulfilling customer orders, the power to charge their credit cards, and the constraint that they must be friendly. But in some cases, they can also be standalone role features. For instance, we can represent a janitor without referencing other roles, as charged with keeping a building clean, and with access to all rooms after 5 pm. Critically, roles do not always need to have all three components. We can represent roles that don’t encode any task (e.g., care recipient), roles that don’t grant any power (e.g., lowest ranking member in a hierarchy), and roles that don’t impose any limitations (e.g., authoritarian dictator). However, at least one of these features must be present. A role with no purpose, power, or limitations has no behavioral or interpretative consequences.

In some cases, occupying a role is accomplished by simply enacting it, and one can leave a role by simply ceasing to behave as expected (e.g., one stops being a friend by no longer enacting the role of a friend). In other cases, entry conditions are more formalized, such as needing to be approved by the occupant of another role. Thus, entry and departure conditions frequently reference other roles that have the power to grant or coerce agents into other roles, as well as to revoke them. But the powers and limitations generated by the role are contingent on actively occupying it; for example, the barista can be a customer while off duty.

Critically, an institutional representation does not need to be a complete representation of the graph’s structure. More often than not, representing a subset of the graph is enough. For instance, our naive representation of universities only references a small set of roles that are most relevant to us (e.g., undergraduates, graduate students, department chairs), without representing the entire complex structure of actual roles. Each representation of a role can also be simplified relative to the true nature of a role, such as an undergraduate representing a professor role in terms of only teaching obligations. However, even when reasoning about a single role (e.g., the janitor), we are aware that the role is situated within a broader system of roles. This knowledge helps us navigate the institution effectively: e.g., if we wanted to change the time of day in which the janitor enters our laboratory, we might speak with them directly, but we might also consider whether the right intervention involves a discussion with someone else, say an office manager. Rather remarkably, we can do this even if we had never previously thought about the janitor’s position in the institutional structure at all; this is a hint as to the generative nature of institutional representations, something we return to below.

4.2. Intuitive institutions, theories of institutions, and real institutions: What’s the connection?

The building blocks of our framework are similar to those used in formal theories of institutions developed in social sciences and philosophy (e.g., Crawford & Ostrom, Reference Crawford and Ostrom1995; Hodgson, Reference Hodgson2006). We believe this is no coincidence. At the beginning of any formal endeavor, theoreticians have no other way to evaluate the plausibility of a proposal other than comparing it against their intuitions. Inevitably, then, seminal formalizations of different phenomena often share crucial similarities to our intuitive theories. To give just two examples, human intuitive physics is surprisingly similar to the first formalizations of physical laws (Battaglia, Hamrick, & Tenenbaum Reference Battaglia, Hamrick and Tenenbaum2013) and, within mentalizing, intuitions about how others make decisions are also strikingly similar to principles of classical economics (Jara-Ettinger et al., Reference Jara-Ettinger, Gweon, Schulz and Tenenbaum2016). The origins of a formal inquiry into social ontology and theories of institutions may be no exception. This suggests that theories of formal institutions, particularly within social ontology, could hold valuable clues as to the nature of our intuitive theories of institutions (e.g., Guala, Reference Guala2016; Crawford & Ostrom, Reference Crawford and Ostrom1995; Searle, Reference Searle2005).

Further, adopting the institutional stance allows for analyzing social relationships in terms of institutional representations even when those representations do not clearly map onto the external institutional structure. This can occur when the external structure may or may not even be institutional (as in the friend group example mentioned above) as well as when the useful institutional representation departs from the formal institutional (e.g. when reasoning about a cafe, it is pragmatically useful to consider the role of customer in relation to the role of barista, even though the customer is not part of the formal institutional structure at all). This highlights the distinction between our focus on the institutional stance as an intuitive theory of person perception and the analysis of formal institutions that other social scientists might undertake.

5.1. Representing collectives

The idea that people not only represent other people’s minds, but also represent social collectives has a long history in cognitive science and philosophy, grounded in the idea that humans have a folk sociology – an intuitive theory of how people organize around notions of groups that help us identify social types, evaluate them, and develop shared expectations for how group members behave (Rhodes & Baron, Reference Rhodes and Baron2019; Hirschfeld, Reference Hirschfeld, Banaji and Gelman2013; Clément, Bernard, & Kaufmann Reference Clément, Bernard and Kaufmann2011; Ritchie, Reference Ritchie2020a; Shutts & Kalish, Reference Shutts and Kalish2021).

Work in folk sociology has revealed that, from early childhood, people have a natural tendency to categorize people. This is most prominently in our tendency to divide the social world into “us” versus “them” (and positively evaluate the “us” in myriad ways; Dunham, Reference Dunham2018). But it also includes a large set of group categories that are essentialized – i.e., thought to be determined by patterns of deep internal, perhaps genetic or biological, similarity (Gelman, Reference Gelman2004; Rhodes & Mandalaywala, Reference Rhodes and Mandalaywala2017), like race and gender (e.g., Prentice & Miller, Reference Prentice and Miller2007) – such that these categories are represented as stable and immutable. These types of social representations in folk sociology are not part of the institutional stance, in that they are intuitively represented as “natural” categories intrinsically embedded in and stable to the person, rather than represented as interconnected roles that one can pass into and out of. At the same time, recent work in folk sociology has noted that not all social categories are essentialized. Many social groups are better defined via deontic or normative properties (i.e., via rights and obligations assigned to members; e.g., Foster-Hanson & Rhodes, Reference Foster-Hanson and Rhodes2019; Kalish & Lawson, Reference Kalish and Lawson2008). Moreover, both children and adults believe that social and cultural forces can produce meaningful social kinds (Noyes & Keil, Reference Noyes and Keil2020), via social agreement (Noyes & Dunham, Reference Noyes and Dunham2017) and formal membership criteria (Noyes & Dunham, Reference Noyes and Dunham2020). Even the most frequently essentialized social categories such as race and gender can be reconceptualized in a role-like manner; indeed, we believe this is precisely what social ontologists have done in arguing that such categories are actually structured representations of hierarchical roles, such that to be conceived of as a woman or racial minority is to occupy a subordinate position (Ásta, Reference Ásta2018; Haslanger, Reference Haslanger1995, Reference Haslanger2000).

We believe that these second kinds of social categories emerge from the institutional stance, and they should be, therefore, understood as reflecting a separate cognitive system that is not supported by essentialized representations (Noyes et al., Reference Noyes, Dunham, Keil and Ritchie2021). Thus, while we agree with this recent work rejecting essentialism as a universal representation of social categories, we diverge from it in two ways. First, past work has suggested that these socially constructed categories are not a primitive type of representation in the human mind, and that they emerge rather late in development (approximately by age seven; Noyes, Keil, & Dunham, Reference Noyes, Keil and Dunham2018, 2020), whereas we believe that institutional structures are evolutionarily ancient and readily available in infancy (although some components involve maturation and/or learning; see Section 6.1). Second, much of this past work has studied social groups as undifferentiated containers of group members in which exchangeable individuals inherit properties (whether traits and behaviors or rights and obligations) directly from the group’s attributes. In our account, these representations have rich internal structure specifying how roles are related to others in specific ways (see also Noyes et al., Reference Noyes, Dunham, Keil and Ritchie2021; Ritchie, Reference Ritchie2020a). In this way, the institutional stance aligns with recent work taking seriously the structured relationships between groups and the distinct patterns of reasoning that emerge from them (Amemiya et al., Reference Amemiya, Mortenson, Heyman and Walker2022; Vasilyeva & Lombrozo, Reference Vasilyeva and Lombrozo2020; Vasilyeva, Gopnik, & Lombrozo Reference Vasilyeva, Gopnik and Lombrozo2018; Yang, Naas, & Dunham Reference Yang, Naas and Dunham2021).

5.2. Representing socially structured behavior

In our account, institutional representations are grounded in representations of roles, each with membership conditions and content. While we are relatively agnostic about the exact internal representational nature of these components, norms are a promising candidate representation that could encode the duties, power, and limitations in a role. It is thus worth considering whether this view, in its strongest form, means that we don’t even need to appeal to roles, and our proposal can be reduced to the idea that people encode agent-relative norms that affect behavior. In what way does the institutional stance go beyond this idea?

Normative systems are critical for social coordination in three ways. First, norms reduce coordination costs by making behavior predictable; if I know what you are likely to do I can easily adapt my behavior to be complementary (and you can do the same). Second, since norm violations undermine cooperation, violators may face reputational costs or retaliation. Knowing this incentivizes compliance and produces a “sense of should,” i.e., an internalized belief that one should comply (Theriault, Young, & Barrett Reference Theriault, Young and Barrett2021). Third, given potential costs associated with negotiating ad-hoc norms on the fly, norms tend to stabilize within groups, allowing coordination to proceed more easily.

In particular, some recent work has suggested that some integration of folk psychology with norms can provide a powerful framework for reasoning about others (e.g., Eickers, Reference Eickers2024; McGeer, Reference McGeer2021, Andrews, Spaulding, & Westra Reference Andrews, Spaulding and Westra2021). Our perspective also asserts that norms are critical to maintaining cooperative social life and extends this claim in two ways. First, even if the institutional stance is composed from basic representation of norms, the way in which norms are structured is not trivial, requiring mechanisms for representing norms as bundled into roles and related to other roles in normatively laden ways. Bundling norms into roles requires the machinery we elaborated on above, i.e., specifying membership conditions, limitations, and/or socially granted powers. Further, the cognitive machinery necessary to structure and productively build complex social structures over roles is not trivial, and is the key to turning the social world’s vast set of norms into a useful interpretive system.

The second and related point of departure is that, in grounding the institutional stance in norms, it suggests the need for a special type of institutional norm that has largely been neglected in the literature. Classical accounts divide norms into two types: injunctive (enforced through social pressure, e.g., being quiet in a library) and descriptive (emergent behavioral patterns without enforcement; e.g., clapping after a play (Kallgren, Reno, & Cialdini Reference Kallgren, Reno and Cialdini2000; Lapinski & Rimal, Reference Lapinski and Rimal2005). Institutional norms have three features that do not fit this dichotomy: (1) they are role-dependent, with their scope defined by a role and its position in the underlying institutional structure (rather than scope being defined by context, statistical prevalence, or degree of social enforcement); (2) adherence to a norm is represented as a causal consequence of occupying a role (i.e., we reason that norm adherence can be explained by the role, a la formal explanation), rather than a merely descriptive pattern or a prudential one generated by fear of retaliation; and (3) failing to adhere to a norm is often seen as evidence that the agent is no longer occupying a role, rather than as a punishable offense or as evidence that the norm is less descriptive of the population (e.g., spotting this morning’s barista sitting at a table at a coffee shop and eating lunch would not lead us to revise our beliefs about barista roles, and we wouldn’t expect them to be subject to punishment for non-compliance, nor would we walk up to them and ask for a coffee). Many norms within the institutional stance may well be special cases of descriptive or injunctive norms. But our point is that reasoning about when a norm becomes relevant or active in social settings requires a latent representation of the institutional structure, such that an account that simply enumerates injunctive and descriptive norms will be insufficient to produce the type of reasoning the institutional stance provides.

Moving up to the representation of a role as a unit, some related work has argued that these representations are instantiated by a capacity to share intentions (in psychology: Tomasello, et al., Reference Tomasello, Carpenter, Call, Behne and Moll2005; Tomasello & Carpenter, Reference Tomasello and Carpenter2007; in philosophy: Searle, Reference Searle2006; Gilbert, Reference Gilbert1992). Consider the task of collaboratively opening a cardboard box, in which one person holds the box while the other cuts it open (Tomasello et al., Reference Tomasello, Carpenter, Call, Behne and Moll2005). Each agent has a set of focal goals they must keep in mind (hold the box steady; cut the tape with the box cutter), but they must also keep in mind the goal of their partner, and adjust their behavior to comply with the higher and lower-level implementation their partner adopts. Once this entire structure is represented, role switching is unproblematic; the holder can become the cutter and vice versa.

Proposers of shared intentionality have argued that this capacity is unique to humans (Tomasello & Carpenter, Reference Tomasello and Carpenter2007; Tomasello & Herrmann, Reference Tomasello and Herrmann2010; Tomasello & Moll, Reference Tomasello and Moll2010) and that it is a cognitive prerequisite for role representations that produce cooperative behavior (Moll & Tomasello, Reference Moll and Tomasello2007; Tomasello, Reference Tomasello2022). Therefore, this view might imply that the institutional stance ultimately depends on mentalistic representations. We do not believe this is the case. Shared intentions are undoubtedly part of how people represent roles in the institutional stance, but a core claim of our perspective is that this needn’t, and often isn’t, the case. The institutional stance allows us to analyze any behavior as if it were governed by latent roles, even when the agents are performing role-like behavior without a representation of roles or joint intentionality. In this sense, the roles within the institutional stance are not Tomasellian roles (Tomasello, Reference Tomasello2020). Thus, we can think of emergent social coordination in terms of roles without committing ourselves to the view that the social creatures are engaging in joint intentionality.

5.3. Representing interactions

Finally, at the lowest level, the institutional stance represents social interactions. One prominent set of alternatives comes from schema and script approaches to social cognition (Bartlett, 1932; Eickers, Reference Eickers2023, Reference Eickers2024; Mandler, Reference Mandler2014; Taylor & Crocker, Reference Taylor and Crocker2022). Schemas refer to knowledge that structures and organizes a particular behavioral domain, which is sometimes analogized to a dramatic script. For example, a supermarket script might describe the types and sequences of activities that take place in a supermarket. This knowledge is organized via the roles that frequently occur in that setting as well as the activities that characterize those roles. For instance, it might specify the role of shopper in terms of activities like getting a shopping cart, gathering groceries, going to the registrar, paying, and leaving with groceries; a role of cashier in terms of things like scanning each item, placing items in bags, accepting payment, and so on. Critically, script approaches provide a domain general account of how information governing familiar domains of behavior is organized. One can in this sense have a script for solitary and non-social activities such as building a campfire.

In our view, script representations are a broad capacity that reflects general intelligence and can be applied to any domain. Indeed, these can be thought of as model-free representations of sequential events, social or not. The institutional representations proposed in our account go beyond simple sequences of how actions typically unfold and so are able to explain behavior not merely as the next line in a script, but rather as the causal consequence of institutional representations. This can be seen by considering the limitations of a social representation that has nothing but scripts. Such a representation would only be useful as long as events unfold as unexpected, but would be of little use when an unexpected event happens. By contrast, the institutional stance also provides guidance as to what to do when an expected social interaction breaks down. For instance, if a barista deviates from the expected behavior by being rude or unhelpful, we are not merely at a loss about what to do. Rather, because we represent the cashier as linked to other roles such as a manager, we can flexibly consider how to intervene (at least, if we decide it’s worth the time). Thus, while script-like representations are a basic form of efficient storage that high-level cognition makes use of in a variety of ways, they become dynamic and generative when embedded within a mental model of role-based behaviors and the relation between them.

Another important framework for thinking about social relationships arises out of the Relational Models Theory (Fiske, Reference Fiske1991), and a recent proposal that Fiske-style relational models are at the heart of an innate “core knowledge” system for thinking about relationships (Thomas, Reference Thomas2024) in which authority, communal sharing, and equality matching are the initial building blocks of more complex understandings of relationships. In our view, these relational models are simple institutional structures capturing a particular configuration of roles and relations. They emerge early and are culturally ubiquitous because the simplicity of their structure and role contents makes them useful tools for understanding real-world phenomena such as ingroup-outgroup structures and dominance hierarchies. Thus, rather than proposing a small number of innate relationship templates, we propose that even infants are able to flexibly construct all of these institutional models, and likely many more, via a particular type of generative grammar operating over roles.

6.1. How are institutional representations acquired?

6.2. When do we deploy the institutional stance?

Given that people can analyze behavior in terms of mental states, when do we choose to apply the institutional stance instead? This can be a challenging problem, because many behaviors can be explained either as the expression of a role or as a combination of an appropriate belief and desire.

At one extreme, it is possible that the institutional stance is pervasively and automatically operating in all or nearly all social settings. This would follow if its evolutionary history and centrality to social cognition run as deep as mentalizing, which has often been described as automatically deployed (Kampis & Southgate, Reference Kampis and Southgate2020). Even in the cases where we are richly mentalizing about others, latent role representations – such as “friend,” “co-worker,” or even “stranger” – might nonetheless provide a set of expectations that could greatly simplify social settings (e.g., they might help us assign traits that simplify mentalizing; Westra, Reference Westra2018, Reference Westra2020). With this in mind, return to our paper’s opening example, where Kate asks Eric to fulfill certain roles when they run into each other at the cinema. The surprise of the untoward request could come not only from knowing that the previous roles are no longer operative, but also because we know and are automatically tracking new roles – “acquaintance” or “friend” – that have very different entailments.

Interestingly, some evidence about the automaticity of these processes comes from classical work that is usually thought to showcase the primacy and automaticity of mentalizing. Consider Heider & Simmel’s (Reference Heider and Simmel1944) classic study showing that people spontaneously interpret the behavior of geometrical shapes as agentive. While some of those interpretations evoked mental state language, many of the original participants also used role-based language, e.g., talking about villains and heroes or parents and children. Moreover, the basic units of analysis often invoke primitive role relations like chasing and fighting, which are detected from low-level statistical movement cues that might be culture-independent (Barrett et al., Reference Barrett, Todd, Miller and Blythe2005). Some more recent evidence has also found that, in self-paced reading tasks, the mention of a role triggers immediate expectations about how the agent is likely to act, whether they are replaceable (i.e., a role can be replaced by another agent), and even what kind of knowledge they might possess (Baker, Dunham, & Jara-Ettinger Reference Baker, Dunham, Jara-Ettinger, Samuelson, Frank, Toneva, Mackey and Hazeltine2024). This suggests that some aspects of the institutional stance might be pervasive and automatic.

The breadth of the institutional stance can also be illustrated by cases in which agents take action to encourage others to apply the stance to them. This is particularly the case when their behavior is the result of a role and where it is critical for observers to recognize that fact (e.g., police officers directing traffic). In other words, to ensure that our own behavior is appropriately interpreted through the institutional stance when relevant, we broadcast the roles we occupy to help resolve potential ambiguity. The efforts to do so have been dubbed “status indicators” (Searle, Reference Searle2006). Visible markers like uniforms – which are pervasive in the social world, such as those worn by police officers, priests, lifeguards, doctors, and baristas – are notable examples. Consider Figure 3, which shows two identical agents taking identical actions in identical contexts. The simple presence of an apron leads to a radical revision of how we interpret the behavior and mental states of the agent on the right. In the first case, we might infer that the person likes olive oil, intends to get some, and is likely to cook a meal with it that week. In the second case, the same behavior no longer reveals any personal attitude towards olive oil, and we instead read the behavior as revealing a role-based task, such as ensuring that the oil has not yet passed the sell-by date.

Figure 3.

Two identical agents taking identical actions in the same context. The single presence of an apron invites observers to take an institutional stance that triggers different inferences, predictions, and expectations. Rather than reading the behavior as revealing an agent’s desire for olive oil, we are more likely to construe it as revealing a role obligation, such as restocking or checking the sell-by date.

In related cases, agents signal their institutional status in other ways, most notably verbally. Emails may begin with a disclosure such as “I’m writing in my capacity as Chair of the Search Committee” to indicate that the action one is taking (perhaps rejecting a candidate) should be attributed to the carrying out of an institutional role rather than the agent’s personal beliefs and desires. In some such cases, speakers even explicitly disavow the mental states that might normally be inferred from the behavior (“it gives me no pleasure to have to write this letter…”), a form of speech that can be interpreted as urging the recipient to consider their behavior the enactment of a role demand rather than a personal desire. These institutional disclosures are even deployed in more informal roles: consider someone in a heated debate declaring that they are simply “playing the devil’s advocate.”

Beyond physical institutional indicators (e.g., uniforms) and linguistic constructions (e.g., explicitly disclosing an institutional role), markers of institutionality might sometimes even be embedded in language itself. For instance, many languages use different pronouns to encode levels of formality (Brown & Gilman, Reference Brown, Gilman and Fishman1960; Giglioli, Reference Giglioli1972) that coincide with institutional roles. Examples of this include the T-V distinction in Indo-European languages (e.g., tú/usted in Spanish), honorifics in Japanese in which relative status is encoded in verb endings (McCready, Reference McCready2019), or Nahuatl honorifics (a four-level morphological system; Hill & Hill, Reference Hill and Hill1978). It is possible that these linguistic markers are used not only for deferential purposes but to signal when an engagement ought to be read through an institutional stance rather than a mentalistic one.

Despite such efforts to encourage or discourage the adoption of the institutional stance, there will remain many cases in which people must determine whether or not to adopt it in the absence of clear status indicators. In these cases, the ability to detect automatic behavior can help us detect scripted behavior that could be caused by a role (Ulman & Bass, Reference Ullman and Bass2024). Knowledge of the institutional structure that one is operating within also likely helps us infer the relevant roles. Imagine, for instance, walking into a coffee shop and having a person in plain clothes greet you and ask you what you want. In this case, the low likelihood that a stranger would be motivated to spontaneously inquire about your general desires, combined with our knowledge of the roles typically found in coffee shops and our awareness that they are frequently not visibly marked, is enough for us to identify a role agent and so adopt the institutional stance with respect to them. In cases like these, it may be that the institutional structure is assumed to be well known, that leads to a decreased use of institutional markers. While these strategies might be particularly powerful for adults, they might also be available to children in some capacity. For instance, young children have expectations about what types of costs are reasonable to incur for others (e.g., ask for help only when the helper can accomplish the goal more easily than you could alone; Grosse, Moll, & Tomasello Reference Grosse, Moll and Tomasello2010; Jara-Ettinger et al., Reference Jara-Ettinger, Schulz and Tenenbaum2020), and violations of such expectations might help reveal roles and the tasks they perform.

Nonetheless, the strategies described above, both identifying institutional indicators and inferring roles from knowledge of institutional settings even in the absence of those indicators, are more likely to work for enculturated adults with rich knowledge of the institutions they encounter. It cannot account for actions within highly unfamiliar or entirely novel institutional structures (a domain which, even if modest in adulthood, will be vast earlier in development). This makes the mechanisms for inferring the hidden institutional structure reviewed in 6.1 critical, a process likely boosted by explicit cultural transmission concerning institutions and their central roles.

6.3. Interplay with mentalizing

6.4. Social constitution

We have argued that the institutional stance is a tool for reasoning about social interactions across a wide range of settings, from the evolutionarily ancient (e.g., hierarchies), through emergent forms of coordination where users don’t think of themselves as performing roles (e.g., a family or group of friends), to formally codified institutions (e.g., universities). In doing so, we have emphasized that the institutional stance is a generative system that, at least in humans, allows us to represent and manipulate arbitrary institutional structures. This ability to treat institutional representations as objects of thought helps explain two uniquely human aspects of social intelligence.

First, the capacity to read complex multi-agent behavior through the institutional stance means that we can apply this capacity as an analytic tool, including to analyze phenomena that we might not initially think of as being institutional. This is an important engine for social change that enables us to reveal hidden role-based systems that were previously unobservable to us or that were initially conceptualized in other ways. For example, the folk psychological conception of race, gender, and caste-based systems is often essentialized, i.e., thought of as natural kinds. Yet many theorists now analyze them in role-based structural terms arguing that they are hierarchical systems of interlocking social roles in which, for example, woman as a structural position defined in subordination to man (Haslanger, Reference Haslanger2000; Rehbein & Söyler, Reference Rehbein, Söyler, Baumann and Bultmann2020). Analyzing complex patterns of social behavior in this way can reveal hidden structures, allowing us to then evaluate them and consider how to modify or intentionally subvert them (Hesni, Reference Hesni, Jones, Schroeter and Schroeterin press). This gives us the ability to discover systems of role-based behavior that are hiding in plain sight, reason about what kinds of consequences the system produces, and, if we determine it to be unfair or oppressive, consider how to improve or erase it. In other words, the generative aspect of human institutional reasoning allows us to imagine and pursue genuine institutional alternatives.

Second, the generative capacity to create arbitrary institutional representations to understand complex forms of social behavior gives a uniquely human capacity, namely the ability to use the institutional stance to actively invent new institutions. Simply put, if we can imagine a new type of role-based social organization and communicate it to others, people may decide to actually enact it, thereby making it real. In this sense, the generative capacity of the institutional stance catalyzed the explosion of complex forms of social organization, such as universities, sports leagues, fraternities/sororities, study groups, and charity groups.

This productive dimension of the institutional stance is on full display when it generates patterns of social coordination that bring fundamentally new kinds of entities into existence. Take, for example, a ballot, the act of voting, and a citizen. These kinds are not merely socially constructed in a narrow sense, such as the sense in which we might say a tool is socially constructed if it required several people to come together to participate in its fabrication. Instead, each of these kinds is socially constituted in a stronger sense in which defining them requires invoking social forces, including social roles and the relations between them (what we here refer to as institutions). As discussed extensively by philosophers in the social ontology tradition (Ásta, Reference Ásta2018; Searle, Reference Searle2005, Reference Searle2006; Hindriks & Guala, Reference Hindriks and Guala2015) and more recently by cognitive scientists (Noyes, Keil, & Dunham, Reference Noyes, Keil and Dunham2018; Noyes & Dunham, Reference Noyes and Dunham2020), these entities have a unique ontological status in that their existence is not just brought about by but actively sustained by social forces. An institutional process can render a piece of paper a ballot, a particular act of marking that paper voting, and a particular individual a citizen; a further change in institutional structure can reverse those changes even without affecting any intrinsic or even any physical facts about the object, action, or person. This contrast becomes clear when we consider institutional kinds in comparison to objects like hammers, acts like hammering, and people who are carpenters. The factors which make these things what they are can be defined without directly making reference to social forces (even if, as we alluded to above, social forces may have been involved in initially bringing them about, say by manufacturing them or communally developing a set of practices). Institutional processes might be involved in regulating the manufacture of hammers, the allowable forms of hammering, or who can perform the role of carpenter. But whatever those processes specify, we would still consider objects designed to deliver force to be hammers, and the act of driving in a nail hammering. On the other hand, a ballot can really cease to be a ballot, and an action can really cease to be voting, as the direct result of institutional processes.

Existing work assessing whether these philosophical distinctions map onto everyday psychology suggests that adults and older children assert that changes in institutional practice (such as what most people do and believe) can make an object like a ballot no longer a ballot, or an action like voting no longer voting. By contrast, they assert that this is not the case for objects like hammers, or actions like hammering (Noyes, Keil, & Dunham Reference Noyes, Keil and Dunham2018; Noyes, Keil, & Dunham, Reference Noyes, Keil and Dunham2020; Noyes, Dunham, & Keil, Reference Noyes, Dunham and Keil2023). In other words, for a special class of institutional kinds, people explain their causal powers not via internal features of the objects or agents, but rather because of the way in which those objects or agents relate to patterns of human intention that are themselves positioned within social structures. This work suggests that people come to see certain classes of objects, actions, and human kinds as institutional in the relevant sense, raising the possibility that the institutional stance is the cognitive basis of this form of social reality, a vital component of the generativity of human social structures.