Maintaining a large social group is challenging. Dunbar proposes three solutions that help buffer against group fragmentation: (1) structural: maintain coordination by following one’s neighbors, (2) behavioral: increase social grooming time, and (3) cognitive: leverage self-control and mentalizing to avoid conflict. While these solutions may be sufficient for maintaining smaller (nonhuman primate) groups around 50 members in size (Dunbar, Reference Dunbar1992, Reference Dunbar2024; Dunbar, Mac Carron, & Shultz, Reference Dunbar, Mac Carron and Shultz2018; Pasquaretta et al., Reference Pasquaretta, Levé, Claidière, van de Waal, Whiten, MacIntosh, Pelé, Bergstrom, Borgeaud, Brosnan, Crofoot, Fedigan, Fichtel, Hopper, Mareno, Petit, Schnoell, di Sorrentino, Thierry and Sueur2014), they are inadequate for explaining how humans establish and maintain cohesive social networks that are far larger. The sheer size of human social networks – which are typically three times as large as the largest social networks seen in primates (Hill & Dunbar, Reference Hill and Dunbar2003) – requires a solution that is robust to coordination and fragmentation problems at larger scales. We suggest that mental abstraction is the fundamental mechanism driving humans’ ability to maintain cohesion in large groups. Abstraction allows people to build efficient and useful representations of how individuals in their community are connected (Basyouni & Parkinson, Reference Basyouni and Parkinson2022; Park et al., Reference Park, Miller, Nili, Ranganath and Boorman2020; Son, Bhandari, & FeldmanHall, Reference Son, Bhandari and FeldmanHall2023). Relying on abstraction to construct a mental representation of one’s social network provides the advantage of stripping away the granular details of individual ties, distilling complex relational patterns into a higher-order map of the broader network structure (Basyouni & Parkinson, Reference Basyouni and Parkinson2022; Son et al., Reference Son, Bhandari and FeldmanHall2023; Xia et al., Reference Xia, Teoh, Nassar, Bhandari and FeldmanHall2025). Abstraction thus enables humans to carve their networks on the major infrastructure scaffolding the community, a prerequisite to solving the unique challenge of maintaining group cohesion in larger networks.

While Dunbar’s structural solution of simply following your neighbors might be sufficient for smaller networks where individuals have limited local connections, this approach would become brittle if scaled up, as the number of neighbors would increase exponentially. This problem is made even more complex when we consider human networks. Unlike those of primates, human networks contain numerous weak ties (Granovetter, Reference Granovetter1973) that rapidly multiply as the network grows, making it nearly impossible to track all neighboring relationships. These weak ties also create structural holes where information is unlikely to flow between more distant contacts (Burt, Reference Burt2014), further undermining group cohesion. To address these challenges, humans can leverage abstract knowledge of the broader structures within a social network. This enables individuals to more easily discern distinct communities and, critically, to detect bottlenecks or gaps where information might flow (or fail to flow) effectively. By identifying structural holes, humans can strategically fill the gap by becoming a bridge between two otherwise weakly connected communities to facilitate cohesion (and in the process, gain access to valuable social resources; Burt, Reference Burt2014; Nan et al., Reference Nan, Cook, Burt and Lin2008). Unlike primates, who align their behavior primarily with immediate neighbors, humans use these abstract network representations to maintain group cohesion by actively addressing structural inefficiencies (Teoh et al., Reference Teoh, Son, Xia, Bhandari and FeldmanHall2025). Such intentional bridging behaviors help humans form a sparser, yet more resilient and cohesive, network structure than primate networks (Pasquaretta et al., Reference Pasquaretta, Levé, Claidière, van de Waal, Whiten, MacIntosh, Pelé, Bergstrom, Borgeaud, Brosnan, Crofoot, Fedigan, Fichtel, Hopper, Mareno, Petit, Schnoell, di Sorrentino, Thierry and Sueur2014).

Similar logic applies to the behavioral solution. While primates maintain social ties through increased grooming, this strategy has an upper limit, especially in larger networks. Given the vast number of social connections humans must navigate, humans instead rely on verbal communication – such as gossiping – as a behavioral solution (Dunbar, Reference Dunbar1998). Transmission of information does not follow the same properties as physical acts (Acemoglu, Ozdaglar, & ParandehGheibi, Reference Acemoglu, Ozdaglar and ParandehGheibi2010; Chen, Castillo, & Lakshmanan, Reference Chen, Castillo and Lakshmanan2014; Daley & Kendall, Reference Daley and Kendall1964), since gossip, for example, can simultaneously reach many people within the network, thus making verbal communication qualitatively distinct from (and more efficient than) grooming. For gossip to strengthen social bonds rather than damage reputations, humans must gossip strategically. Abstract maps of a social network support adaptive decisions about whom to share information with and whom to avoid gossiping with Xia et al., Reference Xia, Teoh, Nassar, Bhandari and FeldmanHall2025). For instance, in a scenario where it may be critical for information to only reach specific parts of the network but not others, an abstract representation of the network would facilitate predictions about the likelihood of gossip spreading along chains of connections stemming from the source, enabling more strategic social navigation.

And the cognitive solution? While better self-control and mentalization help prevent conflicts in groups, the highly dynamic nature of large social networks calls for a cognitive approach that goes beyond immediate, in-the-moment problems. Humans living in large groups must plan ahead and predict future social dynamics to preemptively prevent group conflicts that could have devastating consequences on social cohesion. Equipped with an abstract representation of one’s community, humans can predict the formation (or dissolution) of future social connections (i.e., link prediction), as well as the consequences of those changes in social ties on the overall group dynamics and cohesion. This capacity for future-oriented thinking is particularly useful for mitigating group conflicts caused by negative ties (e.g., enemies), which naturally threaten group cohesion but are nonetheless part and parcel of human life (Offer, Reference Offer2021). For instance, abstract network representations can help humans foresee a united group splitting into polarized factions, even when only a single negative relationship initially exists within the network. This foresight enables humans to identify potential fragmentation in advance and take preventive measures to maintain group stability (Ackermann, Reference Ackermann2003; Axelrod, Daymude, & Forrest, Reference Axelrod, Daymude and Forrest2021; Sherif, Reference Sherif1988). The absence of such cognitive foresight can have catastrophic consequences – as history has taught us. A striking example is the betrayal of Caesar by Brutus, in which an unanticipated rupture of a central tie ultimately led to Caesar’s assassination and the collapse of the Roman Republic (Tempest, Reference Tempest2017).

Mental abstraction offers a comprehensive but parsimonious account for how humans address the challenge of cohesion in large groups and serves as an overarching framework that encompasses the three solutions proposed by Dunbar. While recent empirical research illustrates that humans build abstract mental representations for social networks (Hu et al., Reference Hu, Vives, Bhandari and FeldmanHall2025; Jolly et al., Reference Jolly, Sadhukha, Iqbal, Molani, Walsh, Manning and Chang2023; Parkinson, Kleinbaum, & Wheatley, Reference Parkinson, Kleinbaum and Wheatley2017; Son, Bhandari, & FeldmanHall, Reference Son, Bhandari and FeldmanHall2021, Reference Son, Bhandari and FeldmanHall2023, Reference Son, Vives, Bhandari and FeldmanHall2024; Teoh et al., Reference Teoh, Son, Xia, Bhandari and FeldmanHall2025; Xia et al., Reference Xia, Teoh, Nassar, Bhandari and FeldmanHall2025), little is known about whether nonhuman primates also do so. A key “to-be-tested” hypothesis that falls out of the abstraction solution is that the ability to create increasingly abstract representations of one’s social network should directly scale with the size of their network. While research shows that humans with greater knowledge of abstract social structures move into more influential positions in their social network (Aslarus et al., Reference Aslarus, Son, Xia and FeldmanHall2025), highlighting the benefits of abstraction for navigating larger groups, future studies should examine whether greater mental abstraction capacity is directly linked to larger network size across species. The reverse relationship should also be examined: the larger the network, the greater the demand and utility for building abstract representations of that network.