Abstract: In social systems ranging from ant colonies to human society, behavioural specialization—consistent individual differences in behaviour—is commonplace: individuals can specialize in the tasks they perform (division of labour (DOL)), the political behaviour they exhibit (political polarization) or the non-task behaviours they exhibit (personalities). Across these contexts, behavioural specialization often co-occurs with modular and assortative social networks, such that individuals tend to associate with others that have the same behavioural specialization. This raises the question of whether a common mechanism could drive co-emergent behavioural specialization and social network structure across contexts. To investigate this question, here we extend a model of self-organized DOL to account for social influence and interaction bias among individuals—social dynamics that have been shown to drive political polarization. We find that these same social dynamics can also drive emergent DOL by forming a feedback loop that reinforces behavioural differences between individuals, a feedback loop that is impacted by group size. Moreover, this feedback loop also results in modular and assortative social network structure, whereby individuals associate strongly with those performing the same task. Our findings suggest that DOL and political polarization—two social phenomena not typically considered together—may actually share a common social mechanism. This mechanism may result in social organization in many contexts beyond task performance and political behaviour.
4. Discussion
Our main result demonstrates that, in the presence of homophily with positive influence, the feedback between social influence and interaction bias could result in the co-emergence of DOL and modular social network structure. These results reveal that self-organized specialization could give rise to modular social networks without direct selection for modularity, filling a gap in our knowledge of social organization [55] and mirroring findings in gene regulatory networks, which can become modular as genes specialize [56]. The co-emergence requires both social influence and interaction bias but, if the level of social influence is too high, its pressure leads to conformity, which homogenizes the society. Because this feedback between social influence and interaction bias has also been shown to drive political polarization [22–25], our results suggest a shared mechanism between two social phenomena—polarization and DOL—that have not traditionally been considered together and raise the possibility that this mechanism may structure social systems in other contexts as well, such as in the case of emergent personalities [11,29–31]. Furthermore, the ubiquity of this mechanism may help explain why social systems often have a common feature—modular network structure—that is shared with a range of other biological and physical complex systems [57].
Intriguingly, although our results suggest that diverse forms of behavioural specialization—and the associated modular, assortative social networks—might arise from a common mechanism, depending on their manifestation, they can be either beneficial or detrimental for the group. For example, DOL and personality differences have long been associated with beneficial group outcomes in both animal [5,58–60] and human societies [61] (although it can sometimes come at the expense of group flexibility [62]). Moreover, the modularity that co-occurs in these systems is also often framed as beneficial, since it can limit the spread of disease [63] and make the social system more robust to perturbation [55]. On the contrary, political polarization is typically deemed harmful to democratic societies [64]. Thus, an interesting question for future research arises: if a common mechanism underlies the emergence of behavioural specialization and the co-emergence of a modular social network structure in multiple contexts, why would group outcomes differ so dramatically? Insights may come from studying the frequency of co-occurrence among various forms of behavioural specialization. If the same mechanism underlies behavioural specialization broadly, then one would expect multiple types of behavioural specialization (e.g. in task performance, personality, decision-making) to simultaneously arise and co-occur in the same group or society, as is the case in some social systems, where certain personalities consistently specialize on particular tasks [9,10] or in human society, where personality type and political ideology appear correlated [65]. Then, the true outcome of behavioural specialization for the group is the net across the different types co-originating from the same mechanism and cannot be inferred by investigating any one specific instantiation of behavioural specialization.
While DOL emerged when homophily was combined with positive influence, other combinations of social influence and interaction bias may nevertheless be employed in societies to elicit other group-level phenomena. For instance, under certain conditions, a society might benefit from uniform rather than divergent, specialized behaviour. This is the case when social insect colonies must relocate to a new nest, a collective decision that requires consensus-building [66]. To produce consensus, interactions should cause individuals to weaken their commitment to an option until a large majority agrees on one location. Heterophily with positive influence—preferential interactions between dissimilar individuals that reduce dissimilarity—achieves this dynamic and is consistent with the cross-inhibitory interactions observed in nest-searching honeybee swarms [67]: scouts interact with scouts favouring other sites and release a signal that causes them to stop reporting that site to others. One could imagine that similar dynamics might also reduce political polarization.
Recent work has shown that built environments—physical or digital—can greatly influence collective behaviour [16,18,68–70], but the mechanisms underlying this influence have remained elusive. By demonstrating the critical role of interaction bias for behavioural outcomes, our results provide a candidate mechanism: structures can enhance interaction bias among individuals and thereby amplify the behavioural specialization of individuals. For example, nest architecture in social insect colonies alter collective behaviour [68] and social organization [18] possibly because the nest chambers and tunnels force proximity to individuals performing the same behaviour and limit interactions with individuals performing other behaviours. Similarly, the Internet and social media platforms have changed the way individuals interact according to interest or ideology [16,69,70]: selective exposure to certain individuals or viewpoints creates a form of interaction bias that our results predict would increase behavioural specialization, i.e. political bias. Thus, our model predicts that built environments should increase behavioural specialization beyond what would be expected in more ‘open’, well-mixed environments. This prediction has evolutionary consequences: a nest can increase behavioural specialization without any underlying genetic or otherwise inherent, diversity. Such consequences would further consolidate the importance of built environments—specifically, nests—for the evolution of complex societies. It has been previously argued that the construction of a nest may have been a critical step in the evolution of stable, highly cooperative social groups [71]. Subsequent spatial structuring of the nest would then, according to our findings, bring further benefits to nascent social groups in the form of increased behavioural specialization, e.g. DOL, even in the absence of initial behavioural and/or trait heterogeneity.
Finally, our results shed light on how plastic traits can result in scaling effects of social organization with group size, a finding that tightens theoretical links between the biological and social sciences. Founding sociological theorist, Emile Durkheim, posited that the size of a society would shape its fundamental organization [3]: small societies would have relatively homogeneous behaviour among individuals, but DOL would naturally emerge as societies grew in size and individuals differentiated in behaviour due to social interactions. Similar to Durkheim's theoretical framing, John Bonner famously posited that complexity, as measured by the differentiated types of individuals (in societies) or cells (in multicellular aggregations), would increase as groups grew in size [72]. Bonner argued that the differentiation among individuals was not due to direct genetic determinism but was instead the result of plasticity that allowed individuals to differ as groups increased in size. Our model supports these qualitative predictions and even predicts a rapid transition in organization as a function of group size that results from socially influenced plasticity at the level of the individual. Previous theoretical work showed that DOL could exhibit group size scaling effects even with fixed traits, but these increases in DOL quickly plateaued past relatively small group sizes [5,39]. Our model, along with models of self-reinforced traits [38], demonstrates how DOL could continue to increase at larger group sizes, a pattern observed empirically in both animal [49,73] and human societies [74,75]. For other forms of behavioural specialization, such as emergent personalities or political polarization, the effect of group size is understudied; however, our results suggest similar patterns. Our model further demonstrated that group size can affect social network structure, a dynamic that has only been preliminarily investigated empirically so far [76]. Leveraging new technologies—such as camera-tracking algorithms and social media—that can simultaneously monitor thousands of individuals and their interactions to investigate the effect of group size on societal dynamics could have significant implications as globalization, urbanization and technology increase the size of our social groups and the frequency of our interactions.
---
Modularity is a form of community structure within a group in which there are clusters of strongly connected nodes that are weakly connected to nodes in other clusters. Using each simulation's time-aggregated interaction matrix A, we calculated modularity with the metric developed by Clauset et al. [77]. A modularity value of 0 indicates that the network is a random graph and, therefore, lacks modularity; positive values indicate deviations from randomness and the presence of some degree of modularity in the network.
Frequency of non-random interactions reveals the degree to which individuals are biasing their interactions towards or away from certain other individuals. For a random, well-mixed population, the expected frequency of interactions between any two individuals is pinteract = 1 − (1 − 1/(n − 1))2. For our resulting social networks, we compared this expected well-mixed frequency to the value of each entry aik in the average interaction matrix resulting from the 100 replicate simulations per group size. To determine whether the deviation from random was statistically significant, we calculated the 95% confidence interval for the value of each entry aik in the average interaction matrix. If the 95% confidence interval for a given interaction did not cross the value pinteract, that interaction was considered significantly different than random.
Assortativity is the tendency of nodes to attach to other nodes that are similar in some trait (e.g. here, threshold bias). We measured assortativity using the weighted assortment coefficient [78]. This metric takes values in the range [− 1, 1], with positive values indicating a tendency to interact with individuals that are similar in traits and negative values indicating a tendency to interact with individuals that are different. A value of 0 means random traits-based mixing among individuals.
Posts
No comments:
Post a Comment