Taking sociality seriously: the structure of multi-dimensional social networks as a source of information for individuals

Abstract

Understanding human cognitive evolution, and that of the other primates, means taking sociality very seriously. For humans, this requires the recognition of the sociocultural and historical means by which human minds and selves are constructed, and how this gives rise to the reflexivity and ability to respond to novelty that characterize our species. For other, non-linguistic, primates we can answer some interesting questions by viewing social life as a feedback process, drawing on cybernetics and systems approaches and using social network neo-theory to test these ideas. Specifically, we show how social networks can be formalized as multi-dimensional objects, and use entropy measures to assess how networks respond to perturbation. We use simulations and natural 'knock-outs' in a free-ranging baboon troop to demonstrate that changes in interactions after social perturbations lead to a more certain social network, in which the outcomes of interactions are easier for members to predict. This new formalization of social networks provides a framework within which to predict network dynamics and evolution, helps us highlight how human and non-human social networks differ and has implications for theories of cognitive evolution.

Figure 1.

A social network is composed of b interaction networks between n individuals (here n = 25). Links between individuals are defined as interactions in b different behavioural contexts (dimensions). The clustering coefficient of an individual is calculated by estimating the number of triangular interactions (all members of a triad are connected), given all the possible triads in which that individual is involved. When individuals can interact in several behavioural dimensions, trianglular interactions and triads can form in a number of different ways (inset a: where b = 3). We can see that, in these three cases, if we simply assumed the social network to be the union of the three interaction networks, then the clustering coefficient of all three vertices would be unity, when, in fact, all three examples present very different clustering configurations that have different structural information. Conceptualizing the network as a multi-dimensional object eliminates this problem.

Figure 2.

Paired differences in (a,c) in- and (b,d) out-clustering coefficients for the nearest neighbour (a,b) and grooming (c,d) directed weighted social networks. Captions associated with each boxplot provide the number of paired comparisons calculated (n) and the number of individuals that had a significant change in clustering coefficient after the treatment (asterisk).

Figure 3.

(a) Changes in the standardized joint entropy of the social network under the different treatments and (b) the entropy of each interaction network for each treatment ‘before’ conditions. These values range from zero (highly constrained interactions) to unity (homogeneous interactions). Error bars are jack-knifed standard errors.