Surges of collective human activity emerge from simple pairwise correlations

Collective human behavior drives a wide range of phenomena in the modern world, from spikes in mobile phone usage and online traffic to fluctuating demands on transportation and emergency response infrastructure. However, while the correlated activity of one or two individuals is partially understood, it remains unclear if and how these simple low-order correlations give rise to the complex large-scale patterns characteristic of human experience. Here we show that networks of email and private message correspondence exhibit surges of collective activity, which cannot be explained by assuming that humans act independently. Intuitively, this collective behavior could arise from complicated correlations between large groups of users, or from shared daily and weekly rhythms. Instead, we find that the network activity is quantitatively and robustly described by a maximum entropy model that depends only on simple pairwise correlations. Remarkably, we find that the functional interactions in the model, which are learned exclusively from the timing of people's actions, are closely related to the ground-truth topology of correspondence in the population. Together, these results suggest that large-scale patterns of activity emerge organically from pairwise correlations, which, in turn, are largely driven by direct inter-human communication.