Capturing collective conflict dynamics with sparse social circuits

A central problem in the study of collective behavior is how functionally significant macroscopic properties arise out of individual or microscopic interactions. The most common approach to studying the relationship between the micro and macro in biological systems is dynamical systems and pattern formation (for examples, see Sumpter, 2006; Ball, 2009; Couzin, 2009; Payne et al., 2013). A complementary approach is to treat the micro to macro mapping explicitly as a computation. Amacroscopic property can be said to be an output of a computation if it can take on values that have functional consequences at the group or component level, is the result of a distributed and coordinated sequence of component interactions under the operation of a strategy set, and is a stable output of input values that converges (terminates) in biologically relevant time [Flack and Krakauer 2011][Flack 2014]. The input to the computation is the set of elements implementing the rules or strategies. The input plus the strategies constitute the system’s microscopic behavior. My collaborators and I have developed novel computational techniques, called Inductive Game Theory [DeDeo et al. 2010][Flack and Krakauer 2011][Lee et al. 2014], to extract strategic decision-making rules from correlations observed in the time series data and reconstruct the microscopic behavior. In biological systems, there are typically multiple components interacting. Hence the computation of the macroscopic output is inherently collective, meaning we must examine how different configurations of strategies affect the macroscopic output [Flack 2014]. We describe the space of microscopic configurations using Markovian, probabilistic, “social” circuits [DeDeo et al. 2010][Flack and Krakauer 2011][Lee et al. 2014]. Here, we briefly illustrate this approach using time series data collected on conflict dynamics from an animal society model system—a large, captive group of captive pigtailed macaques (Macaca nemestrina) (n = 47). We start with very simple time series data on which individuals were present in a conflict. This gives us a time series of binary fight participation vectors. Given this input of individual identities, our output is the distribution of fight sizes, which has been shown to have functional consequences for individuals in the study system [DeDeo et al. 2010]. We ask whether individuals decide to join fights based on participation in the last fight (say between their allies and adversaries), what are their decision-making strategies, and how do these strategies collectively produce the distribution of fight sizes?