Decomposing Predictive Information in Social Dynamics

Social behaviors include some of the most interesting interactions in living systems, and we suggest that at the core of these interactions is the process of mutual prediction. Here, we analyze predictive behavior in the context of two male zebrafish engaged in a dominance contest (O'Shaughnessy et al, PRX Life 2024). We adapt a partial information decomposition (PID) to quantify the information between the contestants' past and future, using their 3D velocities. As social behaviors can change rapidly in time, we compute PID using a sliding time window, and we optimize the window size to maximize the total predictive information. Within each window, we use a Gaussian approximation of the joint distribution. We find between-contestant asymmetries in redundant, synergistic and unique information indicative of the dominance relationship emergent at the end of the contest. During the contest, redundant information systematically increases, showing that predictive information is progressively shared between individuals. In contrast, synergistic and unique information, which capture information exchange, are approximately constant. Predictive information decomposition thus provides a quantitative, interpretable perspective which is complementary to game-theoretic models of assessment.