A Socially Aware Bayesian Model for Competitive Foraging

Effectively solving the problem of when and where to forage is critical for survival in many animal species. The task is further complicated when there are other agents, potentially competing for the same limited resources. Previous models of foraging consider agents either in isolation or in groups but without competition. Here, we present a novel Bayesian model for competitive foraging, Socially Aware Bayesian Agent (SABA), that takes into explicit account the presence of other agents for both learning and decision-making. For comparison, we also implement a simple Naive Agent model that completely ignores the presence of other agents. We find that although all models perform well in a stationary environment, converging quickly to the optimal population-level solution, only SABA with the stochastic foraging policy can readily adapt when the environment is non-stationary. These results represent a first step toward cognitively sophisticated representations for learning and decision-making in competitive foraging.