Social foraging is a fundamental collective behavior observed across various species, but studying complex multi-agent dynamics and accompanying neural activity in natural settings remains challenging. Here we present an in silico study of social interaction during foraging using recurrent neural network (RNN) agents trained using multi-agent reinforcement learning (MARL). Our framework explores how different group sizes, food availability, and sensory capabilities shape emergent foraging strategies in grops of 2 to 10 agents. These group sizes are especially relevant to experimental social neuroethology. We find that these variables drive transitions in collective behavior, leading to shifts in cohesion, inequality, and inter-agent influence as environmental pressures change. Neural analyses of agent RNN activities reveal high-dimensional representations of task-related variables, such as distance to food and proximity to other agents, suggesting that decision-making is deeply embedded in sensory inputs and social context. Further, we demonstrate that larger groups lead to decentralized decision- making and a reduction in individual influence, as measured by non-linear Granger causality metrics, while food scarcity amplifies disparities in consumption, particularly in competitive settings. Our framework, which simulates social cognition and collective foraging, should be of broad interest for social neuroethology, offering a scalable platform for investigating complex group behaviors in naturalistic settings.