Representational capacity shapes both the "whether" and "how" of social learning

Abstract

Social learning is widely understood as offering a mechanism to mitigate the costs and risks of individual trial-and-error exploration. This cost-avoidance account implies a framing of social learning as a resource-rational adaptation, which should be most beneficial to populations with limited capacity to learn asocially. But in cases of peer-to-peer transmission, that same limited capacity may hinder the reliability of social information--rendering it less, not more, useful. So how do these conflicting intuitions resolve? Across a series of simulation experiments, we find evidence for an inverse-U relationship where social information use emerges most strongly in populations with moderate capacity relative to the complexity of their environment. Furthermore, we demonstrate a co-evolutionary transition in social learning strategies: as population capacity rises, selection pressure shifts from favouring a success bias to a conformist approach. Our findings suggest that constraints on information-processing play an important role in the emergence of culture, determining not just if a species learns socially, but also how. All code for this project is available at https://github.com/maxtaylordavies/capacity-and-social-learning Author summaryMany animals learn not only from their own trial-and-error exploration, but also by observing others. This "social learning" can be extremely useful, allowing young or inexperienced individuals to avoid the dangers of interacting with the environment themselves. This may be especially true for animals that are less able to learn by themselves due to limited attention or memory. But if these same limits affect those animals peers, the social information available will become less reliable, making social learning less, not more, useful. To reconcile these two opposing factors, we used computer simulations of agents learning which unfamiliar mushrooms were edible. We varied how much information agents could observe about each mushroom, and whether they learned alone or from others. We found that the benefits of social learning were strongest for neither the most nor least capable agents, but rather in between, where others were reliable enough to help but individual learning was still difficult. We also found that the same intervention influenced the best "strategy" for choosing whom to learn from: when agents could observe little information, it was best to copy the single most successful individual; when they could observe more information, it was best to follow the majority. July 1, 2026 1/18