# Maynard-Cross Learning **Entity class:** Swarm-level reinforcement-learning rule **Domain:** Collective decision-making / evolutionary dynamics **Maturity:** Research model ## Definition **Maynard-Cross Learning** is the name Soma, Bouteiller, Hamann and Beltrame gave the macro-level reinforcement-learning update that emerges from a particular weighted-voter model of collective choice. The model is inspired by honeybee nest-site selection: scouts promote options in proportion to their estimated quality, and other individuals imitate locally observed promotions. Individual modeled bees do not need to compare the options' payoffs or execute a reinforcement-learning algorithm. Their changing population proportions follow a Maynard-Smith replicator dynamic that can be read as a single multi-armed-bandit learner updating its action probabilities from many parallel quality samples. ## Scope of the equivalence The equivalence is proved for the specified local revision rule and aggregate policy, with simplifying assumptions including a well-mixed, homogeneous population. The authors explicitly say the weighted-voter model is only a partial description of real bee nest selection: spontaneous commitment, stop signals, switching and quorum effects are omitted or simplified. The macro learner is a mathematical description of the population's choice distribution, not evidence that every bee shares one subjective experience or that human society has been shown to be one agent. The result gives [[wiki/Collective Intelligence|Collective Intelligence]] a precise instance in which local imitation and aggregate learning correspond, while [[wiki/Replicator Dynamics|Replicator Dynamics]] supplies the bridge between levels. [[articles/Rock, Paper, Scissors|Rock, Paper, Scissors]] uses this result as a bounded precedent for thinking across individual and group scales. A broader analogy to institutions or machine collectives becomes a research question about their actual interaction rules, heterogeneity and measured aggregate behavior. ## Source - [Soma, Bouteiller, Hamann and Beltrame, *The hive mind is a single reinforcement learning agent*, *Swarm Intelligence* (17 October 2026)](https://link.springer.com/article/10.1007/s11721-026-00264-7) — formal model, proof, simulations and stated biological and population-structure limitations.