Hive Mind as a Single Reinforcement Learning Agent
A new paper establishes an equivalence between collective decision-making in honey bee swarms and single-agent reinforcement learning. The authors show that the emergent distributed cognition of a bee colony, following simple imitation-based rules, behaves as a single online RL agent interacting with many parallel environments. Specifically, the weighted voter model of bees' waggle dance corresponds to a multi-armed bandit algorithm called Maynard-Cr. This bridges two paradigms: collective decision-making via imitation and trial-and-error learning by a single agent.
Key facts
- arXiv:2410.17517v5
- Announce Type: replace-cross
- Decision-making is essential for intelligent agents or groups
- Natural systems converge to effective strategies via collective decision-making (imitation) or trial-and-error (single agent)
- Paper establishes equivalence between these paradigms using nest-hunting in honey bee swarms
- Emergent distributed cognition (hive mind) from local imitation rules is a single online RL agent
- Weighted voter model of bees' waggle dance corresponds to a multi-armed bandit algorithm
- Algorithm named Maynard-Cr
Entities
—