Sharing Rewards Undermines Coordinated Hunting.

Area of Science:

• Behavioral Ecology
• Computational Biology
• Artificial Intelligence

Background:

• Coordinated hunting is common in animals, with reward sharing often assumed to be key.
• Existing theories rely on correlational data, lacking causal evidence for sharing's role.

Purpose of the Study:

• To causally investigate the role of reward sharing in coordinated hunting using computational modeling.
• To challenge the assumption that prosocial motives like sharing are essential for coordination.

Main Methods:

• Utilized a state-of-the-art Multi-agent Reinforcement Learning (MARL) algorithm for computational modeling.
• Simulated predator groups with varying degrees of reward sharing and individual incentives.

Main Results:

• Selfish agents achieved robust coordination, while reward sharing undermined it.
• Sharing led to the free-rider problem, plateaued coordination at small group sizes, and was not a Nash equilibrium.
• Individually rewarded agents outperformed sharing agents, especially in difficult hunts or larger groups.

Conclusions:

• Reward sharing may be a byproduct of hunting, not a driver of coordination.
• Prosocial motives might be less critical for coordination than previously thought.
• Current AI models (e.g., MARL) assuming reward sharing may misrepresent coordination drivers.