Multi-Agent Empowerment Drives Emergent Group Behaviors
A new study has broadened the notion of empowerment—an internal drive that encourages an agent to enhance its impact on future conditions—to encompass multi-agent systems. Researchers illustrate how this internal motivation can lead to intricate group dynamics without the need for specific engineering. They developed a systematic extension of empowerment applicable to multiple agents and demonstrated an efficient method for its calculation. The team identified distinct patterns of group organization in two different scenarios: a duo of agents linked by a tendon and a manageable Vicsek flock. These findings indicate that empowerment can facilitate behavioral organization at both the individual and group levels, underscoring the promise of intrinsic motivations for scalable collective behavior.
Key facts
- Empowerment is an intrinsic motivation that measures an agent's influence over its future states.
- The study extends empowerment to multi-agent settings.
- Efficient calculation of multi-agent empowerment is demonstrated.
- Two environments were tested: a pair of agents coupled by a tendon and a controllable Vicsek flock.
- Characteristic modes of group organization emerged from empowerment-driven behavior.
- The work shows intrinsic motivations can drive behavioral organization at scale.
- The research is from the field of artificial intelligence and multi-agent systems.
- The paper is available on arXiv.
Entities
Institutions
- arXiv