Symbiotic Evolution: Integrating Inclusive Fitness into Autonomous Multi-Agent Discovery

TL;DR: Let’s make LLM agents act more like evolving species—rewarding not just individual discovery, but also knowledge shared with “genetically related” agents. We’ll test if this “inclusive fitness” approach sparks richer, more diverse innovations than standard reward structures.

Research Question: Can an inclusive fitness-inspired reward scheme, where agents are incentivized for both their discoveries and those of related agents, enhance the diversity and strategic complexity of open-ended discoveries in a CORAL-like multi-agent LLM system?

Hypothesis: Incorporating inclusive fitness principles will encourage agents to balance self-driven search with knowledge sharing, leading to more robust, diverse, and strategically sophisticated discoveries compared to individual or team-based reward structures.

Experiment Plan: - Setup: Extend CORAL to assign each agent a “genotype” (randomly or by clustering initial agent behaviors). Design a reward function that credits both individual discoveries and those made by genetically similar agents (following the framework in Rosseau et al., 2025).

• Tasks: Replicate the original CORAL tasks and introduce new ones with opportunities for both cooperation and competition.
• Data: Track diversity of solutions, discovery rates, and emergence of cooperative strategies.
• Analysis: Compare against CORAL’s standard setup (individual reward), and team-based setups, measuring solution diversity, rate of innovation, and patterns of knowledge reuse.
• Expected Outcome: The inclusive fitness approach will unlock richer solution spaces and more nuanced agent interactions.

References:

• Qu, A., Zheng, H., Zhou, Z., Yan, Y., Tang, Y., Ong, S. Y., Hong, F., Zhou, K.-Q., Jiang, C., Kong, M., Zhu, J., Jiang, X., Li, S., Wu, C., Low, B. K. H., Zhao, J., & Liang, P. (2026). CORAL: Towards Autonomous Multi-Agent Evolution for Open-Ended Discovery.
• Rosseau, A., Avalos, R., & Now'e, A. (2025). Inclusive Fitness as a Key Step Towards More Advanced Social Behaviors in Multi-Agent Reinforcement Learning Settings. arXiv.org.