Dynamic Evolutionary Units: Adaptive Switching Between Group and Individual Evolution

TL;DR: What if agents could fluidly switch between evolving in groups and evolving as individuals, based on the nature of the task or the current stage of learning? The first experiment could let agents self-select their evolutionary unit (group or solo) at each generation, then measure adaptability and performance across varying task complexities.

Research Question: Can allowing agents to dynamically choose between group-based and individual evolution lead to better long-term performance and adaptability than committing to a single evolutionary unit?

Hypothesis: Hybrid evolutionary dynamics—where agents can adaptively switch between group and solo evolution—will outperform static group-only or individual-only paradigms, especially in environments with shifting task demands or resource constraints.

Experiment Plan: Implement a system where agents can self-select their evolutionary unit at each generation (solo or group). Compare performance on benchmarks with varying task complexities and environmental changes. Track task success, diversity of solutions, and speed of adaptation to new environments. Analyze under what circumstances agents prefer one mode over the other.

References:

• Weng, Z., Antoniades, A., Nathani, D., Zhang, Z., Pu, X., & Wang, X. E. (2026). Group-Evolving Agents: Open-Ended Self-Improvement via Experience Sharing.
• Bonforti, A., & Solé, R. V. (2021). Unicellular–multicellular evolutionary branching driven by resource limitations. Journal of the Royal Society Interface.
• Silva, F., Correia, L., & Christensen, A. (2015). R-HybrID: Evolution of Agent Controllers with a Hybrisation of Indirect and Direct Encodings. Adaptive Agents and Multi-Agent Systems.