New AI game theory paper cracks coalition formation complexity

Researchers Šimon Schierrreich and Ildikó Schlotter from arXiv:2608.14461 have published a groundbreaking paper on coalition formation in AI systems, introducing a framework for constrained hedonic games where agents form groups based on friend/enemy relationships. The study examines two preference models: enemy-oriented, which reduces to size-constrained graph coloring, and friend-oriented, which exhibits far more intricate complexity governed by enmity structures rather than friendships.

The complexity further depends on strictness of size requirements and symmetry of relationships. Symmetric settings often yield tractable solutions, while asymmetry introduces intractability—with parameterized complexity analysis mapping these boundaries. The findings provide a complete understanding of which structural properties drive computational hardness, offering critical insights for multi-agent AI systems, social network optimization, and cooperative robotics.