Incentivising Collaboration to Counteract Polarisation in Social Systems

About the Project

Societal polarisation has become one of the most pressing challenges of the 21st century. Growing divisions between social, political, or cultural groups undermine the capacity of societies to address urgent shared problems. Polarisation is closely intertwined with segregation, i.e., the tendency of individuals to interact mainly with similar others (homophily). Extensive evidence, including classic models such as Schelling’s model of segregation, shows that even modest degrees of homophily, typical in everyday behaviour, can lead to large-scale separation and social fragmentation.

This project investigates how small-scale incentives for cooperation can mitigate or reverse such emergent polarisation. The overarching goal is to identify the minimal interventions required to sustain large-scale collaboration in the presence of natural tendencies toward segregation. We will model social interaction as a multi-agent system, using frameworks from coalition and network formation to study how local incentives, such as rewards for diverse collaboration, can promote stable, mixed, and cohesive structures at the population level.

The research will combine a rigorous theoretical component (e.g., methods from algorithm design and computational complexity) with a practical one (e.g., agent-based simulations and testing on empirical data). Particular attention will be paid to budget-limited interventions, robustness under shocks, and the minimal incentive strength needed to maintain collaborative, well-mixed networks.

A further goal is to integrate the findings into systems of collective or distributed artificial intelligence, where similar coordination and alignment challenges arise. The anticipated outcome is a comprehensive toolkit, both theoretical guarantees and implementable design principles, for a broad range of entities, such as digital platforms and community organisations, seeking to foster collaboration across diverse groups.