Multi-Agent AI Simulates Climate Policy Deliberation Across Living Labs

Climate governance involves complex, multi-stakeholder negotiations. To better understand these dynamics, researchers have built a role-based multi-agent simulation that brings together heterogeneous actors — ordinary citizens, advocacy groups, media outlets, and political decision-makers — into a unified virtual environment. The architecture is designed to be modular: it combines empirically grounded cognitive models (HUMAT and MOA) for individual decision-making with socially embedded influence processes (e.g., demographic homophily networks) and institutional strategy modules for NGOs, media agents, and politicians. Political outcomes emerge from the aggregation of multiple signals, including expert input, public mobilisation, party alignment, and media framing. The model is calibrated using synthetic populations derived from real survey data and institutional parameters from Living Lab stakeholder engagement.

While the paper (accepted as a poster at SSC 2026) focuses on architectural design rather than empirical results, the implications are significant. By simulating how different stakeholder motivations and institutional strategies interact, the model could help policymakers test scenarios before implementing real-world climate adaptation measures. It offers a sandbox for exploring the impact of media influence, NGO advocacy, and shifting public opinion on land-use policies. The authors also outline pathways for generalization, suggesting the framework could be applied to other democratic governance challenges beyond climate. This work represents a step toward more realistic, multi-level simulations of collective decision-making in complex socio-ecological systems.