Articulating complexity

Policies unfold in complex adaptive systems: dynamic webs of individuals, institutions, communities, and ecosystems whose interactions are non-linear, strategic, and historically contingent. In such systems, interventions rarely produce proportional or predictable effects. Outcomes emerge from feedbacks, adaptation, and cascades of interaction—and failures often arise through indirect, unexpected pathways.

Traditional tools such as agent-based models have helped formalize these dynamics, but they often rely on simplified rules of behavior and stripped-down representations of decision-making. What they struggle to capture are the political dimensions of real systems: beliefs and values, negotiation and persuasion, strategic language, deception, broken promises, and tactical adaptation.

Instead of representing social actors through simplified rules or numerical variables, we model them as language-based agents powered by large language models, capable of reasoning, arguing, negotiating, deceiving, and adapting—thereby embedding beliefs, values, rhetoric, and strategic thinking directly into complex-systems simulations.

When these agents interact within fully automated serious games, policies can be stress-tested under strategic pressure, coalition formation, misinformation, and institutional conflict, and explored through thousands of simulated runs—transforming complex-systems modeling from static scenario analysis into an experimental approach to policy foresight.

This research programme is carried out through three funded projects, each applying the same core approach to different high-stakes socio-ecological challenges.

• Can we foresee the outcomes of nature conservation policies?
• NatureDiplomacyAI – AI-generated environmental policies for sustainability
• Generative wargaming of socio-ecological warfare

Read more at articulating complexity