Simulating Politics: AI Models for Negotiation and Governance

As part of the Centre for AI in Government (CAIG), this project investigates how Large Language Model (LLM)-based multi-agent systems (MAS) can be designed to simulate political and diplomatic decision-making. With generative AI playing an increasingly prominent role in public sector innovation, there is an urgent need to understand how such systems might credibly model processes like negotiation, deliberation, and institutional discourse.

This research focuses on the epistemological and technical challenges of building interpretable, high-fidelity simulations of political behaviour. Potential, down the line applications for such technology include generating counterfactual scenarios for policy forecasting and training civil servants to navigate complex environments.

The project’s first case study focuses on the European Council, selected for its structured, data-rich deliberative processes. In collaboration with the DICEU project at University College Dublin, led by James Cross, the team is building a digital twin of EU Council negotiations based on annotated transcripts. To assess output reliability, a customised evaluation protocol has been developed to address the specific demands of synthetic political text.

By combining computational experimentation with critical reflection, the project contributes to CAIG’s broader mission of exploring how AI can support democratic governance and responsible innovation in public decision-making.