Development of AI-supported conversation training to foster democratic exchange (DemocraGPT)

Project description

Political conversations are a cornerstone of democratic understanding. However, in a climate characterised by moralised debates, expectations of social sanctions, and perceived restrictions on freedom of speech, an increasing number of people are withdrawing from political discourse—particularly from challenging conversations regarding difficult topics. This jeopardises social cohesion and, in the long term, the legitimacy of democratic processes.

The consortium project DemocraGPT investigates the potential of Large Language Models (LLMs) to strengthen the willingness and capacity to engage in challenging political conversations. The project proceeds from the premise that dysfunctional psychological reactance—resistance to perceived patronisation—constitutes a central hurdle to successful dialogue. Building on this, the project is developing an AI-supported conversation training programme that sensitises individuals to typical reactance patterns, imparts effective conversation strategies, and promotes the ability to constructively tolerate ambiguity and conflict tension.

Within an interdisciplinary collaboration spanning communication science, political science, social psychology, and computer science, the project will produce: (1) a meta-analysis of effective conversation strategies; (2) an innovative, LLM-based training system featuring specially trained reactance archetypes; (3) a large-scale panel study for the empirical evaluation of this training environment; and (4) a publicly usable and evaluated version of the training environment.

This training environment is designed to combine text- and voice-based interactions, automated feedback, gamification elements, and a data-sovereign LLM infrastructure. The aim of DemocraGPT is to create scientifically grounded tools that support people in navigating political differences empathically, tolerantly, and with less reactance, thereby contributing to the strengthening of democratic conversational culture.

Project team