As part of my MSc in Artificial Intelligence at VU Amsterdam, I conducted my thesis research in collaboration with TNO (The Netherlands Organisation for Applied Scientific Research). The goal of the project was to simulate conversations between simulated AI actors and study how persuasive messages influenced opinions over time.
Simulation environment
The project built on a custom simulation environment designed for dynamic interactions between agents and real-time monitoring of conversation flows.
- FastAPI backend handling agent logic, simulation events, and message flow.
- Vite frontend for manually steering simulations, inspecting message threads, and visualising agent interactions in real time.
- Asynchronous message loop letting agents respond sequentially in structured multi-turn conversations.
- Logging and persistence of conversations and metadata per session, for reproducible analysis.
Thesis contributions
On top of this environment, the thesis simulated belief shifts in conversations between receiving and perceiving agents using large language models.
- Agent design: receiving agents modelled on Big Five personality traits and belief systems; perceiving agents on persuasive strategies (logos, pathos, ethos).
- LLM-driven dialogue: GPT-based agents simulating psychologically grounded persuasion scenarios.
- Belief-shift tracking: endorsement changes measured with G-Eval, sentiment evolution tracked with the AFINN lexicon.
- Reasoning analysis: FLAN-T5 classifying the types of persuasive reasoning across dialogues.
- Batch experiments: runs across agent pairings and message lengths, with all outputs logged for structured analysis.
Publication
This work was developed into a paper published at LREC 2026 in Palma de Mallorca: Green Bots versus Red Bots: Evaluating Large Language Models for Simulating Persuasion Dynamics in Online Influence Campaigns.