Natural-Language Multi-Agent Simulations of Argumentative Opinion Dynamics

  • 2021-04-14 09:45:22
  • Gregor Betz
  • 3


This paper develops a natural-language agent-based model of argumentation(ABMA). Its artificial deliberative agents (ADAs) are constructed with the helpof so-called neural language models recently developed in AI and computationallinguistics. ADAs are equipped with a minimalist belief system and may generateand submit novel contributions to a conversation. The natural-language ABMAallows us to simulate collective deliberation in English, i.e. with arguments,reasons, and claims themselves -- rather than with their mathematicalrepresentations (as in formal models). This paper uses the natural-languageABMA to test the robustness of formal reason-balancing models of argumentation[Maes & Flache 2013, Singer et al. 2019]: First of all, as long as ADAs remainpassive, confirmation bias and homophily updating trigger polarization, whichis consistent with results from formal models. However, once ADAs start toactively generate new contributions, the evolution of a conservation isdominated by properties of the agents *as authors*. This suggests that thecreation of new arguments, reasons, and claims critically affects aconversation and is of pivotal importance for understanding the dynamics ofcollective deliberation. The paper closes by pointing out further fruitfulapplications of the model and challenges for future research.


Quick Read (beta)

loading the full paper ...