Capturing Argument in Agent-Based Models.

Agent-based models (ABMs) are widely used to study the complex dynamics and emergent properties of systems with many interacting agents. This includes belief and opinion dynamics as are of relevance to understanding contexts as varied as online social media and the practice of science. This paper argues that such ABMs can capture rich argumentation scenarios in ways that have not been covered in research to date. To clarify the space of potential agent-based models of argument, we distinguish three interrelated notions of argument from the literature. First, refer simply to the propositional content encoded in arguments. Second, describe premise-conclusion relationships that arise between such reasons when asserted as arguments. Third, refer to the deployment of reasons and syllogisms in discussions (be they polylogues or dialogues). We show how modelling each of these three notions of argument naturally involves a continuum of complexity. Specifically, we use the NormAN framework (introduced in Assaad et al. https://doi.org/10.48550/arXiv.2311.09254, 2023), which bases ABMs on the theory of Bayesian networks, as a point of reference and draw out its relationship to other modelling frameworks along each of these dimensions. This provides a novel organising scheme to aid model comparison and model choice, and clarifies ways in which these three notions of argument constrain one another. This shows also that NormAN's Bayesian framework not only captures familiar facets of argumentation, but also allows one to study how dialectical considerations influence population level diffusion of arguments (as we demonstrate with a small simulation study).