Deliberative polling aims to improve collective decision-making by exposing participants to a broad range of arguments before they vote. However, ensuring every voter encounters a representative sample of the argument space—the coverage problem—remains difficult, especially at scale and in adversarial electorates. A new paper on arXiv introduces a method for evaluating solutions to this problem using an LLM-based agentic simulator.
The Agentic Bipolar Argumentation Simulator (ABAS) is grounded in a formal framework that models a poll as a six-tuple consisting of endorsing and opposing justifications, attack and enhance relations, and shareholder- and relation-weights. ABAS simulates N autonomous shareholder agents, each assigned a latent opinion according to desired distributions in [-1, 1]. These agents sequentially vote, choose or author justifications, and optionally submit argumentation-graph links.
The simulator implements recommendations that rank existing justifications by their observable endorsement mass. It evaluates the mechanism's success by coverage, defined as the fraction of the corpus reason-tag set represented in the K recommendations presented to each shareholder. This is framed as a solution to the NP-hard Subsuming Justification Problem.
Reported experiments characterize how four parameters affect coverage and corpus diversity: creativity rate (p_own), recommendation size (K), argumentation density (p_links), and population size (N). In an authenticated electorate where Sybil attacks are impossible and only the relation graph is gameable, the researchers stress-tested the scoring with coordinated strategic voting. The results showed that a tag-flood attack collapses coverage. However, author-count relation weighting through a reversed-PageRank rule resists the flood markedly better than uniform weights.
For enterprise technology leaders, this research highlights the vulnerability of deliberative polling systems to manipulation and offers a quantitative evaluation framework. While the direct application to supply chain or trade is not addressed in the source, the techniques for measuring coverage and robustness against adversarial behavior could inform the design of consensus-building tools in complex, multi-stakeholder environments such as trade policy or logistics network optimization.
The paper is authored by Rwaida Alssadi, Khulud Alawaji, Balaji Kasula, Muntaser Syed, Badria Alfurhood, Markus Zanker, and Marius Silaghi. It is available on arXiv under the preprint identifier 2606.11692.