Learning Through Argumentation: A Personalised Pedagogical Conversational Agent that Maximises Disagreement for Collaborative Argumentation

Argumentation-based learning has long been recognised as an effective pedagogical approach for promoting critical thinking and reflection, while deepening students' understanding of the subject matter. Despite its benefits, providing students with consistent access to high-quality argumentation practice at scale is a significant educational challenge. Recent advances in Large Language Models (LLMs) offer a promising solution by enabling new interactive and personalised forms of learning through argumentative dialogues. This work presents ArgueMate, an LLM-powered Pedagogical Conversational Agent (PCA) that serves as an interactive partner for educational argumentation exercises. ArgueMate is built on the principle of constructive disagreement, using structured dialogue to enhance learning. ArgueMate engages students in structured argumentation by presenting opposing viewpoints, prompting counterarguments, and eliciting reflective reasoning. ArgueMate was developed through an iterative, user-centred design process that included three empirical studies with 80 university students. The methodology included two formative classroom deployments to refine the agent's interaction model, followed by a controlled experiment to compare the effectiveness of ArgueMate with traditional peer-to-peer argumentation. The results revealed two key findings. First, dialogues with ArgueMate can elicit argumentation that is as constructive as dialogues with human peers. Second, engaging with the agent prompted significant shifts in students' viewpoints, confirming its ability to facilitate deep cognitive engagement. By demonstrating that an LLM-powered PCA can serve as an effective argumentation partner that rivals human peers, this work presents a scalable and personalisable model for enhancing critical thinking in education.