The Battle of Wits: Comparing Single-Agent System and Multi-Agent Debate System in Scaffolding Argumentative Writing

• In Event: AER Roundtable Session Friday 1:45 pm - Gold 4
  In Roundtable Session: Human–AI Collaboration and Dialogic Learning

Abstract

This study investigated how a multi-agent debate system supports undergraduate students’ argumentative writing and critical thinking compared to a single-agent system. Forty-seven Chinese students in an English for Academic Purposes course completed three writing tasks under both systems. The multi-agent system simulated dynamic, role-based debates with real-time feedback and structured interaction. Results showed that the system enhanced structured human–AI interaction and enabled students to achieve writing outcomes comparable to their best performance with a single-agent system, while requiring significantly less time. Students reported benefits in idea generation, critical reasoning, and metacognitive growth. These findings revealed distinct instructional effects across single-agent and multi-agent systems, highlighting the pedagogical value of multi-agent frameworks for improving argumentative writing quality, critical thinking, and learning efficiency.

Authors

• Lin Dai, East China Normal University

• Jing Leng, East China Normal University