Scholarlink

<p>Despite the potential of generative artificial intelligence (GAI) to enhance learning, the in-class efficacy of multi-agent systems remains under-examined. This study aims to address student-centered factors influencing their intentions towards GAI by using the Situated Expectancy Value Theory (SEVT) to understand university students’ preliminary knowledge and their perceptions of using the newly developed multi-agent debate system. A 41-item SEVT survey was administered to 121 EFL undergraduates. One intact class (<i>n</i> = 47) participated in an 18-week argumentative writing course supported by four AI agents (AI Tutor, Debate Assistant, AI Debater, AI Debate Expert) which scaffolded claim construction, discourse regulation, and rubric-aligned feedback. Human-AI interaction data (<i>n</i> = 3,975 turns) were coded using a modified Toulmin model. Exploratory factor analysis confirmed a robust six-factor SEVT structure. Survey results indicated high expectancy for success and perceived social support, moderate task value, and low perceived cost. Discourse analysis showed that students predominantly engaged in constructing and refining claims supported by evidence, while counterargument moves occurred less frequently and were often integrated through rebuttal-oriented responses rather than direct refutation. The interviews indicated that students exhibit a measured trust in AI agents and favor a deliberative approach, preferring to scrutinize and qualify AI-generated perspectives over engaging in direct confrontation. The findings suggest that multi-agent debate systems can facilitate reflective and deliberative engagement in argumentation. This research contributes empirically substantiated insights that are valuable for guiding the pedagogical implementation and design of socially supportive, iterative AI-facilitated debate environments.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Too Costly or Worth the Effort? Exploring University Students’ Perceptions of a Multi-agent Debate System in Argumentative Writing

  • 期刊论文

  • 发表日期：

    2026年3月4日

查看全文
  • Jing Leng,
  • Yinuo Li,
  • Jun Qian,
  • Lin Dai,
  • Ronnel B. King

摘要

Despite the potential of generative artificial intelligence (GAI) to enhance learning, the in-class efficacy of multi-agent systems remains under-examined. This study aims to address student-centered factors influencing their intentions towards GAI by using the Situated Expectancy Value Theory (SEVT) to understand university students’ preliminary knowledge and their perceptions of using the newly developed multi-agent debate system. A 41-item SEVT survey was administered to 121 EFL undergraduates. One intact class (n = 47) participated in an 18-week argumentative writing course supported by four AI agents (AI Tutor, Debate Assistant, AI Debater, AI Debate Expert) which scaffolded claim construction, discourse regulation, and rubric-aligned feedback. Human-AI interaction data (n = 3,975 turns) were coded using a modified Toulmin model. Exploratory factor analysis confirmed a robust six-factor SEVT structure. Survey results indicated high expectancy for success and perceived social support, moderate task value, and low perceived cost. Discourse analysis showed that students predominantly engaged in constructing and refining claims supported by evidence, while counterargument moves occurred less frequently and were often integrated through rebuttal-oriented responses rather than direct refutation. The interviews indicated that students exhibit a measured trust in AI agents and favor a deliberative approach, preferring to scrutinize and qualify AI-generated perspectives over engaging in direct confrontation. The findings suggest that multi-agent debate systems can facilitate reflective and deliberative engagement in argumentation. This research contributes empirically substantiated insights that are valuable for guiding the pedagogical implementation and design of socially supportive, iterative AI-facilitated debate environments.