More Correcting or Confounding? When an AI Arguing Agent Informed by Common Student Misconceptions Joins Case-Based Peer-Argumentation in a Real Course

Jérôme Brender, Aitor Perez, Patrick Jermann, Engin Bumbacher, Francesco Mondada, Chenyang Wang
Proceedings of the Impactful and Responsible AI Systems for Education Workshop, PMLR 339:80-89, 2026.

Abstract

Peer-argumentation is an effective and widely adopted learning activity, yet it remains difficult to enact at scale. While pedagogical conversational agents (PCAs) have been extensively studied, their role as arguing peers that engage students in dialogic argumentation is underexplored. To address this gap, we present ArguBot, a didactic AI arguing partner designed to support scalable case-based peer-argumentation activities. ArguBot deliberately adopts an opposing stance, challenging students’ correct claims with common student misconceptions and countering incorrect claims with arguments grounded in course lecture materials, supported by Retrieval-Augmented Generation (RAG). We conducted a semester-long field study in a graduate robotics course of 172 students, in which both in-class peer-peer argumentation and after-class peer-AI argumentation with ArguBot were offered under regular instructional conditions. Analyses draw on pre-post multiple-choice question responses from students who voluntarily participated in in-class activities ($n = 85$) and/or engaged with ArguBot after class ($n = 68$), as well as post-semester perception surveys ($n = 57$). Results show no statistically significant differences between conditions in error correction or correctness preservation. In contrast, students rated the AI arguing partner significantly lower than a human peer, including reporting lower levels of interest/enjoyment. These findings provide early, ecologically valid evidence on the opportunities and challenges of deploying AI arguing agents in higher education to scale argumentation-based learning.

Cite this Paper

BibTeX
@InProceedings{pmlr-v339-brender26a, title = {More Correcting or Confounding? When an AI Arguing Agent Informed by Common Student Misconceptions Joins Case-Based Peer-Argumentation in a Real Course}, author = {Brender, J\'er\^ome and Perez, Aitor and Jermann, Patrick and Bumbacher, Engin and Mondada, Francesco and Wang, Chenyang}, booktitle = {Proceedings of the Impactful and Responsible AI Systems for Education Workshop}, pages = {80--89}, year = {2026}, editor = {Basu Mallick, Debshila and Woodhead, Simon and Wang, Zichao and Ananda, Muktha and Burstein, Jill and Murphy, April}, volume = {339}, series = {Proceedings of Machine Learning Research}, month = {28 Jun}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v339/main/assets/brender26a/brender26a.pdf}, url = {https://proceedings.mlr.press/v339/brender26a.html}, abstract = {Peer-argumentation is an effective and widely adopted learning activity, yet it remains difficult to enact at scale. While pedagogical conversational agents (PCAs) have been extensively studied, their role as arguing peers that engage students in dialogic argumentation is underexplored. To address this gap, we present ArguBot, a didactic AI arguing partner designed to support scalable case-based peer-argumentation activities. ArguBot deliberately adopts an opposing stance, challenging students’ correct claims with common student misconceptions and countering incorrect claims with arguments grounded in course lecture materials, supported by Retrieval-Augmented Generation (RAG). We conducted a semester-long field study in a graduate robotics course of 172 students, in which both in-class peer-peer argumentation and after-class peer-AI argumentation with ArguBot were offered under regular instructional conditions. Analyses draw on pre-post multiple-choice question responses from students who voluntarily participated in in-class activities ($n = 85$) and/or engaged with ArguBot after class ($n = 68$), as well as post-semester perception surveys ($n = 57$). Results show no statistically significant differences between conditions in error correction or correctness preservation. In contrast, students rated the AI arguing partner significantly lower than a human peer, including reporting lower levels of interest/enjoyment. These findings provide early, ecologically valid evidence on the opportunities and challenges of deploying AI arguing agents in higher education to scale argumentation-based learning.} }
Endnote
%0 Conference Paper %T More Correcting or Confounding? When an AI Arguing Agent Informed by Common Student Misconceptions Joins Case-Based Peer-Argumentation in a Real Course %A Jérôme Brender %A Aitor Perez %A Patrick Jermann %A Engin Bumbacher %A Francesco Mondada %A Chenyang Wang %B Proceedings of the Impactful and Responsible AI Systems for Education Workshop %C Proceedings of Machine Learning Research %D 2026 %E Debshila Basu Mallick %E Simon Woodhead %E Zichao Wang %E Muktha Ananda %E Jill Burstein %E April Murphy %F pmlr-v339-brender26a %I PMLR %P 80--89 %U https://proceedings.mlr.press/v339/brender26a.html %V 339 %X Peer-argumentation is an effective and widely adopted learning activity, yet it remains difficult to enact at scale. While pedagogical conversational agents (PCAs) have been extensively studied, their role as arguing peers that engage students in dialogic argumentation is underexplored. To address this gap, we present ArguBot, a didactic AI arguing partner designed to support scalable case-based peer-argumentation activities. ArguBot deliberately adopts an opposing stance, challenging students’ correct claims with common student misconceptions and countering incorrect claims with arguments grounded in course lecture materials, supported by Retrieval-Augmented Generation (RAG). We conducted a semester-long field study in a graduate robotics course of 172 students, in which both in-class peer-peer argumentation and after-class peer-AI argumentation with ArguBot were offered under regular instructional conditions. Analyses draw on pre-post multiple-choice question responses from students who voluntarily participated in in-class activities ($n = 85$) and/or engaged with ArguBot after class ($n = 68$), as well as post-semester perception surveys ($n = 57$). Results show no statistically significant differences between conditions in error correction or correctness preservation. In contrast, students rated the AI arguing partner significantly lower than a human peer, including reporting lower levels of interest/enjoyment. These findings provide early, ecologically valid evidence on the opportunities and challenges of deploying AI arguing agents in higher education to scale argumentation-based learning.
APA
Brender, J., Perez, A., Jermann, P., Bumbacher, E., Mondada, F. & Wang, C.. (2026). More Correcting or Confounding? When an AI Arguing Agent Informed by Common Student Misconceptions Joins Case-Based Peer-Argumentation in a Real Course. Proceedings of the Impactful and Responsible AI Systems for Education Workshop, in Proceedings of Machine Learning Research 339:80-89 Available from https://proceedings.mlr.press/v339/brender26a.html.

Related Material