Utilizing Generative AI to Counter Learner Groupthink by Introducing Controversy in Collaborative Problem Based Learning Settings

Synthesis: Wiss et al. (2025) describe a mixed-methods study that introduced a generative AI agent — CALIE (Collaborative Agent for Learning and Interprofessional Exploration) — into twelve newly formed interprofessional student teams in a virtual problem-based learning session, intentionally prompted to inject controversial viewpoints and counter groupthink. Across 158 survey respondents from seven health-professions graduate programs, the genAI agent positively influenced group dynamics, stimulated critical thinking and engagement, and limited groupthink's potential. The faculty facilitator's stance toward the AI significantly shaped students' acceptance of it, and students' perceptions of the AI as useful, helpful, and part of the team differed significantly and were positively intercorrelated.

Key Findings

• genAI agents can counter groupthink. Students reported that the agent's regularly introduced controversial responses stimulated critical thinking, introduced divergent outside viewpoints, and gave teams a launchpad to analyze the AI's output using their own disciplinary knowledge — steering groups away from conformity-driven consensus.
• Quantitative perceptions. A repeated-measures ANOVA (F(2, 156) = 26.01, p < .001) found students agreed most that AI can be a useful tool (M = 3.49), then that its feedback was helpful (M = 3.21), and least that AI felt part of the team (M = 2.88); all pairwise comparisons differed significantly. The three items were all positively correlated.
• Facilitator influence matters. Students who said their faculty facilitator influenced how they integrated the AI reported more positive AI perceptions on all three measures — e.g., viewing AI as part of the team (M = 3.08 vs 2.64, F(1,156) = 5.96, p = .02) and finding its Feedback helpful (M = 3.44 vs 2.93, F(1,156) = 10.26, p = .002).
• Disagreement with the AI aided teamwork. Even when learners rejected or mistrusted the agent's responses, the act of engaging with or refuting them offered a socially permissible way to speak up, stimulating reflection and conversation and counteracting passivity in unfamiliar teams.
• Mixed bias findings. Perceptions of AI bias were "complex and messy" — some students held personal negative bias toward AI, and it was unclear whether they labeled controversial responses "biased" because they disagreed with the content; bias in interprofessional AI settings is flagged for future research.

Study Design & Method

A mixed-methods study under Yale IRB. 165 graduate health-professions learners (158 opted into the survey) from seven programs (MPH, nursing, AGACNP, DPT, PharmD, DDS, Physician Associate) were divided evenly across twelve small groups, each facilitated by a faculty member. Teams completed a synchronous 180-minute virtual problem-based learning session using the VIPE interprofessional education methodology, during which the CALIE genAI agent was voiced at three points to introduce potentially controversial perspectives. Data came from an online post-activity survey combining quantitative perception items (analyzed with repeated-measures and multivariate ANOVAs) and qualitative open responses.

What this means for practice

• Instructors. Prompt the AI agent to inject dissenting or controversial viewpoints at set points in the session, and treat the disagreement as instructional material: students reported that refuting the agent's responses gave them a socially permissible way to speak up and stimulated critical thinking.
• Instructors. Endorse the agent visibly — students who said their facilitator influenced how they integrated the AI rated it as more part of the team (M = 3.08 vs 2.64, p = .02) and its Feedback as more helpful (M = 3.44 vs 2.93, p = .002) than those who did not.
• Instructional designers. Decide in advance when the agent speaks and script what it will raise; the CALIE agent was voiced only three times, so its role in the team's ongoing discussion has to be a deliberate design choice rather than ambient chat.
• Faculty developers. Coach facilitators to take an open, non-defensive stance toward AI-generated controversy, since facilitator framing was the strongest observed lever on whether students accepted and benefited from the agent.

Limitations

• The genAI output was pre-generated and read aloud at three fixed points, so students interacted with the agent only in brief interjections rather than a participant in the team's ongoing discussion.
• Teams spent about 60 minutes together, a window the authors judge too short for the novelty of the agent to fade and for genuine groupthink effects to appear; longitudinal follow-up was left to future work.
• Data are self-reported perceptions from 158 of the 165 participating learners, with the qualitative strand drawn solely from open-response survey comments.
• Learners came from seven health-professions graduate programs at a single institution, so other professions (e.g., social work, addiction sciences) and non-health disciplines are unrepresented.

Citation

Wiss, A., Showstark, M., Dobbeck, K., Pattershall-Geide, J., Zschaebitz, E., Joosten-Hagye, D., Potter, K., & Embry, E. (2025). Utilizing generative AI to counter learner groupthink by introducing controversy in collaborative problem based learning settings. Online Learning Journal, 29(3), 39–65.