Exploring AI-Supported Disciplinary Mediation in Student Project Teams' Text-Based Communication

Synthesis: Cheng, Chung, Chiu, Lin & Liao (2026) present Spritz, a Discord-based Large Language Models (LLMs) technology probe that mediates disciplinary boundaries in interdisciplinary student project teams, finding that students valued AI as both cognitive support for boundary crossing and a relational buffer — while a central tension emerged when AI moved from neutral mediator to advisor or challenger.

Key Findings

1. The challenge of interdisciplinary sensemaking. Students in interdisciplinary Active Learning must negotiate differences in language, assumptions, priorities, and practices, yet these differences are hard to surface in fragmented text-based team communication where AI tools often become private side channels.
2. Spritz as a mediating probe. Spritz monitors group chat for semantic or pragmatic boundary signals, prompts members to articulate perspectives through private channels, and returns anonymized syntheses to shared discussion. A technology probe study and co-design workshop ran with 12 university students across technical, business, and design backgrounds.
3. Dual value: cognitive and relational. Participants valued AI mediation not only as cognitive support for boundary crossing but as a relational buffer — organizing fragmented discussion, surfacing implicit expectations, clarifying divergent interpretations, and softening interpersonal pressure around disagreement and concession.
4. The neutrality tension. Participants imagined future AI mediators as switchable roles — strategic advisors, cross-domain translators, perspective challengers — but these expanded roles made the neutrality that had made AI acceptable as a mediator unstable once AI began to advise, challenge, or influence team decisions.

What this means for practice

• Learners. Use an AI mediator as a neutral organizer that surfaces and compares positions, and stay alert that this neutrality erodes as soon as the AI begins to advise or challenge the team.
• Let members review, edit, or redact how their private perspectives are synthesized back into the shared discussion, since the private channel was valued precisely because it lowered the facework cost of disagreement.
• Require the AI to label when it is summarizing, translating, advising, or challenging, so the team can apply the right level of trust and scrutiny to each contribution.
• Designers. Make AI intervention negotiable rather than automatic — detecting a boundary is not the same as intervening appropriately — for example by offering a low-friction prompt before pausing team discussion.
• Preserve human ownership of team decisions: an AI mediator should support negotiation rather than present its synthesis as a final judgment.

Limitations

• Exploratory technology probe and co-design workshop with 12 students aged 18–25, recruited through the researchers' personal networks, working through a single predefined virtual discussion session.
• No control group and no quantitative measures: the findings rest on participants' subjective perceptions, and the boundary-detection goal went unevaluated, with no measure of detection accuracy or intervention timing.
• The roles and the product-pitch scenario were predefined and limited to technology, business, and design boundaries, so the findings do not yet generalize to other disciplinary configurations or to the concrete stakes of authentic projects.

Citation

Cheng, C.-J., Chung, Y.-C., Chiu, B.-C., Lin, Y.-H., & Liao, J.-W. (2026). Exploring AI-Supported Disciplinary Mediation in Student Project Teams' Text-Based Communication. (cs.HC).