When Machines Join the Moral Circle: The Persona Effect of Generative AI Agents in Collaborative Reasoning

TL;DR

Generative AI agents can act as moral interlocutors that reshape collaborative reasoning without mandating moral conclusions. The study uses a triadic AV-dilemma task with human-only, supportive-AI, and contrarian-AI conditions, applying Moral Foundations Theory, AKC, BERTopic/DTW, and Epistemic Network Analysis to assess process-level changes. Findings show supportive AI strengthens grounding/qualification in reasoning while contrarian AI broadens moral framing and sustains value pluralism; both AI conditions reduce thematic drift and stabilize discourse, while final decisions align more with initial positions. These results highlight persona-governed AI as a design lever to elicit reflective, pluralistic moral reasoning in educational settings, emphasizing process over outcome and the importance of monitoring discourse dynamics.

Abstract

Generative AI is increasingly positioned as a peer in collaborative learning, yet its effects on ethical deliberation remain unclear. We report a between-subjects experiment with university students (N=217) who discussed an autonomous-vehicle dilemma in triads under three conditions: human-only control, supportive AI teammate, or contrarian AI teammate. Using moral foundations lexicons, argumentative coding from the augmentative knowledge construction framework, semantic trajectory modelling with BERTopic and dynamic time warping, and epistemic network analysis, we traced how AI personas reshape moral discourse. Supportive AIs increased grounded/qualified claims relative to control, consolidating integrative reasoning around care/fairness, while contrarian AIs modestly broadened moral framing and sustained value pluralism. Both AI conditions reduced thematic drift compared with human-only groups, indicating more stable topical focus. Post-discussion justification complexity was only weakly predicted by moral framing and reasoning quality, and shifts in final moral decisions were driven primarily by participants' initial stance rather than condition. Overall, AI teammates altered the process, the distribution and connection of moral frames and argument quality, more than the outcome of moral choice, highlighting the potential of generative AI agents as teammates for eliciting reflective, pluralistic moral reasoning in collaborative learning.

Figures (4)

• Figure 1: Experimental design following an individual–group–individual (IGI) sequence. Participants first made and justified an individual moral decision (pre-discussion), then engaged in a synchronous text-based discussion with either human peers or a supportive (blue)/contrarian (red) AI teammate (discussion), and finally repeated the moral decision task (post-discussion) to assess shifts in reasoning and justification.
• Figure 2: Predicted probabilities of moral language use by moral foundation and experimental condition. Bars represent estimated marginal means (EMMs) from the binomial mixed-effects model, with 95% confidence intervals.
• Figure 3: Temporal evolution of moral themes across discussion windows under each condition. Colour intensity reflects the relative proportion of windows classified under each moral foundation at each time point.
• Figure 4: Epistemic Network Analysis (ENA) results of the relational structure of moral framing and reasoning during discussion. (a) Group centroid distributions with 95% confidence ellipses by condition. (b) Contrarian AI (red) vs. Control (purple), (c) Supportive AI (green) vs. Control, and (d) Contrarian vs. Supportive AI