“I Felt Bad After We Ignored Her”: Understanding How Interface-Driven Social Prominence Shapes Group Discussions with GenAI

Paper Title

“I Felt Bad After We Ignored Her”: Understanding How Interface-Driven Social Prominence Shapes Group Discussions with GenAI

Publication Info

• Topic area: Human-AI collaboration in group discussions
• Keywords: Generative AI, social prominence, group discussions, interface design, human-AI collaboration, proactive AI, voice-based interactions, epistemic vigilance, user control, agent influence

Background and Problem

• Problem / challenge: Despite advancements in generative AI (GenAI) as a participant in group discussions, there is limited understanding of how interface design impacts its influence and role in collaborative settings.
• Significance: Understanding these dynamics is critical to designing systems that enhance group problem-solving without undermining cohesion or enabling undue influence by AI.
• Motivation and related work: Prior research has explored GenAI's assistive roles in brainstorming and decision-making but has largely focused on reactive contributions. Studies on proactive AI participation, especially in voice-based, real-time discussions, remain nascent. Existing work also lacks insights into how interface design influences human-AI group dynamics.

Solution

• Proposed approach: The study introduces "interface-driven social prominence" as a design lens to explore how GenAI's presence and user control shape group discussions. Three collaboration modes (Roundtable, Peripheral, Breakout) were developed to vary these dimensions.
• Novelty:
  1. Introduction of interface-driven social prominence as a framework for designing collaborative GenAI systems.
  2. Development of a voice-based GenAI agent capable of proactive participation in real-time discussions.
  3. Empirical evidence from a mixed-methods study showing how interface design impacts communication patterns, perceptions, and regulation of GenAI influence.
  4. Identification of collective boundary regulation as a key coordination mechanism in human-AI groups.
• Procedure and key techniques:
  1. A custom video conferencing platform was developed with a proactive GenAI agent named Lisa.
  2. Three collaboration modes were designed: Roundtable (high agent presence, low user control), Peripheral (medium presence, medium control), and Breakout (low presence, high control).
  3. A mixed-methods within-subjects study with 18 dyads (36 participants) was conducted, involving three discussion tasks across the modes.
  4. Data collection included video coding, Ordered Network Analysis (ONA), qualitative feedback, and survey responses.

Results

• Concrete findings:
  • Roundtable mode fostered cooperative behaviors but increased agent influence due to social norms.
  • Peripheral mode allowed intentional engagement but often led to inertia in moving the agent out of the shared space.
  • Breakout mode enabled strategic filtering of agent contributions but encouraged competitive behaviors and reduced critical evaluation.
  • Participants with stronger preferences exhibited higher epistemic vigilance in Roundtable and Peripheral modes compared to Breakout.
• Advantage over baselines:
  • The study highlights how interface design can mitigate risks of GenAI persuasion and disengagement common in solitary use.
  • Shared interactions with GenAI fostered critical evaluation and cooperative behaviors not observed in single-user contexts.
• Experiments / evaluation:
  • Participants engaged in three discussion tasks (brainstorming, decision-making, prioritization) with a GenAI agent across the three modes.
  • Data analysis included discourse analysis, ONA, qualitative coding, and survey metrics (e.g., NASA-TLX, epistemic vigilance, social loafing).
• Limitations and future work:
  • Difficulty isolating the effects of agent presence vs. user control due to bundled design choices.
  • Results may not generalize to larger groups or asynchronous settings.
  • Future research should explore multi-agent ecosystems, long-term interactions, and evolving social norms around GenAI.

Summary

This study investigates how interface-driven social prominence shapes the role and influence of GenAI agents in human-AI group discussions. By varying agent presence and user control across three collaboration modes, the research reveals that interface design fundamentally impacts communication patterns, group dynamics, and regulation of GenAI contributions. Findings suggest that shared interactions with GenAI can enhance critical evaluation and cooperation but also introduce risks of undue influence and competitive behaviors. The study underscores the need for thoughtful interface design to balance GenAI's benefits with mechanisms for collaborative boundary regulation and human agency.