SeeSawBot: An LLM-Driven Chatbot Mediating Across Private and Shared Slack Channels to Support Team Dynamics

Paper Title

SeeSawBot: An LLM-Driven Chatbot Mediating Across Private and Shared Slack Channels to Support Team Dynamics

Publication Info

• Topic area: AI-mediated communication in team collaboration
• Keywords: LLM-driven chatbot, cross-space mediation, team dynamics, Slack, human-AI collaboration, boundary objects, emotional labor, group processes, computational mediators, teamwork

Background and Problem

• Problem / challenge: Existing computational mediators focus on when, what, and to whom interventions are directed, neglecting the where dimension of communication (private vs. public spaces). This gap limits their ability to address relational dynamics and spatial transitions in team collaboration.
• Significance: Understanding and designing for where mediation occurs is crucial for improving team dynamics, fostering inclusivity, and redistributing emotional labor in collaborative settings.
• Motivation and related work: Prior research has explored computational mediators for structuring group flow, personalizing interactions, and timing interventions. However, these systems are typically confined to either public or private spaces, failing to navigate cross-space interactions. This study builds on sociological theories of frontstage/backstage behavior and privacy boundaries to address this gap.

Solution

• Proposed approach: SeeSawBot, an LLM-driven chatbot that mediates across private DMs and public Slack channels to support team dynamics by bridging communication spaces.
• Novelty:
  1. Introduces cross-space mediation as a design variable for computational mediators.
  2. Demonstrates how AI can function as both a boundary object (shared artifacts) and a boundary actor (active participation and authority negotiation).
  3. Explores how spatial placement of interventions affects perceptions of autonomy, agency, and legitimacy.
  4. Provides design principles for adaptive, context-sensitive AI mediation.
• Procedure and key techniques:
  • Conducted a formative study (n=10) to inform SeeSawBot’s design.
  • Deployed SeeSawBot in 18 student Slack teams (n=105) over 8 weeks, collecting bi-weekly surveys and post-deployment interviews.
  • Iteratively refined SeeSawBot’s features (e.g., participation nudges, assignment-aware responses) based on user feedback.
  • Analyzed data using thematic analysis to understand cross-space information flows, relational roles, and temporal dynamics.

Results

• Concrete findings:
  • Cross-space mediation enabled distributed sense-making, personalized guidance, and inter-team translation.
  • SeeSawBot redistributed emotional labor by acting as a social gateway, participation balancer, social buffer, and reflective mirror.
  • Teams integrated SeeSawBot into their workflows, with its role evolving across group development stages (forming, storming, norming, performing, adjourning).
• Advantage over baselines:
  • Enabled seamless transitions between private and public communication spaces, addressing relational and spatial gaps overlooked by prior systems.
  • Supported both individual and team-level reflection, fostering inclusivity and reducing social risks in accountability.
• Experiments / evaluation:
  • Bi-weekly surveys (n=63, 51, 57, 52) captured longitudinal feedback on SeeSawBot’s impact on team dynamics.
  • Post-deployment interviews (n=9) provided in-depth insights into perceptions of authority, trust, and emotional labor redistribution.
  • Iterative design process incorporated user feedback to refine SeeSawBot’s features and intervention strategies.
• Limitations and future work:
  • Deployment emphasized naturalistic use, limiting controlled evaluation of specific design factors (e.g., spatial placement).
  • Occasional bugs and unintended behaviors due to LLM-based system variability.
  • Future work could include controlled studies to isolate design effects and expand SeeSawBot’s knowledge base for broader applications.

Summary

This paper introduces SeeSawBot, an LLM-driven chatbot designed to mediate across private and public Slack channels, addressing the overlooked spatial dimension of computational mediation. Through an 8-week deployment in student teams, the study demonstrated how cross-space mediation fosters distributed sense-making, redistributes emotional labor, and adapts to evolving team dynamics. SeeSawBot acted as both a boundary object and a boundary actor, influencing participation, authority, and emotional labor. Findings highlight the importance of spatial placement as a design variable and provide actionable principles for future cross-space computational mediators.