Friend, Foe, or Bot? Exploring Intergroup Dynamics in Hybrid Human-Bot Teams

Paper Title

Friend, Foe, or Bot? Exploring Intergroup Dynamics in Hybrid Human-Bot Teams

Publication Info

• Topic area: Intergroup dynamics in hybrid human-bot teams during competitive interactions.
• Keywords: Human-AI collaboration, hybrid teams, intergroup bias, bot transparency, social dynamics, reciprocity, imitation, spillover, competitive contexts, HCI.

Background and Problem

• Problem / challenge: Limited understanding of how AI teammates influence intergroup dynamics, particularly in competitive settings, and how transparency about bot identities affects social biases and accountability.
• Significance: Understanding these dynamics is critical for designing socially responsible hybrid teams, as AI integration can redistribute accountability and affect trust, cohesion, and fairness.
• Motivation and related work: Prior research has focused on ingroup dynamics and task performance with AI teammates, but little is known about intergroup interactions, biases, and the effects of bot transparency. Existing studies show that bots can trigger human-like prejudice or alter perceptions of human teammates, but the mechanisms in competitive hybrid teams remain unclear.

Solution

• Proposed approach: The study introduces StarHarvest, an online multiplayer game where hybrid teams of humans and bots compete for resources, with bot identities either concealed or revealed. Bots exhibit prosocial or antisocial behaviors to study their impact on intergroup dynamics.
• Novelty:
  1. Systematic comparison of concealed vs. revealed bot identities in competitive hybrid teams.
  2. Analysis of social mechanisms (reciprocity, imitation, spillover) shaping human-bot interactions.
  3. Examination of how bot behavior influences post-game evaluations and resource allocations.
  4. Insights into trade-offs between team cohesion and accountability in hybrid teams.
• Procedure and key techniques:
  1. Participants (240 recruited via Prolific) played StarHarvest in teams of one human and one bot, competing against another hybrid team.
  2. Bots followed one of four strategies: ingroup-biased, outgroup-biased, prosocial, or antisocial.
  3. Conditions varied by bot transparency: "aware" (bots explicitly identified) or "unaware" (bots concealed as humans).
  4. Data collected included in-game actions (locking/unlocking), post-game evaluations (Likert-scale ratings), and bonus point allocations.
  5. Statistical analyses (beta regression, ordinal logistic regression) tested the effects of bot behavior and transparency on social dynamics.

Results

• Concrete findings:
  • Lock rates were higher for opponent bots than humans in the "unaware" condition (30% vs. 24%) but equalized when bots were revealed.
  • Concealed bots fostered smoother coordination but led to spillover effects, where aggression was misdirected at uninvolved teammates.
  • Revealed bots were treated as secondary actors, reducing their influence within teams but triggering human-human retaliation in response to bot-bot interactions.
  • Post-game evaluations and bonus allocations were influenced by bot behavior only when identities were hidden.
• Advantage over baselines:
  • The study highlights how bot transparency reshapes intergroup biases, accountability, and team dynamics, offering insights beyond prior work focused solely on ingroup collaboration.
• Experiments / evaluation:
  • Game design ensured competitive and cooperative interactions, with clear visual feedback on actions and performance.
  • Statistical models tested reciprocity, imitation, and spillover mechanisms across conditions.
  • Post-game measures captured evaluations of performance, collaboration, competitiveness, and harm intention.
• Limitations and future work:
  • Limited demographic diversity (Western participants) and short-term interactions.
  • Simplified game rules and fixed bot strategies may not generalize to complex real-world settings.
  • Lack of verbal communication restricted opportunities for negotiation and conflict resolution.
  • Future studies should explore multi-human/multi-bot teams, long-term interactions, and richer communication channels.

Summary

This study explores how hybrid human-bot teams influence intergroup dynamics in competitive settings, focusing on the role of bot transparency. Concealed bots enhanced team coordination but caused spillover effects, while revealed bots reduced cohesion and accountability but redirected aggression to human partners. The findings highlight trade-offs between cohesion and accountability in hybrid teams, emphasizing the need for careful design to balance social and ethical considerations. These insights are relevant for designing responsible human-AI collaboration in competitive and high-stakes contexts.