Rehearsal: Simulating Conflict to Teach Conflict Resolution

Document Title

Rehearsal: Simulating Conflict to Teach Conflict Resolution

Document Information

• Subject Area: Conflict Resolution and Human-Computer Interaction
• Keywords: conflict resolution, large language models, Interests-Rights-Power (IRP) framework, teaching tools, interpersonal skills, simulated role-playing, generative AI, feedback mechanisms, social computing
• URL: https://doi.org/10.1145/3613904.3642159

Research Background and Problem

• Problem and Challenges: Interpersonal conflicts are inevitable in society. Although cooperative communication for conflict resolution is considered the ideal outcome, developing this skill requires targeted, repeated practice and timely feedback. However, existing training methods are often limited to static learning materials (e.g., case studies) and lack dynamic guidance and practical resources. While expert role-playing is effective, it is costly and resource-intensive.
• Significance: Designing effective, low-cost, and scalable solutions through simulated role-playing to provide more people with practice opportunities equivalent to expert-level conversations could significantly enhance conflict resolution skills, thereby improving collaboration and productivity in workplace and personal relationships.
• Motivation and Related Work: The aim is to leverage rapidly advancing generative large language models (LLMs) to create realistic conflict simulations. However, current LLMs have limitations, such as being overly accommodating to users, providing insufficiently targeted feedback, and lacking guidance in open-ended text generation.

Solution

• Proposed Method:

  1. Developed an interactive system based on generative large language models, named Rehearsal, which teaches users conflict resolution techniques through simulated dialogue scenarios.
  2. Introduced a novel IRP Prompting technique, anchoring dialogue content in the classic Interests-Rights-Power (IRP) framework, a foundational theory in conflict resolution.
  3. Provided real-time feedback and alternative strategy generation, allowing users to explore different dialogue paths and learn from them.

• Innovations:

  • The system applies the IRP framework to constrain LLMs, generating dialogue content aligned with conflict resolution theory, addressing issues of LLMs being overly "compliant" or providing insufficiently targeted feedback in conflict scenarios.
  • The system supports generating "hypothetical" dialogue paths, enabling users to evaluate and observe the effects of different conflict strategies.
  • Designed experiments to assess users' real-world performance rather than merely optimizing theoretical knowledge retention.

• Implementation Steps and Key Techniques:

  1. IRP Prompting: The system employs a multi-step prompting pipeline to sequentially complete message classification, strategy planning, alternative strategy generation, and corresponding response generation.
  2. Conflict Simulation and User Interaction: Users can select predefined conflict scenarios or customize scenarios based on real-life conflicts for simulated dialogues.
  3. Real-Time Feedback: The system provides feedback based on dialogue content, allows users to compare and select alternative strategies, and rates the effectiveness of conflict resolution strategies.
  4. Validation through Experiments and Theory: The system's accuracy and teaching effectiveness were validated through technical evaluations and behavioral studies.

Research Outcomes

• Specific Outcomes:

  • Introduced the Rehearsal system and IRP Prompting technique, demonstrating their potential in generating high-quality conflict simulations and role-based dialogues.
  • Achieved significant improvements in users' actual conflict resolution performance: compared to a control group using traditional training materials, the experimental group using Rehearsal reduced the use of adversarial strategies (Rights and Power) by 67% and doubled the use of cooperative strategies (e.g., Interests and Proposals).
  • The system significantly enhanced users' practical skills in real-world conflict situations, whereas traditional training primarily strengthened static strategy cognition.

• Advantages over Existing Solutions:

  • Eliminates the need for costly human experts in simulated conflict role-playing, offering a low-cost, scalable solution.
  • Provides dynamic and targeted feedback using generative AI, closely aligning with the practical needs of conflict resolution rather than remaining at the theoretical level.
  • Experiments indicated that 35% of users who only watched videos or read materials still struggled with implementation in real-world conflict resolution, while Rehearsal effectively supported practical application.

• Experimental or Evaluation Results:

  1. Among 40 experimental participants, the group using Rehearsal demonstrated significantly higher use of cooperative strategies (3.0 vs. 1.5) and a notable reduction in competitive strategies.
  2. The IRP Prompting technique outperformed control generation mechanisms in classification accuracy (82%) and ecological validity of generated content.
  3. Although there was no significant difference in "textbook knowledge" (e.g., strategy identification and recall) post-practice, users' "practical performance" showed greater effectiveness.

• Limitations and Future Directions:

  • Limitations:
    • The applicability of the IRP framework is limited and may not cover all cultural contexts or specific conflict scenarios, such as those involving danger or mistrust.
    • Simulated dialogue content may be perceived as "scripted" due to generative AI's tendencies (e.g., overly mechanical or unnatural tone).
    • Long-term effects have not been definitively validated (e.g., transferability over time requires further study).
  • Future Directions:
    • Introduce multi-party conflict simulations to adapt the system for complex team interactions and mediation.
    • Consider multi-modal extensions, incorporating audio or video interaction formats.
    • Deepen analysis of frameworks adapted to diverse cultural backgrounds.