"Comforting and Small Like a House Cat, Big and Intimidating Like a Bodyguard": How Women Perceive and Envision AI Companions as a New Harassment Mitigation Approach in Social VR

Research Background and Issues

• Issues and Challenges: This paper focuses on the phenomenon of harassment faced by women in social virtual reality (Social VR), particularly how the embodied and immersive nature of harassment exacerbates its harm, such as virtual "touch" or invasion. These harassment behaviors occur frequently, and existing safety mechanisms (e.g., blocking, muting, reporting, and spatial protection) have limitations.
  • Limitations of Existing Methods:
    1. They exacerbate power dynamics inequality for women on the platform, requiring them to rely on male allies for protection.
    2. Lack of informational, social, and emotional support for women who experience harassment.
    3. Undermine women's identity representation and immersive experiences in Social VR.
• Research Importance: As Social VR becomes an integral part of digital social life, harassment may further escalate, particularly impacting women and other marginalized groups. There is an urgent need for new methods to mitigate harassment.
• Research Motivation and Related Work: Existing studies primarily focus on AI content moderation (e.g., automated reporting systems) but lack in-depth exploration of how AI can address the unmet multidimensional needs (such as emotional support) in harassment mitigation. AI companions, as a non-human form of interaction, have potential protective and supportive functions.

Solution

• Methods and Solutions: The authors propose designing AI as "companions" for female Social VR users to address the limitations of existing safety mechanisms, including four AI companion roles:
  1. Accessible Companions: Provide companionship and protection for female users at any time and in any scenario.
  2. Informational Companions: Offer knowledge and guidance on how to identify and handle harassment incidents.
  3. Emotional Support Companions: Help women recover from the internal harm caused by harassment incidents.
  4. Protective Companions: Provide automatic intervention mechanisms, such as blocking harassment behaviors or taking functional actions against harassers.
• Innovations:
  1. AI is not only a traditional tool for "correction or punishment" but also addresses women's needs for emotional support and social connection.
  2. Proposes new roles for AI companions in preventing harassment, healing trauma, and enhancing experiences, surpassing the singular monitoring function.
  3. Centers on female users, emphasizing their agency and personalized experiences.
• Implementation Steps and Key Technologies:
  1. Conduct in-depth interviews with 20 female Social VR users to gather their visions and needs for AI companions.
  2. Use qualitative coding and thematic analysis to identify core themes.
  3. Develop design principles based on the actual needs of female users and a comprehensive reflection on existing harassment mitigation mechanisms.

Research Outcomes

• Specific Findings:
  1. The study reveals how female users believe AI companions can surpass existing mechanisms to provide unmet protection and support.
  2. Proposes specific capabilities of AI companions envisioned by women and their application scenarios in Social VR.
  3. Introduces three high-level design principles to guide future AI companion design:
     • Design AI companions to promote human emotional connections based on women's agency.
     • Create AI companions that do not appear "foreign" and enhance user comfort.
     • Carefully design AI functionalities to prevent misuse for harassing other users.
• Advantages and Comparisons: Compared to existing harassment solutions, AI companions not only detect and punish harassment behaviors but also fill gaps in emotional and social support while mitigating the negative impacts of current mechanisms.
• Experiment and Evaluation Results: The study, through interviews, reveals female users' specific expectations for AI companion roles, demonstrating the significant potential applicability of the proposed solution.
• Limitations and Future Directions:
  1. Concerns about the "sociality" of AI companions and whether reliance on AI might weaken the quality of real interpersonal relationships.
  2. AI may perform differently across cultural and social contexts, necessitating cross-cultural research.
  3. Technical complexity in ensuring AI can accurately understand social contexts and dynamically adjust functionalities.
  4. Future exploration is needed on how to better integrate AI with existing harassment mitigation mechanisms (e.g., human monitoring and AI content moderation).

Conclusion

The study explores how AI can serve as companions for female Social VR users to address harassment issues, proposing four AI roles to meet unmet protection and support needs. Through analysis of female users' perspectives and real-world experiences, three design principles are introduced to support future AI companion development. This research not only enriches the understanding of AI applications in Social VR but also provides constructive recommendations for harassment mitigation for marginalized groups.