"Chat, Should I Leave Him?" Risks, Rewards, and Roles for AI in Relationship Advice

Paper Title

'Chat, Should I Leave Him?' Risks, Rewards, and Roles for AI in Relationship Advice

Publication Info

• Topic area: Use of generative AI for relationship advice and its implications.
• Keywords: generative AI, relationship advice, psychosocial AI, digital intimacy, human-AI interaction, AI safety, social support, folk theories, co-optation tactics, sociotechnical research.

Background and Problem

• Problem / challenge: The use of generative AI for relationship advice is growing, but its implications for interpersonal dynamics, social support systems, and user well-being are not well understood. Existing research has focused on acute crises like suicidality but has not addressed long-term relational effects or interpersonal risks.
• Significance: Understanding how users interact with AI for relationship advice could inform the design of safer, more effective AI systems that support healthier relationships while mitigating risks like overreliance, misinformation, and social isolation.
• Motivation and related work: Prior research in digital intimacy and psychosocial AI has examined how technology reshapes personal connections, but there is limited empirical work on AI-mediated relationship advice. This paper builds on studies of social support, advice quality, and human-AI interaction to explore this understudied phenomenon.

Solution

• Proposed approach: A qualitative exploratory study examining how people use generative AI for relationship advice, the roles they envision for AI, the risks and benefits they navigate, and the folk theories and tactics they employ.
• Novelty:
  1. A conceptual model and typology of roles users envision for AI in relationship advice (e.g., Interpreter, Judge, Self-Critic).
  2. A risk/benefit analysis of user-AI interactions across these roles.
  3. Identification of folk theories about AI and co-optation tactics users employ to achieve their goals.
  4. Design recommendations for AI systems to support healthier digital intimacies.
• Procedure and key techniques:
  • Recruitment of 25 participants for a questionnaire and 17 for in-depth interviews.
  • Thematic analysis of 90 prompts provided by participants and their interactions with AI.
  • Member checking to validate findings and refine themes.
  • Analysis focused on roles for AI, social support sought, and risk-benefit tradeoffs.

Results

• Concrete findings:
  • Users envision nine roles for AI in relationship advice: Interpreter, Listener, Judge, Self-Critic, Teacher, Therapist, Strategist, Editor, and Executive.
  • AI interactions span three goals: understanding situations, preparing to act, and executing decided actions.
  • Users navigate risks like overreliance, misinformation, and social isolation alongside benefits like affirmation, emotional regulation, and improved communication.
  • Folk theories include beliefs that AI is sycophantic, can access diverse perspectives, and can synthesize hidden truths from data.
• Advantage over baselines:
  • Provides a detailed typology of AI roles and user strategies not previously documented.
  • Highlights nuanced interpersonal risks and benefits beyond acute crises.
• Experiments / evaluation:
  • Data collected through questionnaires and two rounds of interviews.
  • Analysis of user-provided prompts and interview transcripts to identify themes.
  • Member checking to ensure findings align with participant experiences.
• Limitations and future work:
  • Findings may not generalize to all contexts or populations; participants were primarily "power users."
  • AI non-users and child safety were outside the study's scope.
  • Future work could explore broader populations, other contexts (e.g., sex advice), and societal impacts of AI-mediated social support.

Summary

This paper explores how people use generative AI for relationship advice, identifying nine roles AI plays and the risks and benefits users navigate. It highlights users' folk theories about AI and the strategies they employ to achieve their goals. The study provides a conceptual model, risk/benefit analysis, and design recommendations for AI systems to support healthier digital intimacies. While the findings emphasize the potential of AI to assist users, they also underscore the need for safeguards against risks like overreliance and misinformation. This work lays the foundation for future research on AI's role in reshaping social support systems and interpersonal dynamics.