Evaluating the Experience of LGBTQ+ People Using Large Language Model Based Chatbots for Mental Health Support

Title of the Paper

Evaluating the Experience of LGBTQ+ People Using Large Language Model Based Chatbots for Mental Health Support

Paper Information

• Topic Area: Sociotechnical challenges of artificial intelligence in mental health support, specifically targeting applications for the LGBTQ+ community
• Keywords: Large Language Models (LLM), Chatbots, LGBTQ+, Mental Health, Gender Identity, Social Bias, Human-Computer Interaction, Technology Ethics

Research Background and Issues

• Identified Problems or Challenges:

  • The mental health status of LGBTQ+ individuals is significantly poorer compared to heterosexual and cisgender peers (e.g., higher rates of depression and suicidal ideation).
  • LGBTQ+ individuals face daily prejudice, discrimination, and social stigma, which hinder their access to traditional mental health support.
  • Although chatbots based on large language models (e.g., ChatGPT) provide immediate and discreet mental health support for the LGBTQ+ community, embedded biases within these platforms may reinforce stereotypes about LGBTQ+ people.
  • There is a lack of in-depth research on whether these chatbots can effectively and safely provide adequate LGBTQ+-specific mental health support.

• Importance of the Research:

  • As artificial intelligence evolves, these technologies offer potential support systems for marginalized groups, but their risks and limitations significantly impact users' mental health.
  • The mental health needs of the LGBTQ+ community are complex and specific, requiring better-adapted technological solutions.

• Research Motivation and Related Work:

  • Existing online support technologies and mental health applications have insufficient focus on the LGBTQ+ community.
  • While chatbots have made progress in providing psychological comfort, their role in addressing social biases and sensitive topics remains unclear.

Solution

• Research Methods:

  • Conduct interviews with 18 LGBTQ+ and 13 non-LGBTQ+ users to explore their experiences using LLM-based chatbots for mental health support.
  • Investigate how participants utilize these technologies and their advantages and shortcomings.

• Research Questions:

  1. How do LLMs provide value in supporting the mental health of LGBTQ+ individuals?
  2. Do LGBTQ+ individuals have different usage purposes compared to non-LGBTQ+ groups?
  3. Can LLMs meet the needs of LGBTQ+ individuals regarding topics related to their gender and identity?

• Implementation Steps and Techniques:

  • Conduct in-depth interviews and open coding analysis to extract key themes from participant descriptions.
  • Compare the chatbot usage experiences of LGBTQ+ and non-LGBTQ+ participants to examine specific sociotechnical issues.

Research Findings

• Key Findings:

  • LLM chatbots provide immediate support, high availability, and serve as safe spaces for users to engage in intimate conversations and practice social skills.
  • LGBTQ+ users utilize chatbots to explore their identities, rehearse coming-out experiences, and seek guidance on combating discrimination.
  • However, the information generated by these chatbots is often overly generic, failing to deeply understand LGBTQ+ individuals' personalized needs and sometimes offering misleading or dangerous advice.
  • LGBTQ+ participants noted that the lack of real-world support was a primary motivation for relying on chatbots, highlighting the significant impact of social biases on their need for AI-based support.

• Limitations and Future Directions:

  • Limitations:

    1. LLM chatbots lack an understanding of the complexities of LGBTQ+ identities, reflected in the absence of personalized responses and emotional resonance.
    2. Training data biased toward mainstream corpora introduces prejudices, neglecting the needs of minority groups.
    3. Overreliance on chatbots may delay access to professional mental health support.

  • Future Directions:

    1. Technical Optimization:
       • Implement dialogue safeguards with contextual information to avoid generating harmful advice on sensitive issues.
       • Perform more tailored fine-tuning of LLMs to better reflect the specific realities and needs of LGBTQ+ individuals.
       • Develop lightweight models for specific scenarios rather than further scaling general-purpose LLMs.
    2. Decentralized Technology Development:
       • Decentralize LLM development to give communities greater influence in the design of these technologies.
    3. Sociotechnical Solutions:
       • Improve inclusivity in digital spaces and educate key community roles (e.g., online platform administrators) to become LGBTQ+-friendly advocates.
       • Prioritize addressing social stigma and discrimination against LGBTQ+ individuals over mere technological optimization.

• Key Advantages and Unique Contributions:

  • This study demonstrates the potential of LLMs in providing mental health support for marginalized groups (e.g., LGBTQ+ individuals) while also uncovering the sociotechnical and ethical issues of these technologies.
  • The authors advocate for a combined approach of technological and social improvements to create a safer and more effective support environment for the LGBTQ+ community.