Leveraging the Unique and Essential Role of Participatory Social Work Research in Artificial Intelligence (AI) for LGBTQ+ Health

BACKGROUND: As technology evolves, there is growing interest in leveraging large language model (LLM) artificial intelligence (AI) chatbots to address health inequities, especially among LGBTQ+ communities. While AI chatbots offer unique opportunities for social work researchers and practitioners to promote health equity, the development and implementation of AI chatbots require careful consideration as challenges persist in leveraging supervised learning techniques, including data/algorithm biases, privacy concerns, and the need for culturally congruent and ethically sound AI solutions. Social work is uniquely positioned to address these challenges due to its theoretical, methodological, and practical emphasis on understanding individuals within the context of their environment, community engagement and participatory approaches, interdisciplinary perspectives, and social justice and equity. This study explored how an expert and accountability panel helped to address the challenges associated with supervised learning in assessing LLM AI chatbots for LGBTQ+ health.

METHODS: We convened an expert and accountability panel comprising 8-12 scientists, service providers, and community members to address our study's challenges associated with supervised learning. The panel was carefully selected to ensure diverse perspectives and insights, including individuals with lived experience and expertise in artificial intelligence, LGBTQ+ health, social work research and practice, and community advocacy. Through a series of structured meetings, member checking, and feedback sessions the panel engaged in collaborative discussions, critique, and reflection in assessing LLM AI chatbots for LGBTQ+ health.

RESULTS: The expert and accountability panel’s diverse perspectives, expert insight, and feedback helped identify usability issues, content gaps, and potential unintended consequences of AI technology. This included biases (e.g., detecting when misinformation about LGBTQ+ was included in the training corpus), cultural nuances (e.g., intersectionality), and ethical considerations (e.g., handling of urgent situations/prompts such as suicidality; privacy/accidental “outing”) that needed to be addressed to enhance the accuracy, relevance, and cultural sensitivity of chatbot responses related to the specific needs and experiences of LGBTQ+ individuals. The panel's accountability aspect also ensured the research team was responsive, transparent, and ultimately accountable to the LGBTQ+ community.

CONCLUSIONS: The results highlight the value of participatory approaches (i.e. an expert and accountability panel) in assessing, developing, and testing LLM AI chatbots for LGBTQ+ health. By harnessing social work panel members' collective expertise and lived experiences, we addressed complex challenges, creating more responsive, culturally sensitive, and effective processes in AI assessment to improve the development and testing of digital health interventions for the LGBTQ+ community. Implications for research and practice are discussed.