A Scoping Review of the Use of Artificial Intelligence (AI) and Large Language Models (LLMs) Related to LGBTQ+ Health: Implications for Social Work Practice and Research

BACKGROUND: The use of technology in social work practice and research is rapidly growing, particularly the utilization of Large Language Models (LLMs) like ChatGPT. LLMs are being employed in a variety of ways, such as providing evidence-based practice advice and mental health support. For the social work profession to keep pace with these changes, it is necessary to understand the current trends in LLM research across disciplines and the implications for social work research and practice. This scoping review focuses specifically on LLM research related to LGBTQ+ populations, who often face diverse and unique stressors.

METHODS: The scoping review searched for relevant papers in the PubMed, PsycINFO, and arXiv databases using keywords related to large language models (LLMs) such as "conversational Artificial Intelligence (AI)", “generative AI chatbots'', "AI driven chatbots”, and keywords related to LGBTQ+ populations such as "sexual and gender minority", "gay", "lesbian", "trans'', "queer", and "men who have sex with men (MSM)". After removing duplicates, the initial search yielded a total of 47 articles. The abstracts of these 47 articles were reviewed to identify the studies that met the eligibility criteria, which were: (1) published in English, and (2) empirical studies covering both LLMs and LGBTQ+-related topics. Of the 47 articles, seven met these eligibility criteria. To gain a comprehensive understanding of the current research trends, this review did not limit the search to peer-reviewed articles only. It also included preprint articles, in order to capture the latest developments in this emerging field.

RESULTS: The articles covered three main components related to LLMs and the LGBTQ+ community: (1) Evaluating existing biases in current LLMs; (2) Improving and developing LLMs targeted to LGBTQ+ communities; and (3) Experiences of LGBTQ+ individuals using LLMs. Half of the articles focused on ways to evaluate and benchmark anti-LGBTQ+ biases in existing LLMs. This included assessing issues like sexual identity stereotypes, pronoun usage, and biases towards transgender and gender non-conforming individuals. One study looked at enhancing LLMs to provide psychological support for LGBTQ+ youth, while another explored developing community-driven benchmarking tools. Additionally, one study examined the experiences of LGBTQ+ individuals using LLMs.

CONCLUSIONS: The findings indicate that there are implicit biases against LGBTQ+ individuals present in current LLMs. However, there are also efforts underway to improve LLMs and develop models tailored to serving LGBTQ+ communities. This suggests significant potential for social work practice and research to leverage LLMs to provide support and resources for LGBTQ+ individuals, as well as educate social work practitioners on understanding and addressing existing stigma. The implications highlight opportunities to dismantle LGBTQ+ oppression and offer more inclusive, affirmative care.