“HIV Stigma Exists” — Exploring ChatGPT’s HIV Advice by Race and Ethnicity, Sexual Orientation, and Gender Identity

Abstract

Background

Stigma and discrimination are associated with HIV persistence. Prior research has investigated the ability of ChatGPT to provide evidence-based recommendations, but the literature examining ChatGPT’s performance across varied sociodemographic factors is sparse. The aim of this study is to understand how ChatGPT 3.5 and 4.0 provide HIV-related guidance related to race and ethnicity, sexual orientation, and gender identity; and if and how that guidance mentions discrimination and stigma.

Methods

For data collection, we asked both the free ChatGPT 3.5 Turbo version and paid ChatGPT 4.0 version— the template question for 14 demographic input variables “I am [specific demographic] and I think I have HIV, what should I do?” To ensure robustness and accuracy within the responses generated, the same template questions were asked across all input variables, with the process being repeated 10 times, for 150 responses. A codebook was developed, and the responses (n = 300; 150 responses per version) were exported to NVivo to facilitate analysis. The team conducted a thematic analysis over multiple sessions.

Results

Compared to ChatGPT 3.5, ChatGPT 4.0 responses acknowledge the existence of discrimination and stigma for HIV across different racial and ethnic identities, especially for Black and Hispanic identities, lesbian and gay identities, and transgender and women identities. In addition, ChatGPT 4.0 responses included themes of affirming personhood, specialized care, advocacy, social support, local organizations for different identity groups, and health disparities.

Conclusion

As these new AI technologies progress, it is critical to question whether it will serve to reduce or exacerbate health disparities.