AI chatbots show promise for HIV prevention

Recent research into the potential of AI chatbots to bolster HIV prevention revealed that the tools can provide accurate and accessible information to users, but have limitations in terms of response personalization.

The analysis, conducted by the California HIV/AIDS Policy Research Centers in conjunction with the University of California, Berkeley School of Public Health, assesses the ability of four widely-used chatbots -- ChatGPT-4o, Claude 3.5 Sonnet, Gemini 1.5 Flash and Meta AI Llama 3.1 -- to provide high-quality, tailored responses for users with diverse HIV prevention needs.

The researchers prompted each chatbot with six common questions related to HIV prevention, including transmission, testing, pre-exposure prophylaxis (PrEP) and treatment referrals to evaluate response quality. To gauge each model's ability to tailor its responses, the research team created and incorporated three personas related to transgender status, literacy level and location into the prompts.

Doing so helped researchers examine how well each chatbot could deliver accurate, personalized and engaging health information to individuals from groups disproportionately impacted by HIV.

The analysis revealed that the chatbots provided highly accurate, tone-neutral responses across multiple HIV prevention topics, but these responses varied in clarity. The research team indicated that some of the responses did not follow best practices for use of nonstigmatizing language, while others had disjointed flows or unclear conclusions.

In terms of personalization and engagement, the chatbots were able to simplify their responses when asked to do so, but struggled to tailor those responses to the needs of specific populations and users.

The researchers highlighted that these findings underscore the opportunity for chatbot integration in public health services. They noted that when the tools' responses are reviewed and tailored by health professionals, chatbots could be valuable for enhancing the quality and efficiency of HIV prevention, care services and education.

"New and innovative ways to enhance HIV care and prevention efforts are needed, especially to reach younger, tech-savvy groups who may turn to digital solutions for health information," stated study lead Marisa Fujimoto of UC Berkeley, in a press release. "Based on our results, we are cautiously optimistic about the use of AI chatbots for HIV prevention by individuals from communities affected by HIV, community organizations, and health providers. Chatbots are capable of providing reasonably accurate information with few access barriers and could be used best in conjunction with advice from health professionals to optimize information and provide referrals to services."

"Nevertheless, our research also raises important questions about how to ensure that AI chatbots provide inclusive guidance that addresses the needs of communities disproportionately affected by HIV, like those seeking gender-affirming care," Fujimoto cautioned.

Shania Kennedy has been covering news related to health IT and analytics since 2022.