Ana Lorena Guerrero Torres

This session showcased the potential transformative impact of artificial intelligence (AI) across diverse settings/contexts. Speakers presented projects where AI contributed to improved HIV care and service delivery by reaching underserved populations for HIV through testing or screening of NCDs, improving referrals and support to healthcare providers, and optimizing drug combinations.

Doris Macharia from Orbis International in South Africa shared their experience implementing AI tools such as Cybersight AI and HealthPulse AI. Cybersight AI, a non-invasive, open-access tool for detecting eye diseases in persons with HIV, has reduced screening time and improved referral follow-through rates in Rwanda and Vietnam. Zhao Ni from Yale presented the development of AI-Chatbot to enhance HIV testing among vulnerable populations in Malaysia. Iterative process involved stakeholder engagement and feedback to refine script and algorithm. Despite ongoing evaluations, the project demonstrates AI's potential to provide real-time access to HIV services, especially in communities with high stigma. Peter Wang from National University of Singapore showcased an AI platform designed to optimize drug combinations, crucial for combating AMR. This platform rapidly identifies effective drug combinations and its dose ratios. Discussion included tailoring AI algorithms to community needs, ethical considerations, privacy concerns, and importance of integrating AI tools into health systems. The speakers/audience highlighted how AI tools need to consider/involve communities, their needs throughout their development and implementation. Finally, even when these AI tools may be open-access and easily available, decision-makers need to invest in other components: “If you don’t have the appropriate camera, the [AI tool] won’t work”.

This session highlights the transformative potential of AI in reaching vulnerable people and achieving HIV treatment goals. However, to fully achieve this, we must address some important contexts throughout the development, implementation and evaluation of these AI tools. These include: (1) all AI tools must ensure privacy concerns are addressed so that vulnerable populations are not placed at increased risk of violence, stigma or discrimination; (2) communities must be included throughout as this allows tailoring tools to their specific needs and enhances trust and acceptability; and (3) AI success requires interdisciplinary collaboration, particularly when working within the whole health system.