Centering Equity in Health AI: LLM Design Strategies for Impactful Solutions (Preprint)

Le résumé fourni par la source

UNSTRUCTURED This article addresses the growing need to incorporate health equity considerations into the design and development of large language models (LLMs). While existing guidelines for artificial intelligence (AI) development often focus on ethical and technical aspects, they frequently overlook health equity, risking the potential for LLMs to exacerbate existing health disparities, or limiting LLM impact in promoting health equity. We propose a novel framework of design strategies that explicitly promote health equity through LLMs and LLM-supported health solutions. This framework aims to provide actionable guidance for developers, researchers, policymakers, and others to integrate health equity considerations into every stage of the LLM development process. Through a literature scan of work across academic, industry, and government sources, we developed an outline of key strategies to promote health equity in LLMs. This article presents a comprehensive framework of eight design strategies that can be used to promote health equity through LLMs. By integrating these strategies into the LLM design process, developers can design LLMs to improve health outcomes for all, while mitigating the risks of perpetuating existing inequities. The framework also highlights the importance of collaboration among stakeholders, including healthcare providers, patients, community members, and researchers, to ensure that LLMs are developed and deployed in a manner that is equitable and beneficial for all. LLMs hold immense promise for transforming healthcare and improving health outcomes. However, ongoing research, collaboration, and a commitment to developing robust metrics will be needed to thoroughly assess the impact of LLMs on health equity. While the proposed framework provides a starting point for this work, future research should focus on evaluating the effectiveness of these strategies, and refining them to further promote their implementation.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.