Advancing the democratization of generative artificial intelligence in healthcare: a narrative review

Background and Objective: The emergence of ChatGPT-like generative artificial intelligence (GenAI) has dramatically transformed the healthcare landscape, bringing new hope for the democratization of artificial intelligence (AI) in healthcare-a topic that has not been comprehensively reviewed. This review aims to analyze the reasons propelling the democratization of healthcare GenAI, outline the initial evidence in the literature, and propose future directions to advance GenAI democratization. Methods: We conducted a deep literature search for GenAI studies using Google Scholar, PubMed, ChatGPT, Journal of the American Medical Association ( JAMA ), Nature, , Springer Link, and Journal of Medical Internet Research (JMIR). JMIR ). We performed an abstraction analysis on the nature of GenAI versus traditional AI and the applications of GenAI in medical education and clinical care. Key Content and Findings: (I) A detailed comparison of traditional and GenAI in healthcare reveals that large language model (LLM)-based GenAI's unprecedent general-purpose capabilities and natural language interaction ability, coupled with its free public availability, make GenAI ideal for democratization in healthcare. (II) We have identified plenty of initial evidence for GenAI democratization in medical education and clinical care, marking the start of the emerging trend of GenAI democratization in a host of impactful applications. Applications in medical education include medical exam preparation, medical teaching and training, and simulation. Applications in clinical care include diagnosis assistance, disease risk prediction, new generalist chatbots, treatment decision support, surgery support, medical image analysis, patient communication, physician communication, documentation automation, clinical trial automation, informatics tasks automation, and specialized or custom LLMs. (III) Responsible AI is essential for the future of healthcare GenAI. National initiatives and regulatory efforts are working to ensure safety, efficacy, accountability, equity, security and privacy are built into healthcare GenAI. Responsible GenAI requires a human-machine collaboration approach, where AI augments human expertise rather than replaces it. Conclusions: The democratization of GenAI in healthcare has just begun, driven by the nature of GenAI and guided by the principle of human-machine collaboration. To further advance GenAI democratization, we propose three key future directions: integrating GenAI in medical education curricula, democratizing GenAI clinical evaluation research, and building learning health systems (LHS) with GenAI for system- level enforcement of democratization. Democratizing GenAI in healthcare will revolutionize medicine and significantly impact care delivery and health policies.