On 25 February, the Alliance for Health Policy and Systems Research, in collaboration with AcademyHealth, convened a global webinar to examine how artificial intelligence (AI) is reshaping health policy and systems research (HPSR).
The session moved beyond abstract debates about technological disruption and focused instead on how AI is already being applied across the research lifecycle – from evidence synthesis and data analysis to national health system management and workforce development.
Opening the event, Michael Gluck, Vice President at AcademyHealth, underscored the shared mission of both organizations, noting that “better evidence leads to better health,” but that too often research remains disconnected from the decisions that shape people’s lives. He emphasized that the emergence of AI presents an opportunity for global exchange and collaboration across institutions facing similar pressures.
Kumanan Rasanathan, Executive Director of the Alliance, acknowledged that conversations about AI often oscillate between optimism and alarmism. There are those who see AI as a pathway to access and equity, he observed, and those who fear dystopian outcomes. Yet, he stressed, the reality is that societies – and health systems – are already undergoing rapid transformation. While much attention has focused on AI in service delivery, he pointed out that far less attention has been given to how AI is being used through, and even collapsing, the whole research cycle. For HPSR, this shift is particularly significant.
Diving in to concrete examples of how AI is already shifting the practice of HPSR, Heather Ames, a Senior Researcher at the Norwegian Institute of Public Health (NIPH), offered a detailed account of how machine learning tools were introduced at NIPH into systematic review workflows during the COVID-19 pandemic. Faced with an explosion of new evidence and urgent policy demand, her team sought “to produce evidence faster without compromising quality”.
Using off-the-shelf tools to support study screening, prioritization and risk-of-bias assessment, the team gradually embedded machine learning into routine practice. The most striking change, however, was not simply speed. AI altered how teams worked. Review stages became more fluid. Teams worked more intensively and in parallel. Timelines shifted, requiring new forms of communication with commissioners.
Crucially, this transformation depended on institutional support. Ames noted that leadership provided protected time for experimentation and accepted that not every innovation would succeed – an approach that allowed teams to build trust in the tools while maintaining methodological rigour.
Responding to her presentation, Amirhossein Takian, Senior Scientist at the Alliance, reflected that the experience demonstrated something broader: integrating AI is not just a technical upgrade but an organizational change process and ultimately a governance decision. For HPSR institutions, the question is therefore not only whether an algorithm performs well, but whether the integrity and interpretive depth of research are preserved.
William Hoyos Arango, Director of Strategic Health Monitoring and Evaluation at the Ministry of Health in El Salvador, brought the discussion into the realm of national policy and implementation. El Salvador recently adopted an AI law and is piloting tools for lung nodule detection, stroke diagnosis and mammography screening across public hospitals. Beyond clinical use, Hoyos described how AI systems are enabling administrators to query national health databases using natural language rather than specialized coding – expanding access to system-level data analysis.
At the same time, he was clear that innovation must be accompanied by oversight. AI in health, he argued, should be treated and evaluated like any other health technology, with careful attention to effectiveness, safety and equity. The country chose to act early on regulation not because it had mastered the field, but because it recognized the risks and wanted to establish safeguards from the outset.
The session itself offered a small demonstration of innovating with AI in practice. The Alliance piloted a live, low-cost AI interpretation tool, developed with the help of AI, between English and Spanish during the webinar. Responding to Hoyos’ presentation, Keisuke Nakagawa, Director of Strategic Impact and Growth at the University of California at San Diego Health’s Jacobs Center for Health Innovation noted that he had followed the remarks using the AI interpretation feature and described it as “such a living example in real time of how technology is allowing us and enabling us to collaborate globally”.
For Nakagawa, the substance of Hoyos’ remarks was equally striking: “we are having the exact same conversations in San Diego, California, as you are having in El Salvador”. The tools may differ and the infrastructure may vary, but the underlying questions – about workforce readiness, evaluation, governance and equity – are shared.
Bianca Frogner, Director of the Center for Health Workforce Studies at the University of Washington shifted the focus to the research workforce itself. AI is increasingly being used to assist with coding, statistical translation across platforms, debugging and drafting manuscripts. These applications can reduce time spent on repetitive technical tasks and potentially shorten the path from analysis to publication.
Yet she cautioned that expanded data volumes and automated text generation bring new risks. Concerns about data integrity, uneven access to computational resources and the responsible use of generative tools are becoming part of everyday research practice. Ultimately, she reminded participants that AI remains a tool, and that people retain responsibility for interpretation and judgement.
Responding from the perspective of health systems research in China, Roman Xu, Professor of Global Health and Health Systems of Southern Medical University, argued that the implications for HPSR extend beyond technical skills. Training researchers to use AI tools is important, he suggested, but so is building the capacity to evaluate and govern them. Health policy and systems researchers must be able to ask how algorithms perform in local populations, how bias is monitored over time and how AI systems are integrated into broader service delivery strategies.
Overall, the conversation showcased how artificial intelligence is already shaping how evidence is generated, synthesized and translated into policy. Research institutions are adjusting workflows. National health systems are experimenting with new analytical tools. Researchers are renegotiating what counts as authorship, expertise and methodological rigour.
For HPSR, this moment is less about technological adoption and more about direction-setting. The field has long grappled with questions of equity, accountability and context. AI introduces new tools into that landscape, but it does not change those guiding principles. However, it does present the possibility of transforming who is able to undertake HPSR and where this is done, potentially expanding its reach and impact.
In her closing reflections, moderator Ivor Horn underscored the leadership dimension of AI adoption, pointing to the importance of governance that creates “safe work environments” and asking what it means to allow teams to test and learn as new tools are introduced.
The webinar closed with a recognition that the choices being made now – in research teams, ministries of health and academic institutions – will shape how AI ultimately functions within health systems. For the HPSR community, the work ahead lies in ensuring that innovation strengthens the link between evidence and decision-making. As part of that effort, the Alliance is currently developing a manual on the responsible use of artificial intelligence in health policy and systems research, aimed at supporting institutions to navigate questions of quality, equity and governance in a rapidly evolving landscape. The question is not whether AI should or will be used for HPSR, as this is a given – rather, it is which elements must remain human-led, what constitutes acceptable and equitable use, and how can AI enable HPSR to have greater impact on improving health and health equity.