One question ran through the launch of the Oxford Global Health Seminar Series: how can AI be harnessed to improve global health without reinforcing existing inequalities?
Opening the session, Prof Caesar Atuire, Co-Director of Oxford Global Health, described the series as a space for researchers to exchange ideas, challenge assumptions and generate new collaborations around some of the most pressing global health challenges of our time.
The first seminar focused on AI, ethics and global health – a topic that reflects both the rapid pace of technological change and the complex questions that accompany it.
Making AI work for the people who need it most
AI is often presented as a solution to some of global health's most pressing challenges. But what happens when the data needed to build those systems are scarce, incomplete or unevenly distributed?
Prof Tingting Zhu explored the potential of AI to support healthcare delivery in resource-constrained settings, highlighting how AI systems can identify patterns in data and support decision-making. Drawing on examples from her work in countries including Vietnam and the Philippines, she demonstrated how digital tools and wearable technologies can help address challenges associated with limited healthcare infrastructure and scarce data.
Zhu also discussed approaches such as synthetic data generation, which can help researchers continue developing and testing models when real-world data are difficult to obtain. Throughout the discussion, she emphasised the importance of grounding technological innovation in local realities and practical needs.
"If a paper and pen works better, that is a better solution," Zhu said.
Can AI systems escape the inequalities embedded within their data?
Prof Angeliki Kerasidou examined the ethical dimensions of AI in global health, focusing on questions of bias, representation and fairness. Many AI systems are trained using datasets that disproportionately represent populations from high-income countries.
"The question is not whether AI works but who it works for," Kerasidou maintained.
What was clear from the two presentations is that data scarcity is both an engineering challenge and a question of justice.
Missing data often reflect populations and experiences that have historically been overlooked, meaning that tools that perform well for some groups may perform far less effectively for others. Questions of equity, trust and accountability, Kerasidou argued, are not peripheral concerns but central to how AI systems are designed, deployed and evaluated.
Participants also explored whether synthetic data can help address underrepresentation or risk reproducing existing biases at scale.
Respondents Dr Fanqi Zeng and Dr Munib Mesinovic helped draw connections between the two presentations, highlighting how questions of data scarcity, fairness and representation can be understood as both technical and ethical challenges. Their reflections helped open a wider discussion about who benefits from AI and whose priorities shape its development.
Beyond technology: whose problems are we trying to solve?
Building on Kerasidou's reflections on ‘techno-solutionism’, the discussion raised questions about power and priorities. Who decides which health challenges deserve AI solutions, and how are research agendas shaped by funding, data availability and commercial incentives?
While AI offers exciting possibilities, participants questioned whether technological solutions always address the issues that communities themselves consider most urgent. If local priorities are access to clean water, healthcare infrastructure or workforce capacity, then sophisticated algorithms may not necessarily provide the most effective solution.
The discussion turned to the growing role of industry in shaping AI development. While commercial investment has accelerated innovation, participants noted that commercial incentives can influence which problems receive attention and how technologies are presented. This makes it critical to ask not only whether a tool works, but whose interests it serves and what problems it is designed to address.
Trust, responsibility and the future of AI in global health
Questions of explainability, accountability and governance surfaced repeatedly throughout the discussion. Alongside technical questions about how AI systems make decisions were broader concerns about responsibility, transparency, public trust and who should determine whether an AI system is ready for deployment.
Zhu highlighted the importance of understanding the uncertainty of AI systems and recognising their limitations. Rather than treating AI as a replacement for human judgement, she argued that developers should design systems that communicate confidence levels and support informed decision-making.
Questions of responsibility remain unresolved. When an AI system contributes to a healthcare decision, who is accountable if something goes wrong? The developer, the healthcare provider, the institution deploying the technology or the system itself? As AI becomes increasingly embedded within healthcare systems, such questions will become ever more important.
AI is entering a world already marked by inequality and unequal access to resources. The challenge is not to assume that it will solve those problems, but to ensure that it does not exacerbate them.
The inaugural seminar highlighted how questions about AI in global health extend far beyond technology. By bringing together perspectives from engineering, ethics, sociology and global health, the discussion generated as many new questions as answers.
About the Oxford Global Health Seminar Series
The Oxford Global Health Seminar Series brings together researchers from across disciplines, divisions and career stages to explore some of the most pressing challenges in global health. Designed as a discussion-led forum rather than a traditional lecture series, it aims to create space for new ideas, critical reflection and collaboration, while ensuring that early-career researchers play an active role in shaping the conversation.