AI in healthcare risks could exclude 5 billion people; here’s what we can do about it

Artificial intelligence (AI) is rapidly transforming healthcare. From early cancer detection to personalized treatment recommendations, AI systems seek to make medicine faster, more precise and more affordable.

In high-income countries, these tools are already being tested in hospitals, research centres and clinics, and the results are impressive. Yet, for most of the world’s population, nearly 5 billion people living in low and middle-income countries, the benefits of medical AI remain out of reach.

Instead of narrowing health inequality, current approaches to AI development risk widening it. Most AI health systems are built on data sourced from a narrow set of populations, which means billions of people, mainly from the Global South remain largely invisible in diagnostic models, risk assessments and treatment algorithms.

Without corrective action, AI may inadvertently reinforce existing structural inequities in global health rather than helping to overcome them.

It is well known that data is central to AI development. In healthcare, this typically refers to electronic health records, imaging scans, genomic information or biometric signals collected from millions of patients.

According to a report by Deutsche Welle, more than 80% of genetics studies include only people of European descent, which represents less than 20% of the world’s population, a situation many experts find highly disproportionate.

Similarly, most health data used to train AI models comes from patients in the United States, parts of Europe and China. Even though it is only with comprehensive and diverse datasets that AI health systems learn to recognize patterns and make accurate predictions across demographics.

AI systems trained on biased data often perform poorly when applied to populations with different genetic backgrounds, disease prevalence or environmental exposures. For example:

This creates a troubling situation, namely that the populations who stand to gain the most from scalable, low-cost diagnostic tools are the least represented in the systems being built.

The dangers of exclusion posed by biased AI health systems are not just technical; they have tangible, real-life consequences. A cancer detection algorithm that misses tumours on darker skin isn’t just a technical error; it is a life-or-death issue. If medical AI systems are deployed without addressing these issues, it could lead to:

AI has the potential to transform entire economies. But especially in healthcare, its power will only be realized equitably if it is built on foundations that reflect global diversity.

Building inclusive AI health systems is no mean feat. It requires coordinated action across governments, international organizations, private sector innovators and local communities. To achieve this, these priorities should be taken into consideration:

Representative data is the core of inclusive AI. To ensure AI systems serve populations in Africa, Asia and Latin America, we must invest in creating and curating datasets that capture diverse demographics. This could involve supporting local hospitals and clinics in digitizing records. Rwanda’s Digital Health Initiative is an exemplary step in this direction.

Also, encouraging collaborations that make datasets interoperable across borders while respecting data sovereignty is vital. International initiatives such as the Global Alliance for Genomics and Health have shown that cross-border cooperation on data sharing is possible.

Similar frameworks are needed to ensure that broader health data informs AI models, reflecting global realities.

Even when inclusive datasets exist, they are only useful if local healthcare systems can deploy and adapt AI tools effectively. Many medical facilities in the Global South lack the digital infrastructure from reliable internet connectivity to secure data storage required to support advanced AI.

Equally important is human capacity. Clinicians, data scientists and policymakers need training and resources to evaluate AI tools, customize them for local contexts and ensure they align with clinical workflows.

Building local innovation ecosystems around AI in healthcare ensures solutions are not only imported but also created within and for the communities they serve.

Alongside investments in digital infrastructure and local capacity, it is imperative to establish governance systems and regulatory frameworks that would ensure fairness in the design, engineering, deployment and use of these systems.

India’s National Health Blueprint is another strategic initiative aimed at addressing these gaps.

The risk of excluding 5 billion people from the future of AI-enabled healthcare is real but it is not inevitable. The choices made today by policymakers, innovators and healthcare leaders will determine whether AI deepens the existing divides or helps bridge them.

International organizations such as the World Health Organization, in its Global Digital Health Strategy, emphasise the importance of digital health equity. Technology companies and research institutions are increasingly recognizing the need for diverse data. However, progress remains uneven and the scale of the challenge demands more global cooperation.

AI holds extraordinary promise in healthcare. It can help clinicians detect diseases earlier, guide treatment decisions and extend quality care to communities far from medical centres. However, for these benefits to be accessible globally, we must address the structural inequities surrounding how AI health systems are currently developed.