By Vishnu Narayan Earlier this year, our team at HealthTechAsia had the privilege of contributing perspectives from the Global South during the United Nations Stakeholder Consultation on the Global Dialogue on Artificial Intelligence Governance.
The consultation brought together governments, scientists, industry, civil society, and international organisations to discuss one of the defining policy questions of our time: how can societies govern artificial intelligence in ways that advance innovation while protecting the public interest?
The recently released Preliminary Report of the Independent International Scientific Panel on AI should therefore be welcomed not simply as another technical assessment, but as evidence of an international conversation that is becoming noticeably more mature.
The report remains preliminary. It does not attempt to settle every debate surrounding artificial intelligence, nor should it. Its significance lies elsewhere. It reflects a growing recognition that governing AI requires more than evaluating increasingly capable models. It requires understanding the institutions, environments, and social contexts within which these systems ultimately operate.
For healthcare, this evolution may prove particularly important.
UN consultation calls for enforceable AI governance as Global Dialogue approaches
Healthcare has always been more than technology
Much of the early discussion surrounding artificial intelligence in healthcare understandably focused on technical capability. Could algorithms detect disease more accurately? Could language models assist clinicians? Could AI accelerate scientific discovery?
These questions remain important. Yet healthcare has never depended on technology alone. Clinical judgement, functioning referral pathways, trained professionals, public trust, and institutional capacity have always determined whether innovation ultimately improves patient care.
One of the report’s strengths is that it consistently reflects this broader understanding.
Its discussion of AI-enabled diabetic retinopathy screening illustrates the point well. The reported success did not arise solely because an algorithm identified retinal disease. It succeeded because patients could be referred into functioning care pathways, clinicians remained involved, and health systems possessed the capacity to act on the information generated.
Technology created opportunity. Institutions converted that opportunity into better health outcomes.
Not every AI System deserves the same governance
Another encouraging feature of the report is its willingness to distinguish between different forms of artificial intelligence. For several years, discussions around AI regulation have often treated healthcare AI as though it represented a single category of technology.
The report adopts a more nuanced position. Task-specific clinical systems designed to solve clearly defined medical problems are assessed differently from general-purpose conversational models capable of discussing almost any health-related question.
This reflects an important principle of good public policy. Different risks require different forms of governance.
A clinically validated imaging algorithm operating within a regulated healthcare system presents fundamentally different governance questions from a conversational chatbot providing informal health advice to millions of people online.
Recognising these differences allows regulation to become more proportionate while preserving public trust.
The conversation has become more global
Perhaps one of the report’s most welcome developments is its treatment of the Global South. For many years, international discussions often positioned developing countries primarily as future recipients of AI technologies. The report adopts a broader perspective.
Experiences from India, Rwanda, Kenya, and other lower-resource settings are presented not simply as implementation examples, but as important sources of governance insight. They demonstrate that successful deployment depends upon local languages, workforce capacity, referral systems, institutional design, and social realities that differ considerably across countries.
This represents a meaningful shift.
Healthcare systems operating under conditions of constrained resources frequently encounter governance challenges earlier and more visibly than highly resourced environments. Their experiences therefore contribute valuable evidence for the international community rather than merely illustrating implementation difficulties.
Equally encouraging is the report’s recognition that bridging AI divides cannot be achieved through access to computational resources alone.
Representative datasets, regulatory expertise, local evaluation capability, and scientific capacity increasingly determine whether countries can govern AI according to their own health priorities and public interests.
Participation ultimately depends upon institutional capability as much as technological capability.
Trust requires more than AI literacy
The report also approaches AI literacy with welcome balance. Helping clinicians, regulators, and healthcare professionals understand artificial intelligence is undoubtedly necessary. Better understanding can reduce both overconfidence and unnecessary hesitation when adopting new technologies.
Yet the report wisely avoids suggesting that education alone can resolve governance challenges. Healthcare professionals should not be expected to compensate for systems that remain opaque, poorly evaluated, or insufficiently governed.
Responsibility must continue to be shared.
Developers should design systems appropriate for high-risk environments. Regulators require the expertise necessary to evaluate increasingly complex technologies. Healthcare organisations must establish appropriate oversight, while governments create the institutional conditions within which responsible innovation can flourish.
Public trust has always depended upon this collective responsibility. Artificial intelligence should be no different.
A useful direction of travel
Every international report inevitably reflects compromise. Diverse perspectives rarely produce complete consensus. That should not be viewed as a weakness.
Rather, it reflects the reality that AI governance cannot be developed by any single discipline, institution, or country acting alone. One of the most encouraging aspects of this preliminary report is that it appears to move the international conversation away from asking only what artificial intelligence is capable of doing towards asking under what conditions societies can deploy it responsibly.
For healthcare, this distinction is particularly valuable.
Health systems have long demonstrated that successful innovation depends not only on scientific excellence but also on governance, institutional trust, professional judgement, and public accountability. Artificial intelligence does not change those principles. If anything, it reinforces them.
As the report enters the next stage of consultation, post the recent UN Global Dialogue on AI Governance, there remains an important opportunity for governments, researchers, clinicians, industry, and civil society to continue refining these ideas together. The quality of the process itself deserves recognition. Bringing together such a diverse range of scientific expertise and regional perspectives is no small undertaking, and the report reflects that collective effort.
The next phase of AI governance will almost certainly depend less on building increasingly capable systems than on strengthening the institutions responsible for deploying them wisely. For healthcare, that may prove to be the most important lesson of all.
Matthew Brady