AI in healthcare: Singapore has the tools. Now comes the hard part

SINGAPORE – An elderly patient arrives at a polyclinic with a nagging cough. Her chest X-ray is taken, fed through an AI system and flagged for urgent review within minutes. Her doctor is already looking at it when she sits down.

Elsewhere, a medical social worker finishes her consultation notes before the patient leaves the room – a secure AI transcription tool listened during the consultation and drafted them for her. At midnight, a worried parent gets a clear answer about where to take a feverish child, without waiting on hold on the phone.

These are not visions of the future. They are happening in Singapore today.

And yet, for every team that has made artificial intelligence work, there are others still wrestling with the same pilot they launched two years ago. The technology is not the problem. So, what is?

That question was put to more than 60 health leaders, doctors, researchers and technology experts who gathered in Singapore in 2025 for an advisory think-tank. It was convened by the Health Empowered by AI Launchpad at the Centre for Healthcare Innovation and hosted by NHG Health.

Their answer, published in February in a report, AI For Health: Converting Momentum Into Muscle, is blunt: The biggest barriers to the adoption of AI in healthcare are not technical. They are human. And unless Singapore learns to address them, the country risks ending up with a health system full of impressive-sounding tools that few people can explain, fewer trust and almost nobody uses.

As Minister for Health and Coordinating Minister for Social Policies Ong Ye Kung puts it in the report’s foreword, Singapore’s goal should not be to build the most algorithms, but to become the place where AI in healthcare is implemented well – safely, equitably and at scale. The word “implemented”, while small, carries much weight.

The ‘pilot’ problem

Walk into almost any hospital in Singapore today and you will find AI at work somewhere – a risk-scoring tool here, chatbot there, a scheduling system running quietly in the background. What you will find less often is evidence that these tools have changed how care works for patients.

The reason is familiar to anyone who has watched new technology enter large organisations. Good ideas start in one team and often stay there. All too frequently, a neighbouring department builds something similar, unaware it has already been done. Learning never spreads. The result is a health system with hundreds of experiments and not enough lasting change.

But the problem runs deeper than coordination. A tool that works well in a demonstration often struggles in the reality of a busy ward at 2am. It may not fit how staff execute their work on the ground in high-pressure and intense environments.

It may ask them to do something extra rather than ease their workload or processes. Or staff simply may not trust it – and in healthcare, that hesitation is not irrational. It is often just the right instinct.

Why governance is part of the answer – and part of the problem

Healthcare takes safety very seriously, and rightly so. But there is a difference between oversight that protects patients and oversight that buries promising ideas in process.

Traditional approval systems were built for stable technologies – devices that behave the same way every time, in every setting. AI is different. It learns. It changes. It performs differently depending on who is using it and where. Applying the same lengthy sign-off process to an administrative chatbot as to a system that helps diagnose cancer creates delay without adding safety where it is needed.

One participant in 2025’s discussions put it simply: Governance should fuel confidence, not restrain it.

The same applies to the people expected to use these tools. Doctors, nurses and administrators in Singapore are already experimenting with AI – often without waiting for their institutions to catch up. That curiosity is worth encouraging.

But without clear guidance, practical training and the reassurance that raising concerns will be welcomed rather than ignored, some staff forge ahead confidently while others are quietly left behind.

What is actually working

The health teams making real progress share one habit: They start with the problem, not the technology.

Singapore is not short of AI tools. What is harder to find is focus. The most effective deployments have been anchored in specific, practical frustrations – the documentation that keeps clinicians at their keyboards long after patients have left, the chest X-ray that needs a second pair of eyes, the patient who cannot find a simple answer without calling a busy helpline.

At Woodlands Hospital, an AI transcription tool that captures the conversation between a professional and the patient, drafts notes and maybe even makes recommendations in the future has already cut documentation time for medical social workers by more than 40 per cent.

A smarter nursing scheduling system in NHG Health has reduced the time spent building rosters by 83 per cent, improved staff retention and freed up resources equivalent to nearly 60 full-time positions each year.

NHG HealthBot, which provides AI-enhanced patient support, is now available securely on WhatsApp, and is on track to handle nearly a quarter of routine health inquiries, reducing pressure on call centres while improving access for patients.

None of these succeeded because the technology was extraordinary. They succeeded because someone had identified a real problem, worked with front-line staff to understand it, and built a solution around their needs rather than around what was technically possible.

Singapore’s moment

Singapore is better placed than most countries to get the use of AI right. Its public health system is relatively coordinated. The digital infrastructure is strong. And the Republic has a clear national interest in making healthcare more sustainable as its population ages and the pressure on hospitals and polyclinics grows.

But these advantages will not translate automatically. The harder work lies in the less visible foundations: digital systems that different parts of the health system can share; approval processes that are calibrated to the level of risk rather than being uniformly slow; training that reaches every nurse and administrator, not just the tech-savvy few; and honest evaluation that asks whether patients and staff are genuinely better off, not just whether the project was completed on time and on budget.

It also requires a willingness to talk openly about what has not worked. Health systems that make real progress are those that treat a failed deployment as a lesson rather than something to be quietly shelved.

More than momentum

There is no shortage of excitement about AI in Singapore’s health system, and much of it is justified. The technology is capable of things that would have seemed remarkable only a few years ago.

But excitement alone will not change what it is like to wait for an appointment or sit with an ageing parent in an emergency department. Nor will it prevent a nurse having to work a double shift or give back the hours that nurses and doctors lose every day to paperwork rather than patients. What will change those things is the quieter, harder work of building organisations that can adopt new tools safely, learn from experience and improve steadily over time.

Turn momentum into muscle

In Singapore, all the conditions are there. Used well, AI need not simply make healthcare faster or cheaper. It can also make it more reliable, more human and more sustainable: a system where technology protects the time that doctors and nurses need to simply be present with their patients.

That is worth working towards.