How do we make a success of Artificial Intelligence in health?

Following his 2025 REAL lecture, Hetan Shah, Chief Executive Officer of the British Academy, says developing AI interventions will require meaningfully involving the public, investing in data and IT, and adopting proven technology worthy of our trust.

The government recently published its AI Opportunities Action Plan with a view to ‘mainlining AI into the veins’ of the nation. Given this major political focus on AI, it is timely to review how we enable its successful adoption in health. Especially given the UK’s poor record on the digital transformation of public services.

The breathless rush to ‘unleash’ AI should not prevent policymakers from using their critical faculties to evaluate it – otherwise we are in danger of a solution looking for a problem. We must begin by asking what the most effective (and cost-effective) intervention is for the policy problem we are trying to fix.

Make AI trustworthy

If AI is the intervention of choice, we need high-quality evaluation to screen out the inevitable snake oil that emerges in any lucrative market, as well as malfunctioning tools through so-called ‘hallucination’ and other errors. The proliferation of small-scale AI pilots across the health system may end up being less than the sum of their parts without high-quality assessment. With a multitude of NHS trusts, what is the role for the centre in providing a ‘one and done’ evaluative service to reduce duplication? 

Alongside this we need a stronger public voice to help shape how technology is used in health services. There is no shortage of techniques on how to do this – many pioneered by the late Simon Burall to whom I wish to dedicate my REAL lecture. Unfortunately, this is far from being a priority right now. There is no need to be scared of the public – often they are more supportive of change than is assumed and understanding their concerns can help improve deployment. Loss of public support can destroy well-intentioned projects, as shown with the infamous care.data scheme. Without clear governance and accountability frameworks, there are also dangers that we sow the seeds of a future Post Office Horizon scandal today. 

Remember public services are a complex system 

A naïve view that technology will ‘save the day’ still persists, but we must ask why it will be different this time given the NHS is littered with failed digital transformation projects. We need to begin by recognising that adoption of technology in public services is not easy. Overlaying AI onto failing services will not work, and often we do not need new technologies but to better use and diffuse existing technologies. 

Then there is a danger that during a time of constrained public finances, AI is seen merely as a way of making the NHS more efficient – rather than part of a wider project to transform the service to become more preventative and community based. We should be careful not to automate away those things that are less measurable and hence hard for AI to optimise, such as relationships and community. 

Innovation will require transformation of whole services and that requires investment in management and currently weak IT systems. Data is the unsexy cousin of AI but is a priority area that needs addressing, as comprehensively set out in Cathie Sudlow’s recent review.

Shape the market

The health service is in a good position to shape the market as a powerful single purchaser, but has so far not done enough to signal what it wants. This includes articulating the role of technology in prevention, social care and mental health. AI is also likely to replicate existing biases in data unless we expressly direct it to reduce inequalities. And if we want AI to change power dynamics within the health system, rather than reinforce them, we will need to make an explicit effort to shape technology in this direction. AI will not do so if left to its own devices.

Shaping the market will require will require effective procurement strategies, especially when negotiating with a small number of very powerful technology companies. The UK should remember the tech giants don’t hold all the cards – our high-quality health data is valuable and should not inadvertently be given away. At the same time, how can the NHS help support the UK’s homegrown start-up ecosystem, when it is traditionally not seen by some as a good funder of innovation?

Think broadly about AI’s impact on health

When thinking about improving our health we must look beyond the NHS to consider wider social and economic factors that history suggests are hugely important for our health. The jury is still out on whether AI give us all better quality jobs or lead to massive labour market disruptions. How will AI impact the wider environment we live in? There are growing concerns, for example, about the resource costs of the data centres that power AI. 

Fast-changing technology is also impacting the information systems that inform the public’s decision making about health. Although we have more information available to us than ever before, AI is affecting how we consume it in complex ways, including through algorithmic recommendation systems. How can we shape our changing AI-driven information environment so it supports people’s health rather than undermining it?

Where now?

It is clear that AI is a major priority for the UK government as a way to improve public services. But policymakers should keep their heads and not be naïve. History shows technological change in public services can be hugely valuable but that it is not easy, and the UK has a patchy record. 

For it to be different this time, we need to meaningfully involve the public, invest in our data and IT systems, and not fall for snake oil technologies, but adopt scientifically proven interventions worthy of our trust. Let’s use this opportunity to signal that we want technology that can reshape the health system towards prevention, reduce inequalities and rebalance power relationships rather than reinforce them.