Priorities for an AI in health care strategy

The potential of AI is generating significant excitement in health care. With its ability to help discover new drugs, diagnose illness faster and more accurately and revolutionise clinical note taking, AI has understandably captured the attention of policymakers and practitioners looking to tackle some of the health service’s most complex challenges. As the prime minister noted in the run-up to 2023’s AI Safety Summit, ‘AI could solve problems we once thought beyond us’. Generative AI systems such as ChatGPT and Bard – which can generate text, images and other data – have attracted particular attention, with industry, practitioners and policymakers exploring how these powerful systems might be used in health care.

But while there is hope, there is also concern. The World Health Organization recently warned of the risks associated with generative AI in particular, calling for rigorous oversight to prevent AI systems leading to errors, patient harm and the erosion of public trust. In a safety-focused environment like health care, there is significant focus on how to avoid harmful outcomes, as well as how to determine who is accountable when things go wrong. Researchers have also warned about the potential of AI to further exacerbate health inequalities, for example in minority ethnic groups.

And despite AI’s promise, long-standing challenges with the implementation and evaluation of health care technology continue to slow down service transformation. Not only is there still work to do to get the basic digital infrastructure right – an essential foundation for the effective use of AI – but AI brings its own complications and challenges. While most service transformation will ultimately be led by staff and patients at the front line of care, political energy, ­­commitment and resources are vital to support this change and create a safe, enabling environment. We argue that a dedicated strategy for AI in health care is needed to coordinate current fragmented efforts in the NHS in England. Here, we set out six priorities this strategy should address and some of the steps to do so. 

The health service is fertile ground for technology-driven improvement. With the ability to coordinate reforms and control costs centrally, access broad, in-depth and life-long health care datasets and draw on the UK’s world-class science base, the NHS should be well placed to drive the development and adoption of technologies like AI. 

Although there has been a lot of governmental focus on AI in the UK, to date much of this has been cross-sectoral – such as the National AI Strategy, the associated AI Roadmap and the formation of the AI Policy Directorate under the Department of Science, Innovation and Technology – largely centring on the safety and regulation of AI and efforts to promote innovation. While there are important rationales for tackling key issues relating to AI cross-sectorally, this work has not enabled a coordinated focus on the potential of AI to improve health care and the specific opportunities and challenges AI presents in this field. 

The need for coordination is one of the many reasons for a dedicated AI in health care strategy. In recent conversations with over 50 senior stakeholders across the NHS, academia and industry, we heard recognition of the important pockets of work on AI underway within the NHS. But we also found broad agreement that progress in developing and deploying AI in the NHS is being hampered by the lack of an overarching strategy and agreed focus, as well as a lack of coordination and collaboration across national-level agencies and organisations.

A dedicated strategy would not only support a coordinated focus on AI in health care and improve alignment across the landscape but would also help identify and address the gaps in work currently underway and galvanise greater progress on key challenges such as infrastructure, skills and spread. For example, because health care innovation sits under both health care policy and industrial strategy, industrial strategy considerations often result in a significant focus on the development of innovations and far less on the underlying infrastructure and capability needed to ensure their effective adoption and spread in the NHS – which is ultimately what is needed for improved health outcomes in the UK.

It is important to emphasise that a significant component of the AI challenge in the NHS relates to adoption and spread, with AI tools often trialled in a small number of leading providers but with no clear route to scale them up. This increases the likelihood that organisations considered ‘behind the curve’ on innovation will continue to fall behind and that inequalities in provision and access will grow as some trusts become much more technologically advanced than others. Progress on this front could be one of the big prizes of a dedicated strategy. Importantly, this will require a focus not just on future AI but on how we can make the most of existing AI tools over the next few years to realise their benefits across the NHS. Without the right tools, resources, infrastructure and policy environment, and without the required coordination between different parts of the system, the health service will struggle to absorb AI technologies in ways that have lasting impact.

Perhaps most importantly and urgently, escalating demand for health care driven by an ageing population and a rise in long-term conditions and multi-morbidity – combined with significant workforce shortages – means it is critical the NHS do all it can to capitalise on the opportunities presented by technology and AI to support staff and improve care quality and productivity. A dedicated strategy is needed to galvanise efforts to harness AI for the benefit of the NHS and ensure the health service remains strong and sustainable for the future.

England is not alone in being in the early stages of developing such a strategy, but having a national health system provides an enviable opportunity to develop a comprehensive and coherent approach more quickly. A strategy will also need to be flexible to accommodate rapid developments in this field, such as general purpose AI (systems such as large language models that have a wide range of possible uses and can be applied to many different tasks). Policymakers and practitioners will need to ensure that policies and infrastructure put in place now are able to support longer-term innovation and aspirations. But uncertainties about future innovation should not prevent the NHS from capitalising on the benefits of a more coordinated and comprehensive approach today. 

With the power and promise of AI also comes significant risk, particularly in a safety-critical sector like health care. Without proper safeguards, AI could cause or magnify serious harms to individuals, organisations or communities, especially where it is used in patient care. We have already seen this play out – for example, in the case of a risk-prediction algorithm that failed to recommend medical treatment for black patients. There could be detrimental impacts beyond patient care too, such as the significant carbon emissions, water consumption and e-waste generated by developing and running AI models.

Appropriate safeguards and standards are therefore essential to reduce risks and avoid harmful outcomes. Recent international conversations have focused heavily on AI safety, with global collaborations agreed at both the UK’s 2023 AI Safety Summit and the AI Seoul Summit in 2024, although the agreed standards remain voluntary and, at least in the UK, have not been reinforced through new legislation.

While safety remains an integral part of AI development, the transformative potential of AI means our ambitions must extend beyond simply avoiding harmful outcomes to ensuring AI development has a positive impact on society and the environment. To date, attention in the UK has largely focused on developing ethical principles and safety guidance and balancing regulation and innovation, but less on the broader impacts of AI technologies. 

Responsible AI is an approach that can provide a framework to ensure that the development of AI is not only trustworthy, designed with ethics in mind and with any risks minimised, but also ultimately benefits people, society and the planet. This approach is being adopted elsewhere, such as in Denmark and Japan, where responsibility and social benefit are central principles in their national AI strategies.

To harness the potential of AI in health care, progress is needed on multiple fronts. Here, we highlight six critical challenges an AI in health care strategy should prioritise, together with some of the actions needed to address them. 

1. The use of AI should be shaped by the public, patients and health care staff to ensure it works for them

AI is creating new functionalities that have profound implications for health care. For example, the use of AI for decision making and advising patients, and the use of AI-enabled robots to assist with health care tasks, both have important implications for accountability, patient experience and the quality and ethos of care.

Developing and using technologies in ways that are compatible with high-quality care, and that command the confidence of patients and the public, will be critical for realising their benefits speedily. Conversely, poor user experience, social and ethical concerns and a lack of public support can all act as significant constraints on service transformation. In addition, engaging patients and staff, particularly those often underrepresented in clinical research and tech development, in the design and development of AI technologies and their application to health care is critical to help prevent bias and ensure new uses of technology are rooted in a deep understanding of users’ needs and work well for them. 

Previous Health Foundation polling showed that while more of the public would like to see greater use of AI in health care than less (40% compared with 13%), there is definitely further to go to build public support around both AI and the use of health data. For example, a recent Health Foundation survey found that the public currently have relatively low levels of trust in the use of health data by both government and commercial organisations. 

To build public confidence, AI not only needs to be safe and well regulated, but there needs to be active engagement with the public, patients, carers and staff to inform decisions about how it should be used in health care. While there have been a number of important national and regional engagement initiatives on aspects of AI such as data sharing, with more currently underway (such as those around the Federated Data Platform), our research highlights that further engagement with the public and NHS staff are needed to build understanding of AI and address risks and concerns. There is a chance here to learn from approaches elsewhere, such as Canada’s proposal (as part of its Pan-Canadian Health Data Strategy) for a national public assembly to provide advice to government on AI in health care and its Digital Charter in Action, which was developed through engagement with the public. There is also more to be done to encourage and support technology developers to work closely with a wide range of patients and staff to identify needs and co-design solutions. As the Whitehead Review of Equity in Medical Devices recommended, engagement with, and the co-design of, AI should involve diverse groups of patients and the public and consider equity, transparency and fairness. 

An AI in health care strategy should be based on a deep understanding of what people in the UK think about AI-driven health technologies. It should ensure there are effective mechanisms in place for engaging patients, the public and NHS staff on relevant topics as they arise to inform high-level decision making. It should also involve patients and staff in the co-design of AI solutions if we are to harness their potential in a way that works for all.

2. The NHS must focus AI development and deployment in the right areas

AI has significant potential to transform health care across a broad range of pathways and processes. But with finite resources and attention available to support AI development and deployment, there will be real advantages in targeting efforts where they can have the most significant impact – something a dedicated strategy can create consensus on.

So far, attention has largely centred on the development and regulation of AI for the discovery of new treatments, vaccines and drugs; improving diagnoses, particularly in diagnostic radiology; and detecting, preventing and screening for cancer. While these are all critically important areas, there has to date been comparatively little focus on the potential of AI to help reshape pathways, promote prevention and health, and improve operational and administrative work (notwithstanding some important national initiatives such as the new government AI Incubator). To have a better chance at achieving the significant service transformation the NHS needs, as well as the widespread productivity gains needed over the coming years, the focus on AI must incorporate these wider aspects of health care. This will be particularly important in areas such as managing long-term conditions and frailty, given that the number of people affected by these conditions is growing, with 9.1 million people in England projected to be living with major illness by 2040.

The emphasis on more politically attractive clinical AI continues to overshadow the significant potential of AI technologies for operational and administrative work. From helping to optimise hospital flow to creating discharge summaries, planning community worker staffing and routes and analysing qualitative patient feedback (see Box 1), AI systems can transform administrative and back-office functions. And without some of the clinical risks or regulatory burdens of AI used as a medical device, these impacts could be realised faster.

With so much potential, how should the NHS decide where to invest its efforts? A twin-track approach is needed: first, supporting the most promising innovation and experimentation currently happening in innovative teams and provider organisations; and second, setting out a small number of high-level priorities for AI use and supporting the demonstration, testing and spread of these tools. The first approach is needed to ensure new applications of AI can continually emerge and be investigated in contextually appropriate ways. The second is essential to focus scarce attention and resources in a way that makes a difference on specific challenges the NHS needs to address – such as improving high-volume pathways (which treat a high number of patients with comparatively less complex conditions) or freeing up time to care for staff. It is crucial that priority setting of this kind is supported by both effective horizon scanning so the NHS is aware of the latest trends and emerging technologies, and effective demand signalling from health care staff and NHS provider organisations as to their challenges AI could help with.

Stronger demand signalling from health care staff and NHS provider organisations will also be important where AI development and deployment happens locally. At present, the range and availability of AI technologies is driven primarily by industry or focuses on where good data are readily available (such as with imaging). While this increases the likelihood of being able to procure ‘off-the-shelf’ solutions, it also risks overwhelming organisations that are unsure which system will be best for them while also reducing the likelihood those systems will match specific local requirements and populations. Health innovation networks can play a critical role here, linking developers with NHS teams and providers.

An AI in health care strategy should support local innovation and experimentation while also setting out a small number of high-level priorities where AI can help address key challenges the NHS faces (administrative and operational as well as clinical). It should also support the demonstration, testing and spread of these tools. As part of this, a strategy will need to maintain effective horizon-scanning functions and provide opportunities and mechanisms for NHS staff and provider organisations to signal where AI could help most.

3. The NHS needs data and digital infrastructure that will enable it to capitalise on the potential of AI

If the NHS is to deploy AI at scale, it needs the right underpinning digital and data infrastructure with the right level of maturity. This includes effective mechanisms to collect, manage and provide access to the range of data needed to develop AI models. It also includes the wider digital infrastructure needed to deploy AI, including hardware and software, connectivity, cyber security and interoperability, as well as the associated workforce capabilities. While progress has been made, recent assessments show that, overall, data and digital infrastructure in the NHS is not yet adequate, and digital maturity is generally low, with only a small number of exceptions. So there is still further to go to get the underlying basics right.     

One of the foundations of digital maturity is the use of an electronic patient record (EPR) system – both as a means of data collection and as a platform into which to integrate AI tools. While EPR system coverage has increased to around 90% of secondary care providers, some providers still do not have an EPR system, and others may need to upgrade theirs to use the full functionality of AI. The experience of the US, where EPRs are more mature, shows that further work is needed to unlock the functionalities they could provide in the NHS, such as with clinical decision support tools. 

Data are critical for the effective development, assessment and ongoing improvement of AI systems, and access to that data needs to be safe, ethical, secure and supported by the right governance models. But there remain significant and longstanding barriers to accessing and sharing health data that are limiting both innovators’ abilities to develop and train AI models and the NHS’s ability to spread impactful interventions across the country. 

The importance of data for research and technology development has long been recognised, with a number of ambitions set out in Ben Goldacre’s ‘Better, broader, safer’ review and the NHS England data strategy. While significant funding has been committed to improve data linkage, build subnational secure data environments and develop a national Federated Data Platform, more work – including more investment and support – will be needed if the ambitions set out in the Goldacre Review and NHS England data strategy are to be achieved. 

A particular barrier is the lack of standardised approaches to data access across the health service, often meaning that accessing the appropriate data for developing or deploying AI systems in different NHS trusts (or even within the same trust) is both cost and time prohibitive. The NHS Research Secure Data Environment Network – which provides secure access to health data for research and development purposes, including for the development of AI algorithms – is one approach that could make a difference here. However, it is a relatively new initiative, so more time is needed to see what kind of impact it makes. More radical proposals such as the creation of a National Data Trust jointly controlled by the NHS and government have also been put forward. This is an area where it will be important to compare with and learn from other nations, such as Denmark, which has also set up a secure data environment. 

Another challenge is representativeness. Many datasets for training algorithms are not representative of the populations for whom the algorithms may be used, risking discrimination and harms from in-built biases. While initiatives like the STANDING Together partnership (see Box 2) have made recommendations on how to ensure datasets for AI testing and training are diverse and inclusive and promote AI generalisability, there is currently no systematic approach for building these into practice, and adherence to guidance remains voluntary.

Another potential way to improve the availability of data for developing and testing AI models while protecting patient confidentiality is through artificial or synthetic datasets. NHS England has made artificial datasets available through its artificial data pilot scheme, and the Medicines and Healthcare products Regulatory Agency (MHRA) has a similar initiative – although further work is needed to assess the viability of this approach at scale, including the risks associated with using synthetic data, such as bias amplification or the absence of robust methods for auditing data quality. There is an opportunity to learn from other countries here, including Norway, which is also using synthetic datasets to support AI system testing and development. 

One current debate is whether charging for data access could generate greater resources for the NHS, and a range of potential mechanisms have been put forward to achieve this, including one-off payments in exchange for data access. The idea of charging for access to and use of NHS data is not new: the Clinical Practice Research Datalink, which provides access to primary care data for researchers, including from commercial organisations, operates on a cost-recovery basis. NHS England has also committed as part of the NHS Secure Data Environment programme to ‘always charge a fee for accessing health data’. However, efforts to go beyond cost recovery have – so far – not proven successful, and past experience shows that this an extremely complex area to navigate. Further work is clearly needed to explore the viability of this idea, including engaging the public.

An AI in health care strategy should ensure the NHS’s digital infrastructure is fit for purpose and set out how processes can be standardised and improved to allow efficient access to high-quality data for the development of AI systems. Such access should be based on a proportionate approach to data security and privacy that effectively balances risk and opportunity. 

4. The use of AI in the NHS must be underpinned by high-quality testing and evaluation

A robust approach to the evaluation of AI-based health technologies is important to ensure the safety of AI tools and build the confidence and trust needed to encourage their adoption and spread. However, the complex and adaptive nature of AI algorithms and how they interact with people and processes in health care – as well as the speed at which AI systems can develop – mean that conventional evaluation methods are often not suitable. 

Some AI is built on adaptive algorithms that can change their behaviour after deployment, leading to differences in performance. Meanwhile, static AI models may become less accurate with demographic shifts and changing health care needs (for instance, AI models developed to predict hospital length of stay using pre-pandemic data may now be less accurate, as length of stay has increased in recent years). AI can also be used for a different purpose than originally intended. Indeed, AI is increasingly likely to be a ‘foundation model’, meaning its specific purpose is not designated at the point of design. In addition, AI models may be trained on datasets not representative of the populations they will be applied to, risking poor outcomes for particular social groups (see Box 3, which looks at this phenomenon in the case of diabetic retinopathy screening). Beyond the nature of algorithms themselves, the complexity of implementing AI in a highly socio-technical system like health care also poses challenges for the evaluation of AI models.

To ensure that AI technologies are safe and effective and remain so, ongoing surveillance and evaluation – as well as regular revalidation – may be needed for as long as an algorithm is used. This would require sufficient transparency for appropriate scrutiny during AI deployment – accounting for any changes in purpose or variation in performance across user groups.

While there has been work to explore the requirements for evaluating and reporting evidence around AI as a medical device – such as the SPIRIT-AI and CONSORT-AI initiatives, NICE’s Evidence Standard Framework for Digital Health Technologies and the development of specific evaluation frameworks for AI – further work is required to standardise these approaches and support their translation into practice. As Morley and colleagues note, while much is known about what should be done when it comes to evaluating AI, the challenge is establishing how to do it. 

In addition, not enough is yet known about the factors that enable the successful adoption of AI in health care. Successful adoption and implementation of technology is an area where our research has shown that the challenges – which relate not just to skills and equipment but new ways of working, patient interaction and wider culture – are often underestimated. So, uncovering the enabling ingredients of change should be an important focus for formative evaluations of AI implementation. For example, there is currently no systematic way for NHS organisations to assess their capability and readiness for AI deployment. A readiness assessment framework could help improve decision making about AI deployment and reduce risks. Singapore, for instance, has developed tools and resources for enterprises to assess their organisational readiness to adopt AI.

Keeping up with the pace of innovation is another challenge. A new strategy must encourage greater investment in evaluation capacity to ensure the NHS and innovators are able to test and evaluate AI both pre- and post-deployment. This should include rapid evaluation approaches that can reduce the time between innovation and implementation, an area our analysis shows has historically lacked funding and support. And given that the success of AI in practice depends on how well it performs in live health care settings, not the laboratory, a strategy must also consider how to broker and support more opportunities for testing AI technologies in real-world settings. 

An AI in health care strategy must support the further development of evaluation frameworks appropriate for AI and boost the capacity to evaluate AI as it is developed and implemented in the NHS.

5. The NHS needs a clear and consistent regulatory regime for AI

Effective regulation of AI in health care needs to ensure safety and provide consistency and clarity for developers and users. Under the current UK regulatory system, there is a wide and fragmented range of rules, principles and proposals that are failing to provide the necessary clarity and a need to address regulatory gaps and improve coordination among regulatory bodies.

A variety of regulatory approaches are possible. The European Union has introduced horizontal legislation, the EU AI Act, that classifies health care AI as ‘high risk’ – meaning that all uses will require third-party conformity assessments and be subject to strict regulatory obligations before they can be put on the market. The Act also provides for the establishment of coordinated 'regulatory sandboxes’, which allow for the testing of AI in controlled environments across the EU to support innovation.

In contrast, the UK government argued in its 2023 white paper that dedicated, horizontal AI regulation is not needed, as sectoral health care bodies such as the MHRA, NICE, National Institute for Health and Care Research (NIHR) and Care Quality Commission can address AI through existing approaches. (While there have been some indications the government may be changing tack on this, it seems that any additional legislation will likely only focus on the safety of more advanced foundation models, such as large language models.) There are also signs of a split in approaches across the UK, with Scotland introducing mandatory registration for all public-sector uses of AI, similar to the EU.

While sector-specific regulation can be more agile, this approach makes consistency and coordination across regulators especially important. For example, while the new AI Safety Institute will seek to ‘carefully examine, evaluate and test new types of AI to understand what each new model is capable of’, it is so far unclear what role it will play in evaluating existing AI within health care.

Important work is already happening to improve both the regulatory framework and the institutional landscape. This includes recommendations by the Regulatory Horizons Council in 2022 for effective AI regulation and a 2023 roadmap by the MHRA to ensure clear regulatory requirements for software and AI to protect patients. A new NIHR-funded Incubator in AI and Digital Healthcare hosted by the University of Birmingham is building national capacity in responsible innovation and regulatory science, and the MHRA’s Software and AI as a Medical Device Change Programme is considering how to build concepts like transparency and explainability into the regulation of AI medical devices. The NHS AI Lab has also been attempting to streamline regulation through a new online advice service, the AI and Digital Regulations Service.

Despite such efforts, there is no shared overarching strategy to guide the alignment of these existing workstreams, and confusion remains among innovators and clinicians as to the UK’s regulatory stance. There is particular concern among clinicians as to where clinical liability sits when algorithms are used in clinical decision making, so providing regulatory clarity here will be essential. It is also worth considering where international alignment could help; given that many UK-based companies will also seek to sell into EU markets, they may work to meet EU standards by default.

An AI in health care strategy needs to create a regulatory framework that provides clarity and consistency for AI developers and users. It must prioritise the coordination of sectoral regulators, bringing all relevant bodies together under an agreed approach that addresses gaps and overlaps. 

6. The health care workforce must have the right skills and capabilities to capitalise on AI

AI has the potential to boost both productivity and job quality if developed and deployed in the right way. But realising these benefits will rely on health care workers having the skills, knowledge and capacity to implement and use AI effectively. Both clinical and non-clinical NHS staff will need a range of education and training to ensure they can capitalise on the potential of AI. Given the pace at which AI is evolving, this cannot be addressed solely through student curricula; training and development will be needed throughout careers. 

Following the 2019 Topol Review, Health Education England (HEE; now part of NHS England) set out an education and training approach that recognised how the precise capabilities needed will differ according to someone’s role and level of interaction with AI. For example, the capabilities required to implement AI in a care pathway will look different to those required to explain an AI system to a patient. 

To put this approach into practice, HEE recommended the creation of AI-specific content for curricula and the development of roles and career paths for specialist AI health care workers. Action here is particularly important. As there are currently no specific career paths that allow clinicians to specialise in AI or digital health care, and as NHS pay is often unable to compete with the salaries of digital data and technology roles outside the NHS, the health service risks losing talented staff to other sectors or not attracting the talent needed in the first place. 

Despite fears about jobs, much research suggests that – in health care at least – AI is more likely to transform the nature of work than lead to redundancies. A recent Health Foundation survey showed that NHS clinicians are on balance optimistic about the potential of AI to save them time in their work within the next 5 years. However, AI use could also have negative impacts on the health care workforce, such as skill loss due to tasks being performed by AI or diluting aspects of work that staff might find fulfilling, such as the quantity or quality of patient contact. The benefits and risks of AI also may not be felt evenly across the workforce. Our research shows attitudes towards AI differ across professional groups, with health care assistants, for example, feeling more negative about the impact of AI than medical staff. So, it will be important to understand staff concerns and support role development for those occupational groups most likely to be affected. 

A new approach to NHS management will also be required to deliver on the AI agenda. Managers will need to play a much greater role in facilitating and continually assessing innovation as part of their everyday work. This will require not only a cultural shift but also training that provides the necessary knowledge and skills, such as for technology appraisal and implementation, quality improvement and collaborative leadership.

Ultimately, there needs to be a shared vision for how professions and occupations – as well as new roles – should develop with greater use of AI. And NHS staff themselves should play a central role in the development of this vision, in partnership with their colleagues, employers, trade unions, professional and representative bodies, patients and the public.

An AI in health care strategy must set out concrete plans to equip the current and future workforce with the skills needed for using AI, develop career paths that allow health care workers to specialise in AI and empower staff to shape the evolution of their roles. 

We have set out the case for an AI in health care strategy that would provide greater coordination and direction for the NHS, industry and other stakeholders. We also outlined six priorities for such a strategy.

For government and national NHS bodies, addressing these priorities will require coordination of currently fragmented initiatives and responsibilities, along with action on a range of fronts. A strategy should not only support local innovation but also set out a small number of high-priority areas where AI can help tackle key challenges the NHS faces and focus attention and resources on supporting the demonstration, testing and spread of these tools. It must also enable a clear and consistent regulatory regime and improve the NHS’s capacity to implement, use, evaluate and spread AI. We also argue it is essential to involve patients, the public and NHS staff in both decisions about AI and the design and development of technological innovations themselves.

Progress on the priorities we set out should deliver a range of important benefits:   

• Patients and the public will have a voice in decisions about AI design and deployment, be confident that any AI used in the NHS is safe and effective and that the use of health data is secure and trustworthy, and – if it is deployed well – benefit from AI’s potential to improve care quality and experience.
• Staff and health care providers will have the skills and capabilities needed to capitalise on AI, the ability to shape the development of new technologies to ensure AI effectively tackles the challenges they face, and – if it is deployed well –benefit from AI’s potential to improve job quality and system performance.
• Industry will get a clearer view of what the NHS needs and how to navigate the range of regulations, standards and obligations, and have opportunities for closer collaboration and partnership with the NHS, patients and the public in developing new technologies.

The prize for establishing – and delivering on – a clear and compelling way forward for AI in health care is significant. It would not only help the UK economy thrive, but also help achieve better health outcomes and patient experience and a more efficient health service.