AI in health care must earn the public’s confidence
About 4 mins
The government’s 10-Year Health Plan commits to completing the move from an analogue system of paper, pagers and fax machines to a technology-enabled service that ‘liberates staff from admin and allows people to manage their care as easily as they bank or shop online’. Artificial intelligence (AI) is central to this vision, through applications such as analysing test results, dictating notes and creating summaries, and predicting risks. National policymakers are also increasingly focused on the economic potential of AI development in the UK.
Turning this vision into reality is not straightforward. Deploying technology in the NHS often involves difficult choices about what we value most and what may have to change to reap the benefits on offer. Is it better to have stricter rules governing AI’s use, even if that potentially deters companies from developing AI in the UK? Or is it better to have more relaxed rules that could encourage companies to develop here? These questions are as political and ethical as they are technological.
So where does the public land when competing priorities collide? We asked 8,000 members of the UK public aged 16 years and older, to weigh up five trade-offs that policymakers, NHS leaders, health professionals and AI developers may face. While these are unlikely to be strict ‘either or’ choices in real life, we used simplified scenarios to illustrate what the public prioritises when tensions arise. This research formed part of our annual survey of public and NHS staff attitudes to technology in health care.
Safety and evidence outweigh the desire to accelerate
Across the trade-offs we tested, a consistent pattern emerges. When asked to choose, the public tends to prioritise safety, oversight and assurance over speed or economic opportunity.
Policy debates often focus on the tensions between creating robust safeguards and a permissive environment for AI innovation. When asked whether the UK should prioritise strict rules on AI, even if that might deter some companies from developing AI here, or instead encourage investment through more relaxed regulation, nearly three-quarters of respondents (71%) favoured stricter rules (Figure 1). Only 29% were willing to accept looser regulation in order to attract AI companies. This suggests limited public support for approaches that frame AI policy in health primarily around competitiveness or economic growth.
Figure 1
A similar picture emerges on evidence standards. Almost three-quarters of respondents (72%) said it mattered more that there is a substantial body of evidence proving AI tools work, even if that means they take longer to become available (Figure 1). Fewer than 3 in 10 (28%) prioritised rapid availability in the absence of strong evidence. This highlights the importance the public places on diligence and evaluation. Many are willing to accept slower progress if it reduces risk and increases confidence that technologies are safe and effective.
Speed comes second to human oversight
This preference for assurance extends to how AI is used in clinical care. When asked whether it was more important for patients to receive test results as quickly as possible, or to wait longer so that a human could check the outputs of an AI system, 70% of respondents prioritised human oversight (Figure 1). Only 30% prioritised speed without human checks.
This reinforces the broader pattern seen in the survey: where trade-offs relate to safety and effectiveness, the public tends to prioritise stronger diligence and safeguards – human checks, high evidence thresholds and strict rules – over potential benefits such as speed and economic opportunities by a margin of around 70% to 30%. Even where AI promises performance and efficiency gains, the public appears to want those gains embedded in systems that retain clear human oversight.
The public is more evenly divided on other trade-offs
Previous Health Foundation research suggests lower levels of support for technologies that appear to come between people and health professionals. To explore this further, we asked respondents whether people should be able to access health advice as quickly as possible, even if it comes from an AI system such as a chatbot, or whether advice should always come from a person, even if that means waiting longer. While a slim majority (55%) favoured receiving advice from a health professional, a substantial minority (45%) prioritised speed through AI-generated advice (Figure 2).
Figure 2
Another possible trade-off relates to explainability, where the ‘black box’ nature of some AI models can make it difficult to understand how the data going into the system are turned into decisions and recommendations. If performance gains come at the expense of transparency and explainability, clinicians may struggle to interpret recommendations and patients may have less confidence in them. When presented with this trade-off, respondents were evenly split (50%–50%) between prioritising AI tools being as accurate as possible and being able to explain how those tools produced their results (Figure 2). This balance shows that opaque systems may struggle to build trust, reinforcing the importance of considering how AI is integrated into care pathways.
As policymakers and NHS leaders develop new approaches to governing AI, including through the forthcoming NHS AI roadmap and the National Commission into the Regulation of AI in Healthcare, these findings offer a clear signal about what matters most to the public.
Across the scenarios tested, people consistently prioritised strong evidence, clear oversight and meaningful human involvement in decision making ahead of speed or efficiency. Approaches that build trust and enable safe and reliable integration of AI matter because research suggests that when technologies are seen as high risk, people may discount or even reject their potential benefits.
AI will only take root in the NHS if it earns our confidence.