The team set out to answer a deceptively simple question: does the public view AI-driven care differently from human-centred care, and does that difference shift depending on where a person sits along the dementia care journey? To probe this, the researchers employed a repeated-measures experimental design, a technique in which each participant evaluates multiple carefully constructed scenarios rather than a single one. This approach allows researchers to isolate the effects of specific variables while controlling for individual differences in outlook. The scenarios, presented as short vignettes, varied along three dimensions. The first was the stage of care, aligned with the five stages of NHS England’s Dementia Well Pathway, which traces the trajectory from early awareness and prevention through diagnosis, living well with the condition, and ultimately dying well. The second dimension was what the authors call system centrism, contrasting an AI-based care system with a human-based one. The third was the level of involvement, distinguishing between moderate support, in which the system assists, and full control, in which it takes charge.

The choice of participants was itself a deliberate act of strategic targeting. Rather than sampling the general population, the researchers focused on adults aged 55 to 64, a cohort that faces an elevated risk of developing dementia within the coming decade and is also among the most likely to encounter these technologies as patients, carers, or family members in the near future. Their attitudes therefore offer something close to a preview of the real-world reception that AI care systems can expect as they move from prototype to deployment. Each participant rated every scenario on two separate scales: how acceptable the arrangement would be, and how plausible or likely they judged it to be in practice. This dual measurement is important, because acceptance and perceived likelihood can diverge in revealing ways. A technology might be seen as desirable yet improbable, or inevitable yet unwelcome, and such gaps shape both consumer behaviour and policy debates.

The results paint a nuanced picture in which the public’s verdict on AI care depends heavily on context. Across the board, acceptability was sensitive to both the stage of care and the type of system involved. The most striking statistical feature was a significant interaction effect between system centrism and level of involvement. In plain terms, scenarios involving full human involvement were consistently rated as more acceptable than any AI alternative, while AI involvement was viewed favourably only when it operated at a moderate, supportive level. When AI was framed as assuming full control of care, approval collapsed. This pattern suggests that the public is not rejecting the technology itself so much as rejecting the prospect of human replacement, a distinction that carries substantial weight for how AI products in this space are designed, marketed, and regulated.

Crucially, the interaction effect did not remain constant across the dementia care pathway; it intensified as the pathway progressed. The largest discrepancies between human-centred and AI-centred care emerged in the later stages, particularly those described as Living well and Dying well, where full-control AI scenarios received the lowest acceptability ratings of the entire study. This gradient makes intuitive sense once considered from the perspective of patients and families. In the earliest stages, when a person is still largely independent, an AI system offering reminders or monitoring may feel like a helpful convenience. But as dementia advances and the stakes of care decisions rise, the emotional and ethical weight of those decisions grows, and the public appears increasingly unwilling to see machines occupy the central role. End-of-life care, in particular, seems to be regarded as territory where human presence, empathy, and judgement are not optional extras but essential elements.

The second outcome measure, perceived likelihood of actualisation, followed a related but distinct pattern. Participants judged scenarios to be more likely under conditions of moderate involvement than under full control, suggesting that the public already senses, perhaps correctly, that the realistic future of AI in dementia care lies in assistance rather than autonomy. Human-centric scenarios were also rated as more likely than AI-centric ones, and this gap widened in the later stages of the pathway. In other words, respondents not only preferred human involvement as dementia progressed; they also expected the health system to deliver it. This alignment between preference and expectation may partly reflect public familiarity with how care currently works, but it also signals a potential credibility problem for any organisation promising fully autonomous AI care in this domain.

Underlying these findings are deeper psychological currents that the authors identify as trust, autonomy, and public understanding. Misconceptions about what AI can and cannot do continue to circulate widely, and concerns about losing user agency, the sense of being in control of one’s own life and care, remain a powerful brake on adoption. The study’s results suggest that these concerns are not irrational anxieties to be engineered away but legitimate value judgements that any acceptable care technology must respect. Acceptance peaked precisely when AI was positioned to augment human input rather than substitute for it, supporting what the researchers describe as hybrid models. In such models, AI handles scalable, routine tasks such as continuous monitoring, data collection, and early-warning alerts, while human carers retain authority over consequential decisions and provide the interpersonal dimension of care that machines cannot replicate.

The practical implications for NHS England and for digital health developers are considerable. With over half a million recorded diagnoses and pressures on health and social care systems intensifying, scalable solutions are urgently needed, and AI-enabled smart home systems offering cognitive support, real-time monitoring, and decision-making assistance are among the most promising candidates. But the study indicates that the route to adoption runs through augmentation, not replacement. Products that emphasise moderate involvement, that keep humans visibly in the loop, and that are tailored to the specific stage of care a patient occupies are far more likely to win public acceptance than systems promising full automation. Developers who ignore the stage-sensitivity of attitudes, deploying the same one-size-fits-all automation across the entire pathway, risk encountering the sharpest resistance exactly where care needs are greatest.

There are also implications for communication and public engagement. Because misconceptions about AI continue to limit acceptance, improving public understanding of how these systems actually work, what data they collect, and what decisions they can and cannot make could help close the gap between perceived and actual capabilities. The finding that participants rated moderate-involvement scenarios as more likely than full-control ones suggests the public’s intuitions are already broadly realistic, which gives communicators a foundation to build on. Framing AI as a tool that supports carers and preserves patient agency, rather than as an autonomous decision-maker, aligns with both the evidence and the values expressed by the very demographic most likely to use these services.

Ultimately, this study offers a measured but firm message: the future of AI in dementia care in England will be decided not by the sophistication of the algorithms but by the willingness of the public to accept them, and that willingness is conditional, stage-dependent, and firmly anchored in the preservation of human involvement. As AI-enabled care technologies mature, the organisations deploying them would do well to remember that the people they serve, particularly those approaching the age of greatest risk, want machines to lend a hand, not to take the wheel, and they want that human hand to remain steady through every stage of the journey, from the first worrying signs to the final days.

Subject of Research: Public acceptance of AI-enabled versus human-centred dementia care across NHS England's dementia care pathway stages

Article Title: Examining public acceptance of AI versus human-centric dementia care across NHS England’s dementia pathway stages

Article References: O’Fee, T., Vijaykumar, S., & Craig, M. (2026). Examining public acceptance of AI versus human-centric dementia care across NHS England’s dementia pathway stages. PLOS Digital Health, 5(10), e0001088. https://doi.org/10.1371/journal.pdig.0001088

Image Credits: AI Generated

DOI: 10.1371/journal.pdig.0001088

Keywords: artificial intelligence, dementia care, NHS England, public acceptance, digital health, smart home systems, care pathway, human autonomy, hybrid care models, health technology, end-of-life care, public trust