Labour market modelling suggests that the impact of new technologies will be felt differently across industries and occupations. A 2021 report by PwC predicted the risk of job displacement in health and social care from ‘AI and related technologies’ would be lower than that in many other sectors. In fact, when set against the backdrop of escalating patient demand, the report anticipated that health and social care will see the largest net employment increases of any sector over the next 20 years, with technology largely proving ‘complementary’.
There are several potential factors behind this more positive outlook – including factors that underpin the nature of health care work itself.
Technology struggles to replicate the traits or competencies in health care
Firstly, many tasks in health care are difficult to automate because they require traits or competencies that AI and other technologies currently struggle to replicate. Recent research from Open AI, Open Research and the University of Pennsylvania (2023), for instance, found jobs requiring critical thinking skills are less likely to be impacted by current large language models. Critical thinking is central to much of health care, where staff must weigh up the benefits and risks of different possibilities, approaches and solutions. For example, important nuances may be required in translating patient symptoms into diagnoses and treatments. While AI such as large language models can assist critical thinking – for instance by supporting clinicians’ education, training and professional development or by distilling high volumes of research to generate health advice – this is different from actually doing the critical thinking. Other key competencies needed in health care, such as creativity and negotiating skills, are likewise difficult to automate.
Social and emotional intelligence are also essential components of high-quality care, enabling staff to empathise, communicate effectively and meet patient needs. Analysis by the Office for National Statistics (ONS) in 2019 – which found that medical practitioners were one of the three occupations at lowest risk of automation – noted that health-related words such as ‘patient’ and ‘treatment’ frequently appeared in the task descriptions of jobs at low risk of automation. The ONS suggested this reflects the dimension of ‘working with people’ and ‘the value added by humans in these roles, which is difficult to computerise’. Again, emerging research indicates AI could support empathetic communication – for instance, by generating draft responses to patient questions – but this is different to being empathetic, which requires the ability to read and understand the feelings of other people, and to express and reason with emotion.
Health care is seen as intrinsically ‘human’
A second factor is that in the UK, as in many cultures, health care is seen as intrinsically ‘human’. Given the considerable value attached to the interpersonal dimension of care, some activities – such as communicating a diagnosis of serious illness or comforting a patient – cannot be delegated to machines without undermining the quality and ethos of care. To take another example, while some patients may be happy with AI making clinical decisions in areas such as triage, others might feel that having a human listen to and consider their case is an important component of being treated respectfully and compassionately. Health care is not a product, but a service that is co-designed between professionals and patients and built on trust. Thus, human relationships assume particular significance in areas like care planning, where the need for genuine partnership may pose important constraints on the use of automation.
A 2021 study from the University of Oxford looked not only at which health care activities could be automated, but also at what health care practitioners thought about the desirability of automating them. Interestingly, several activities ranked high for automation potential but low for automation desirability – typically those involving a ‘high level of physical contact’ with patients (such as administering anaesthetics or examining the mouth and teeth). Many health care tasks sit at the intersection of attending to a patient’s physical, mental and social needs, and this likely influences attitudes towards automation.
Even where a task could be automated it doesn’t necessarily follow that it should be. In the study by Open AI, Open Research and the University of Pennsylvania, researchers noted the difficulty of making predictions about the impact of large language models on activities where there is ‘currently some regulation or norm that requires or suggests human oversight, judgment or empathy’ – a description that characterises much of health care.
Few health care roles consist of wholly automatable tasks
A third reason why there is lower risk of widespread job displacement in health care is that automation applies to tasks not to roles per se, and few health care occupations consist wholly of automatable tasks. A Health Foundation-funded study into the potential of automation in primary care found that, while there were a small number of roles (such as a prescription clerk) that were likely to be heavily impacted by automation, no single occupation could be entirely automated.
Where only specific tasks can be automated, staff can adapt by focusing on other tasks or expanding their roles. Research by Goldman Sachs (2023) on the exposure of different industries to automation and generative AI predicted that the occupational categories of ‘healthcare support’ and ‘healthcare practitioners and technical’ will be largely complemented rather than replaced by AI – precisely because of the mix of tasks involved. Similarly, research by Accenture (2023) suggested that, compared to many other industries, a smaller share of health work has high potential for automation, but a bigger share has a high potential for ‘augmentation’ by technology.