Clinicians' attitudes to AI and expected use
In 2024, clinicians were aware of AI far more widely than they used it in practice. They expected it to speed discovery and lift menial work, feared misinformation, critical errors and the erosion of their critical thinking, and wanted guardrails before relying on it, and their institutions had told them little about how it may be used.
Awareness and use of AI
Elsevier's 2024 study of attitudes towards AI drew over a thousand responses from clinicians in 80 to 85 countries and asked a large group of researchers many of the same questions. It was released in July 2024. Later in 2024, its authors suspected that its figures would by then have shifted considerably, the field being fast-moving. It was aimed at practising clinicians, many of whom also did research.
Ninety-five per cent of clinicians described themselves as aware of AI in some form, against 97% of researchers. Only 8% called themselves very familiar with it, having used it a lot, against 14% of researchers. Fifty-one per cent of clinicians said they had used AI, but nearly a quarter used it in their practice, against 37% of science and technology researchers. The study's authors suggested that the lower figure reflected the carefulness to be expected of clinicians. Sixty-six per cent of clinicians expected to use AI over the next two to three years, and 71% expected its impact on their area to be significant or transformative. Throughout, the researcher figures were a little higher than the clinician figures, including for expected use and impact.
Clinicians' use of the different AI tools was relatively low. ChatGPT was the tool clinicians were most familiar with and had used most, at 19%, and awareness and use dropped off considerably for Bard, Bing Chat and Microsoft Copilot. Below the top ten, which ended at about three or four per cent, came a long trail of other tools, and use was generally quite low at the time.
Expected benefits and harms
Clinicians' overall attitudes to AI were mixed but generally positive: they perceived many benefits, but believed false information would wreak havoc. One clinician's comment, which the study's authors took to reflect clinicians' general thinking, accepted AI as the future but found it very worrying.
The benefit clinicians mentioned most was the acceleration of knowledge discovery, much as researchers did. Ninety-two per cent expected AI to save money for federal organisations and for business. They also expected it to raise the quality of their work and leave them more time for higher-value work.
Among the disadvantages, clinicians feared that AI would be used for misinformation broadly. They were concerned that it would cause critical errors, mishaps or accidents, in which AI, probably used for good reasons, might hallucinate or misinterpret something, with a particularly severe impact. Clinicians in particular were concerned that AI would erode their critical thinking and be used to make clinical decisions, and they believed human oversight should be involved. They also expected a lot of disruption to the scientific process and, more broadly, to society.
Tasks clinicians would use AI for
Clinicians broadly supported the use cases the study explored, and their comments described AI as alleviating menial tasks. Clinicians were asked which tasks would benefit most from AI in publishing, executing research, the consumption of scientific content, teaching, clinical activities and funding activities. In publishing, they named formatting the manuscript and perhaps help finding a journal. In doing research, they named writing code, checking data against prior research, generating new research hypotheses, searching the literature on very specific topics and summarising the clinical evidence. In teaching, they named identifying real-world examples to use. In clinical activities, they named reviewing a patient's history across multiple data points and helping to identify the correct treatment approach.
Offered a trusted AI system that assessed symptoms and identified possible disease conditions, 94% of clinicians said they would use it and would probably act on its recommendations for confirmatory tests. Approval was generally high across a number of such use cases. Clinicians held strongly that AI is a tool and a means to an end, which helps them make a better-informed decision by making the data quickly available to them.
Conditions for trusting AI
Clinicians showed a strong need for guardrails before they would trust AI more. They accepted that they needed to change, but did not want to lose the relationship with the patient, and needed AI tools to be high-quality and ethical. Asked what they would need in place to be comfortable using AI tools, clinicians put being paired with human expertise at the top, at 86%. They also wanted to be able to turn off a tool's AI functionality and to be told when a tool is using AI. Seventy-three per cent expected the tools to be based on high-quality content rather than on content generally collected by large language models.
In 2019, healthcare professionals felt uncomfortable about relying on AI-powered software, a gap that one article held needed to be bridged.
Institutional guidance on AI use
Institutions had generally communicated little to clinicians about the rules and regulations for using AI in their work. Over a third of clinicians were not aware of any restrictions their institutions were introducing, a share a little higher than among researchers. Nearly half were not aware of any plans their institutions were putting in place, and a further 14% said there were no preparations at all. Clinicians said they needed and wanted their institutions to put plans in place. The gap they named was expertise in how to use AI in their institution, which they saw as limiting their use of it. The study's authors suggested that it is for organisations and institutions to work out the guardrails for AI in the cases where it helps clinicians. They expected these guardrails to evolve and become stronger over time, and clinicians' uncertainty about what their institutions were doing to clear probably within the next two to three years. They reported that many institutions had moved their guidelines forward since the study.
Differences by region
Clinicians' views differed little by therapy area, and the differences by geography were much greater. Clinicians in the United States, perhaps the country at the forefront of AI, were perhaps the most conservative about its benefits and the most critical of how it could be used negatively. Europeans also scored high on the negative components, though lower than Americans, while clinicians in Asia-Pacific embraced AI much more than either.
The medical writer's role
One view in 2024 was that knowing clinical researchers' perspective on AI helps medical writers understand their expectations, assumptions and misassumptions, and that there is a role for organisations and their membership in taking a leadership position in building up expertise to help educate the researchers they work with.