A recent Nature Medicine study investigated the public perception of artificial intelligence (AI)-based tools designed to provide digital medical advice.
Study: Influence of believed AI involvement on the perception of digital medical advice. Image Credit: MUNGKHOOD STUDIO / Shutterstock.com
The role of AI in medicine
To date, several AI-based systems have been developed for medical purposes. For example, AI-based tools enable the analysis of medical images, such as X-rays and magnetic resonance imaging (MRI) scans, and the prediction of drug interactions.
Recently developed AI-based large language models (LLMs) have been used to generate medical advice. For example, ChatGPT, a popular LLM application created by OpenAI, offers medical information without consulting professional physicians. ChatGPT, particularly the ChatGPT 4.0 version, is associated with high accuracy in diagnosing disease.
One previous study revealed that clinicians assessed responses to medical queries generated by LLMs and considered these answers to be of high quality. In fact, the responses generated by AI-based LLMs were considered to be more empathic than answers provided by human physicians. Importantly, none of the clinicians in this study were aware of the authorship of the responses they evaluated.
Even after generating high quality data, significant reservations have been observed among different stakeholders in using AI-based applications. Thus, it is imperative to assess how the general public perceives AI-generated healthcare advice.
About the study
The current study used two experiments to explore how the public reacts to LLM-generated medical advice in a controlled experimental setting. Whereas the ‘study one’ cohort comprised 1,050 participants from various nationalities, the ‘study two’ cohort included 1,230 participants from the United Kingdom. These cohorts were used to assess how identical medical advice labeled as human, AI, or human physician + AI was perceived by the participants.
The human physician + AI label, which included information generated by a human physician in collaboration with AI, was developed based on the assumption that AI will not replace but support human competencies in the future. In study two, an individual’s willingness to follow the provided medical advice was evaluated. The study participants’ desire to test the AI tools that offer medical advice was also assessed.
Study findings
The study findings indicate that the general public perceives physicians as the most authentic source of medical information compared to AI-based tools. ' Human physician’s advice was perceived as significantly more empathic than advice provided by ‘AI’ and ‘human physician + AI.’ Similarly, ‘human’ advice was rated as significantly more reliable than ‘AI’ and ‘human physician + AI’ advice.
Mixed-effect regression analyses indicated that the comprehensibility ratings were not affected by the author's label. In general, the study participants were significantly less willing to follow medical advice believed to be generated by AI tools.
Consistent with previous reports, the current study highlighted the importance of mutual demonstration of care and respect achieved through patient-physician interactions. The aversion to AI-based tools for medical information could be attributed to the perception of these tools as ‘dehumanizing,' which is reflected through lower empathy scores for AI-labeled advice. Another reason for resistance to AI-generated medical advice could be due to ‘uniqueness neglect,’ in which patients perceive that AI may fail to consider an individual’s unique characteristics.
Conclusions
Medical advice provided by ‘human physicians’ was perceived as more empathic and reliable but not as comprehensible as ‘AI’ and ‘human physicians + AI’ advice.
The current study had certain limitations. For example, all study participants were asked to adopt the perspective of other individuals, which meant they could not formulate their inquiries.
Furthermore, the assessed dialogs had only one question and an associated response. Therefore, this study's experimental setting failed to capture the extensive interactions that typically occur in face-to-face doctor-patient consultations. Future research should consider more interactive and less controlled environments.
Consistent with previous research, the study findings suggest a bias against medical advice labeled as AI-generated, irrespective of whether it is supervised by a human physician or not. This raises notable concerns, especially considering rapid advancements in the use of AI in healthcare and the potential for human-AI collaboration.
To mitigate public concerns, a larger group of stakeholders will need to be engaged, including insurance providers and physicians. The issue of framing the involvement of AI in delivering medical advice is crucial, as a recent study has shown that when patients are convinced that human physicians would remain unequivocally involved in the decision-making position, there is a higher level of trust in medical advice offered.
Journal reference:
- Reis, M., Reis, F., & Kunde, W. (2024). Influence of believed AI involvement on the perception of digital medical advice. Nature Medicine; 1-3. doi:10.1038/s41591-024-03180-7
Metabolic imaging detects ovarian cancer subtypes and predicts chemotherapy response