The introduction of large language models (LLMs) such as ChatGPT to the general public has led to the fastest adoption of any technology in history, with individuals frequently reporting using the technology when experiencing medical issues, according to a recent study. Two national surveys examining trust and acceptance of medical artificial intelligence (AI) have found that while most people are reluctant to use AI tools to diagnose their health condition, they see potential in the technology's ability to help diagnose cancer. The findings, published by Sobolev et al in a Center for Economic and Social Research and USC Leonard D. Schaeffer Institute for Public Policy & Government Service working paper, suggest that real-world encounters with generative AI may foster readiness for AI-enabled health care by increasing familiarity, confidence, and nuanced acceptance, according to the study authors.

Study Methodology

The researchers used data collected from the Understanding America Study (UAS), a nationally representative, probability-based panel of individuals aged 18 years and older. For the purposes of this study, two UAS datasets were merged and conducted in two different time points in 2024. Neither survey was designed to exclusively focus on medical AI, but rather on AI adoption in daily life and on acceptance and trust of different scientific advancements and AI technologies.

In the first survey, “AI Adoption and Trust in Medical AI,” the researchers surveyed 10,035 individuals, who ranged in age from 18 to 103 years and were mostly male (48.9%). A second survey, “A Vignette Study of AI for Cancer Diagnosis,” included a subsample of 713 participants who completed both surveys, who ranged in age from 18 to 93 years and were mostly male (53%).

Results

The researchers found that among all participants in the two surveys (10,035), 47.9% had heard of AI chatbots such as ChatGPT that can create human-like text, and 24.3% reported having used the tool. When asked about general trust in medical AI compared to human experts, only 15.1% of respondents trusted AI to provide a health diagnosis as much as or more than a human professional or expert. Men and individuals with a college education were more likely to express trust in an AI-based diagnosis, whereas older respondents reported lower trust.

When exposure to and personal use of AI were included in the analysis, both variables showed strong positive associations with trust: those who had heard of or had used AI were significantly more likely to trust AI for a medical diagnosis than those who had not. Gender remained a significant correlate of trust, while age and education did not.

Among the subsample of 713 participants who read a description of medical AI use in a cervical cancer diagnosis, respondents indicated relatively high potential and excitement, moderate understanding and trust, and low fear of the technology.

Prior exposure to and use of AI were positively associated with higher understanding, trust, potential, and excitement.

“Together, the findings support the vision of human-AI collaboration to augment clinical decision-making and diagnosis and emphasize the need to study the evolving sociotechnical aspects of AI technology,” concluded the study authors.

Michael Sobolev, PhD, of the University of Southern California Schaeffer Center and Cedars-Sinai Medical Center, is the corresponding author of this study.

Disclosure: Funding for this study was provided by the Samuel Oschin Cancer Center at Cedars-Sinai Medical Center. For full disclosures of the study authors, visit papers.ssrn.com.