Americans are increasingly willing to accept artificial intelligence in healthcare, but they want physicians involved in decisions affecting their care, according to two recent national surveys.
An October 29 Pew Research Center report found that 60% of 3,488 U.S. adults surveyed were uncomfortable with AI helping physicians make diagnoses or treatment decisions. However, respondents were more accepting of AI applications when humans reviewed the results.
A November 9 West Health–Gallup report reinforced that distinction. Among 5,330 adults who had visited a primary care provider during the preceding year, 56% were comfortable with AI making diagnostic or treatment decisions when their healthcare provider could override them. Only 9% were comfortable with AI making those decisions without provider review.
Although the surveys asked different questions, both found that patients strongly favored human oversight of consequential medical decisions.
UC San Diego Health Emphasizes Physician Oversight
UC San Diego Health offers a local example of how healthcare organizations are addressing these concerns.
The health system uses AI to draft clinical notes and patient communications and identify patients at risk for sepsis. It obtains patient permission before using AI-assisted note-taking during clinical visits.
UC San Diego Health also evaluates AI applications through a multidisciplinary governance process involving clinical, ethical, safety, equity, and technology experts. The organization monitors AI performance and states that clinicians retain responsibility for medical decisions.
In October, UC San Diego Health announced a collaboration with Harvard Medical School and Anthropic to develop an oversight system for evaluating recommendations generated by patient-facing AI. The technology remains under development.
Implications for Independent Physicians
Independent physicians face similar questions about clinical responsibility, although smaller practices may lack the resources of major health systems.
The findings suggest that medical groups should explain how they use AI, establish procedures for reviewing clinically significant recommendations, verify AI-generated documentation, and monitor systems for errors.
Practices should also clarify who remains accountable when AI contributes to clinical decisions.
Neither survey establishes that physician oversight directly improves patient outcomes or increases trust. Both demonstrate, however, that patients distinguish between AI operating independently and AI functioning under meaningful clinical supervision.
For San Diego physicians, maintaining that distinction may prove essential to adopting AI while preserving patient confidence and clinical accountability.
