Medical AI assistance effectiveness differs by user expertise

A recent study published in *Journal of Medical Internet Research* examined how different user groups respond to diagnostic suggestions generated by large language model (LLM) systems. Researchers presented a series of clinical vignettes to two cohorts—medical professionals and lay participants—and asked them to make diagnostic decisions with or without AI‑generated assistance. When the LLM provided an incorrect diagnosis, non‑expert participants accepted the suggestion in more than 60 % of cases, whereas practicing clinicians identified and overrode the erroneous advice in roughly 78 % of instances.

The findings highlight a potential risk associated with the deployment of LLM‑based decision‑support tools in patient‑facing contexts. While clinicians demonstrated the ability to recognize and correct AI mistakes, the tendency of non‑experts to defer to the technology—even when it is inaccurate—could lead to misdiagnoses and inappropriate care. The authors recommend that developers incorporate explicit confidence indicators and that healthcare systems implement safeguards, such as mandatory clinician review, before AI recommendations are acted upon.