In a significant trial involving 100 patients, Google’s conversational AI system, AMIE, demonstrated its capability to conduct pre-visit medical interviews and generate diagnostic insights comparable to human physicians. This study marks a potential shift in how AI could streamline clinical workflows in urgent care settings.
Published recently on the arXiv preprint server, the research assessed AMIE’s performance in a live clinical environment at an ambulatory primary care clinic within Beth Israel Deaconess Medical Center. The trial involved adult patients scheduled for non-emergency urgent care visits, who interacted with the AI system through secure text-based chats up to five days prior to their appointments.
Throughout the interactions, a board-certified internal medicine physician monitored AMIE in real-time, ensuring patient safety and providing clarifications as needed. Notably, the study found no safety incidents requiring intervention during all 100 patient interactions, indicating that the AI can operate safely under supervision.
AMIE exhibited a marked improvement in patient attitudes toward medical AI, as indicated by the General Attitudes toward AI Scale (GAAIS), with scores showing a significant positive shift following the interactions (p < 0.001). Blinded physician reviewers later evaluated the differential diagnoses generated by AMIE, finding them to be of comparable quality to those produced by human clinicians.
While AMIE performed effectively in generating diagnostic insights, human clinicians outperformed the AI in crafting management plans that were practical and cost-effective. This disparity underscores the importance of contextual patient information, which human doctors utilize but was less accessible to the AI during the study.
AI’s Role in Addressing Physician Shortages
The increasing integration of AI tools in healthcare comes at a time when many nations are grappling with a shortage of primary care physicians. As workloads surge and burnout rates among medical professionals climb, researchers are actively exploring how advanced computational solutions, particularly large language models, can alleviate these systemic pressures.
Previous studies have shown promise for AI systems like AMIE in controlled environments, indicating their potential to engage in nuanced clinical reasoning. However, critics caution that real-world clinical practices often present challenges that standardized simulations cannot replicate. Patients possess diverse communication styles and varying levels of health literacy, factors that must be considered before widespread AI adoption in clinical settings.
The current study is pivotal as it validates AMIE’s application in a busy primary care clinic. The AI not only gathered patient histories effectively but also produced clinical summaries that were shared with the attending physician prior to the patient’s visit.
After the consultations, an independent panel reviewed the accuracy and safety of the management plans devised by AMIE and compared them to those created by human clinicians. Evaluators found no significant differences in the quality of diagnoses (p = 0.6) or safety of management plans (p = 1.0). However, they noted that human clinicians excelled in creating more practical (p = 0.003) and cost-effective (p = 0.004) plans.
The findings indicate that AMIE can serve as a valuable supervised clinical assistant, effectively collecting patient information and aiding in diagnostic processes. While the technology has not yet reached a stage where it can operate independently, this research supports the perspective that AI can complement clinical workflows rather than replace human practitioners.
As the healthcare landscape continues to evolve, further research is essential to ensure the safe integration of AI into routine medical practice. Larger studies across multiple sites will be necessary to confirm the effectiveness, safety, and generalizability of AI applications in diverse patient populations.
Source: Brodeur, P., et al. (2026). A prospective clinical feasibility study of a conversational diagnostic AI in an ambulatory primary care clinic (Version 2). arXiv. DOI, 10.48550/ARXIV.2603.08448, https://arxiv.org/abs/2603.08448v2
See also
Norm Ai Launches Legal AGI Lab to Shape AI Governance in High-Stakes Industries
NSWCPD Applies AI to Predictive Maintenance, Enhancing Submarine Machinery Health
70% of Logistics Professionals Anticipate AI Transformation in Transport by 2026, Says Microlise
OpenAI Launches GPT-Rosalind to Accelerate Drug Discovery and Biological Research
UNC’s AI-Enabled Ultrasound Gains FDA Clearance to Enhance Maternal Healthcare Access
