Artificial intelligence is increasingly integrating into daily life, from composing written content and answering inquiries to aiding in the creation of images and software. Recently, researchers have started investigating a crucial new application: can AI assist medical professionals in caring for patients?
A study published in the journal Nature highlights that advanced AI systems might support doctors across many aspects of patient care. The research introduces two innovative AI tools named MIRA and AMIE, both capable of performing various medical tasks. These range from collecting patient information and diagnosing conditions to recommending treatments and planning follow-up procedures.
The significance of these developments is heightened by global healthcare challenges such as doctor shortages, aging populations, and escalating chronic illnesses. Physicians often face long hours and limited direct time with each patient, prompting hope that AI could help alleviate some of these burdens.
Most existing healthcare-focused AI tools target narrow functions—for example, detecting cancer in scans or summarizing patient records. However, real-world medical decision-making involves complex reasoning and multitasking—collecting symptoms, reviewing histories, ordering tests, interpreting results, and adjusting treatments as conditions evolve. Researchers questioned whether AI could handle such multifaceted reasoning.
The first system, MIRA (Medical Intelligence for Reasoning and Action), was developed by a team led by Jakob Kather. Tested with real-world data from over 500 emergency department cases, MIRA interacted with virtual patients—responses modeled after actual clinical notes—by asking questions and gathering information much like a clinician would. It had access to a database of more than 85,000 possible actions, including ordering tests, reviewing results, recommending therapies, prescribing medications, scheduling procedures, and deciding hospital admissions. Impressively, MIRA achieved nearly 88% accuracy in diagnoses, outperforming a diverse panel of six physicians, who averaged about 78%.
The second AI system, AMIE (Articulate Medical Intelligence Explorer), was developed by Google researchers using the Gemini AI framework. Unlike earlier platforms, AMIE could analyze multiple patient visits over time, tracking disease progression and treatment effectiveness. In a comparison with 21 primary care doctors using 100 simulated patient scenarios across five specialties—based on UK clinical guidelines—AMIE performed on par with experienced physicians in overall reasoning. In some instances, it even demonstrated superior precision, especially in testing and treatment recommendations, and adhered closely to established guidelines. Additional testing with a new benchmark called RxQA, focused on medication reasoning, showed AMIE outperformed physicians on particularly challenging cases.
Despite these promising results, experts caution that neither AI system is ready to replace doctors. Medical practice encompasses much more than decision-making from data—communication, understanding individual patient preferences, recognizing atypical cases, and making nuanced judgments in real-world settings are beyond the current scope of AI.
Limitations of the studies include their reliance on virtual scenarios rather than real clinical environments. Future research must establish whether these systems can perform robustly in busy hospitals and clinics with real patients and unpredictable situations.
The emerging view suggests that conversational AI tools might serve as valuable assistants rather than replacements, helping healthcare providers gather information, generate differential diagnoses, confirm treatment guidelines, and streamline routine tasks. If further validation confirms their safety and effectiveness, AI could become a vital partner in enhancing healthcare quality and efficiency, particularly in regions facing healthcare worker shortages.