AMIE, Google's medical AI, matches doctors
A study published in *Nature* shows that AMIE, Google's medical AI, matched **21 primary care doctors** in the reasoning used to manage diseases. The system performed better on plan accuracy and alignment with clinical guidelines, although it is not yet ready to treat patients on its own.
Google has tested AMIE, its medical AI system, on a task that begins after diagnosis: helping manage a disease over time. In a study published in Nature, the system matched primary care doctors in the overall reasoning behind their plans and performed better on accuracy and alignment with clinical guidelines.
Managing a disease involves much more than identifying what is wrong with a patient. You need to track changes in symptoms, review treatments, adjust medications, and apply recommendations that may change. This work is usually spread across multiple appointments and a large amount of medical documentation.
How AMIE works
AMIE combines two components based on Gemini models:
- A conversational agent that speaks with the patient in real time and maintains an empathetic dialogue.
- A reasoning agent that consults hundreds of pages of clinical guidelines and treatment formularies, the documents that set out which medications can be used and under what conditions.
The idea is for both to work together. While one component gathers information from the patient, the other compares that data with the available medical knowledge to develop a follow-up plan.
What the study showed
The research was blinded and used actors playing the role of patients. Specialist doctors compared AMIE's responses with those of 21 primary care doctors.
The system did not outperform the professionals in every area. In the overall reasoning used to manage the cases, its results were similar to those of the doctors. However, it scored significantly higher in two specific areas: the accuracy of its plans and their alignment with clinical guidelines.
That does not show that AMIE is ready to treat patients on its own. The study took place in a controlled environment and does not amount to a test of routine use in hospitals or medical practices.
What could change for you
If these kinds of systems work in real-world settings, they could help prepare for appointments, summarize how a disease has progressed, and check whether a plan follows the latest recommendations. For someone with a chronic condition, that could mean more organized follow-up and less time for doctors to spend reviewing documentation.
The final decision would still belong to a professional. AI can find information and point to options, but it can also misinterpret a symptom, overlook an important detail, or apply a guideline out of context.
Google says it is now studying how to bring AMIE into clinical settings and has launched a national study to evaluate AI in virtual care with real patients. The next step will be to see whether the results obtained with actors hold up when records are incomplete, unexpected changes occur, and medical decisions have real consequences.