ResearchApplications 🇺🇸 24.07.2026 02:05

SymptomAI: A Conversational AI Agent for Initial Symptom Assessment from Google Research

Google/DeepMindGoogle/DeepMind
Researchers at Google Research have introduced SymptomAI, an experimental conversational AI agent based on Gemini Flash 2.0 for symptom assessment and differential diagnosis. In a national study involving 13,917 participants, the system demonstrated differential diagnostic accuracy at least as good as real physicians in over 50% of cases, with agentic questioning strategies significantly outperforming passive mode. Additionally, AI diagnoses correlated with physiological signals from wearable devices (Fitbit), opening possibilities for large-scale research.
Google Research has published a study on SymptomAI, the first-of-its-kind conversational AI agent for differential diagnosis and symptom checking as part of a national study. 13,917 participants described their symptoms to one of five agent variants based on Gemini Flash 2.0, which differed in the degree of freedom in conducting the interview. Two weeks later, participants reported diagnoses received from their physicians. Clinical expert evaluation showed that SymptomAI's differential diagnoses were deemed more accurate by physicians than those made by other clinicians in over half of the cases. All active interview agent strategies significantly outperformed the baseline no-prompting mode. A correlation was also found between AI diagnoses of infectious diseases and changes in physiological signals from Fitbit (cardiovascular, respiratory, skin temperature, sleep quality) several days before the consultation, confirming the objectivity of the assessment. The researchers note that SymptomAI is a research tool and its diagnoses are not official clinical conclusions.
Source: Google Research — original
Our earlier posts on this topic ↓
Fresh news