SymptomAI: Towards a conversational AI agent for everyday symptom assessment
Google says SymptomAI beat clinician transcript reviews on top-five differential accuracy in a 13,917-person study.
Participants described symptoms to one of five Gemini Flash 2.0-based agents, then some reported later diagnoses from healthcare providers. Clinician reviewers preferred SymptomAI’s differential diagnoses over other clinicians’ in more than half of cases. Agent-led follow-up questions outperformed a base chatbot-style condition. Google also found Fitbit biosignal shifts around respiratory-infection cases flagged by SymptomAI, while stressing the system was research-only and not a confirmed medical assessment. Google Research's note
Participants described symptoms to one of five Gemini Flash 2.0-based agents, then some reported later diagnoses from healthcare providers. Clinician reviewers preferred SymptomAI’s differential diagnoses over other clinicians’ in more than half of cases. Agent-led follow-up questions outperformed a base chatbot-style condition. Google also found Fitbit biosignal shifts around respiratory-infection cases flagged by SymptomAI, while stressing the system was research-only and not a confirmed medical assessment. Google Research's note
score 5