SymptomAI: Conversational AI for Everyday Symptom Assessment
A recent study featured on arXiv (2605.04012) presents SymptomAI, a collection of conversational AI agents designed for comprehensive patient interviews and differential diagnosis (DDx). Utilizing the Fitbit app, researchers randomized 13,917 participants to engage with five distinct AI agents. The resulting dataset reflects a wide range of communication styles and a realistic variety of medical conditions. Among the participants, 1,228 received a diagnosis from a clinician, with 517 of these cases undergoing further review by a panel of clinicians over more than 250 hours of annotation. The accuracy of SymptomAI DDx was significantly higher (OR = 2.47, p < 0.001) compared to the control group, highlighting AI's potential in routine symptom evaluation.
Key facts
- SymptomAI is a set of conversational AI agents for patient interviewing and differential diagnosis.
- The study was deployed via the Fitbit app.
- 13,917 participants were randomized to interact with five AI agents.
- The corpus captures diverse communication and realistic illness distribution.
- 1,228 participants reported a clinician-provided diagnosis.
- 517 participants were further evaluated by a clinician panel during over 250 hours of annotation.
- SymptomAI DDx were significantly more accurate (OR = 2.47, p < 0.001) than controls.
- The study is published on arXiv with ID 2605.04012.
Entities
Institutions
- arXiv
- Fitbit