This video showcases the Medical Bayesian Kiosk, an exploratory demonstration developed by Dan Bohus and Eric Horvitz at Microsoft Research in 2010. The system integrates advanced AI-powered medical diagnosis with sophisticated multiparty conversational capabilities. Utilizing Bayesian networks developed in collaboration with Dr. Richard Behrman, the kiosk dynamically infers and updates differential diagnoses based on new symptoms, intelligently selecting the most informative next questions. The demonstration illustrates the system's ability to understand and adapt to the ages, relationships, and varying medical knowledge of participants—such as a mother and her young son—while engaging them empathetically about health concerns. The project also explored the potential integration of such technology into broader healthcare workflows.
Summarizing the Medical Bayesian Kiosk (2010) Demonstration:
0:12 Introduction: The "Triage Kiosk," developed by Microsoft Research (Dan Bohus and Eric Horvitz) in 2010, introduces itself as being in touch with "the best doctors in the world" and asks the user's purpose.
0:24 Patient Information Collection: A mother indicates she is there for her child, Manov, who is six, almost seven years old.
0:26 Symptom Elicitation (Initial Complaint): The mother states her child has been experiencing diarrhea.
0:45 Targeted Questioning (Sequential Reasoning): The kiosk proceeds to ask a series of diagnostic questions, dynamically adapting to the respondents (asking Manov if his "tummy has been hurting" after the mother initially reported no complaints about abdominal pain).
1:32 Preliminary Diagnosis & Recommendation: Based on the input (diarrhea, slight fever, tiredness, stomach pain, darker stool, no vomiting or weight loss), the kiosk provides an initial assessment of "not that concerned" and suggests scheduling a follow-up appointment with a doctor for confirmation.
1:45 Appointment Scheduling: The kiosk proposes a specific time (Friday at 10:00 a.m.) for the appointment, which the mother accepts.
1:50 Empathetic Closing: The kiosk concludes the session empathetically, wishing Manov well.
Key Capabilities: The system demonstrates AI-powered medical diagnosis using Bayesian networks, multiparty conversational understanding, dynamic question selection, and adaptation to user profiles (e.g., age, relationship, medical knowledge).
Broader Vision: The project aimed to explore how such AI technology could be integrated into healthcare workflows to enhance the diagnostic process.