Medical Bayesian Kiosk (2010)

Опубликовано 11 августа 2025, 16:11

The Medical Bayesian Kiosk was an exploratory demonstration developed by Dan Bohus and Eric Horvitz that merged advances from Microsoft Research’s Situated Interaction (microsoft.com/en-us/research/p... effort with AI-powered medical diagnosis. The system integrated multiparty conversational capabilities with sequential diagnostic reasoning, employing inferences from Bayesian networks developed in an earlier collaboration with pediatrics expert Dr. Richard Behrman of Stanford University. The kiosk inferred and updated differential diagnoses as new symptoms were considered, and dynamically selected the most informative next questions to guide a diagnostic session. The demonstration illustrated how a multiparty dialog system could understand and adapt to the ages, relationships, and differing medical knowledge of participants—in this case, a mother and her young son—while engaging them jointly and empathetically about health concerns. It also explored how such technology might be incorporated into broader healthcare workflows and systems.