A peer-reviewed PNAS study finds that human-AI collectives diagnose clinical vignettes more accurately than physicians or AI systems alone. The study combines 40,762 physician differential diagnoses from the Human Dx platform with outputs from five large language models across 2,133 case vignettes.
Physicians question the Human Diagnosis Project’s crowdsourced approach despite support from the American Medical Association. Their concerns center on whether the project’s output is validated and on expectations for artificial intelligence in medicine, as the initiative competes for a $100 million grant.
The Human Dx platform combines machine learning with crowdsourced advice from thousands of doctors across 70 countries to help individual physicians improve diagnostic accuracy. It targets specialist access for safety-net hospitals and clinics, drawing on its physician network alongside AI.
