Two AI Systems Beat Doctors in Nature Papers — but a Key Result Warns Specialized Medical AI May Age Poorly as Base Models Improve
Summary
Two papers published in Nature on June 18 show AI agents outperforming physicians: Germany's MIRA system achieved 87.8% diagnostic accuracy versus specialists' 71–78.1% across 311 emergency cases, while Google's AMIE outperformed 21 primary care physicians on treatment appropriateness (95% vs. 72%) and guideline adherence. A critical buried finding: AMIE's specialized medical scaffolding delivered substantial gains over Gemini 1.5 Flash but those advantages 'almost vanished' when applied to newer Gemini 2.5 Flash, suggesting purpose-built clinical AI scaffolding may become redundant as base models improve — a result that complicates long-term clinical AI investment theses.
Originally reported by the-decoder.com
Read the original article →Original headline: Two AI Systems Beat Doctors in Nature Papers — but a Key Result Warns Specialized Medical AI May Age Poorly as Base Models Improve