Google Deepmind's "AI co-clinician" beats GPT-5.4 in blind doctor tests but still trails experienced physicians

Editor's take

Google DeepMind's experimental AI co-clinician demonstrated diagnostic accuracy surpassing a hypothetical GPT-5.4 in simulated patient scenarios, though it fell short of seasoned human doctors. This development highlights the incremental progress in specialized AI for healthcare, a sector grappling with physician burnout and increasing patient loads. The implications extend to how AI can augment, rather than replace, human expertise, potentially democratizing access to sophisticated diagnostic support in underserved areas if scaled effectively.

The continued gap between the AI and experienced physicians underscores the nuanced understanding and implicit knowledge that human clinicians possess, particularly in complex or ambiguous cases. Further research will need to investigate how the AI's error patterns differ from human physicians and whether these differences represent critical safety concerns. Observing how DeepMind integrates this technology into clinical workflows and addresses regulatory hurdles will be key to understanding its real-world impact.