US Pharma MarketingAI models that can match or outperform physicians on text-based diagnostic tasks should be judged more by how they safely improve patient outcomes in real-world care than by benchmarks or demonstrations, Flinders University researchers argued.
Subjecting AI to human doctor standards?
Flinders University researchers stress that patient outcomes remain the true measure of success, not benchmarks or demos. AI models that can match or outperform physicians on text-based diagnostic tasks should be judged more by how they safely improve patient outcomes in real-world care than by benchmarks or demonstrations, Flinders University researchers argued. A commentary published in Science by Erik Cornelisse, a PhD candidate, and Associate Professor Ashley M Hopkins, a leader of Flinders' Clinical Cancer Epidemiology Lab, examines new evidence that reasoning models can approach physician-level clinical reasoning. Cornelisse and Hopkins wrote the commentary following a Science study showing that…
Source: MobiHealthNews