Tool Generates Doctors’ Notes So Well That Physicians Couldn’t Tell the Difference

Researchers at the University of Florida and NVIDIA have developed an artificial intelligence (AI) program, GatorTronGPT, that can generate doctors’ notes with such accuracy that human physicians struggle to distinguish them from notes written by other doctors. In a proof-of-concept study, physicians correctly identified the author of patient notes—some written by human doctors and others generated by the AI program—only 49% of the time.

The researchers trained supercomputers to create medical records based on the GatorTronGPT model, which functions similarly to ChatGPT. The AI models aim to support healthcare workers and enhance efficiency in medical research and care. The study, published in npj Digital Medicine, suggests the potential for AI to play a significant role in healthcare, but the researchers acknowledge the need for extensive data and computing power to build such models. The GatorTronGPT model was trained on de-identified UF Health medical records from 2 million patients, producing clinical text resembling doctors’ notes.