In a groundbreaking development, a new artificial intelligence (AI) computer program has demonstrated the ability to generate AI Doctors’ Notes so convincingly that human physicians couldn’t distinguish between notes written by their peers and those crafted by the AI, according to a recent study. This proof-of-concept study, conducted by a team of 19 researchers from NVIDIA and the University of Florida, could pave the way for AI to significantly enhance efficiency and support healthcare workers.
In the study, physicians reviewed patient notes, some authored by actual medical doctors and others generated by the new AI program. Astonishingly, the physicians correctly identified the author only 49% of the time. The AI model, named GatorTronGPT, was trained on supercomputers and functions similarly to ChatGPT, a widely used language model.
GatorTron models, available for clinical research on the open-source AI website Hugging Face, have garnered over 430,000 downloads for their free versions. The lead author of the study, Yonghui Wu from the University of Florida’s department of health outcomes and biomedical informatics, emphasized the unique capabilities of GatorTron and GatorTronGPT in powering various aspects of medical research and healthcare.
“In healthcare, everyone is talking about these models. GatorTron and GatorTronGPT are unique AI models that can power many aspects of medical research and health care. Yet, they require massive data and extensive computing power to build. We are grateful to have this supercomputer, HiPerGator, from NVIDIA to explore the potential of AI in healthcare,” Wu said.
To overcome the challenges posed by medical records, which require safeguarding patient privacy and are highly technical, the researchers used health records from two million patients, comprising 82 billion useful medical words. The GatorTronGPT model was trained on this dataset using the GPT-3 architecture, allowing it to generate clinical text resembling medical doctors’ notes.
The potential applications of a medical GPT are vast, with one notable idea involving the replacement of manual documentation with notes recorded and transcribed by AI. Achieving such parity with human writing involves weeks of programming supercomputers with clinical vocabulary and language usage, based on an extensive dataset of billions of words.
As AI continues to make strides in the healthcare sector, the study published in the journal npj Digital Medicine marks a significant step towards harnessing the power of AI to streamline and enhance various aspects of medical research and patient care.
