UF and NVIDIA advance AI-generated medical notes

This research, showcasing the transformative potential of AI in healthcare, involved an assessment by physicians who reviewed patient notes. These notes were either authored by medical professionals or the AI. The assessment found that the physicians were only able to correctly identify the author 49% of the time, highlighting the AI's exceptional proficiency.

This research, conducted by a team of 19 experts from NVIDIA and the University of Florida, has been published in Nature Journal npj Digital Medicine – the study provides a glimpse into a future where AI could greatly enhance healthcare efficiency.

The AI, developed on the GatorTronGPT model and similar to ChatGPT, underwent rigorous training on supercomputers. The GatorTron model, available on the Hugging Face open-source AI platform, has garnered over 430,000 downloads.

Dr. Yonghui Wu from the UF College of Medicine has emphasised the complexity and innovative nature of these AI models, and acknowledges the crucial role of NVIDIA's HiPerGator supercomputer in their development.

“In health care, 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 health care.”

The Malachowsky Hall for Data Science and Information Technology, named after UF alumnus and NVIDIA co-founder Chris Malachowsky, represents a $150 million investment, signifying a successful public-private partnership between UF and NVIDIA. This collaboration was further strengthened by a substantial investment in 2021 to upgrade UF's HiPerGator supercomputer.

The AI's training involved processing and analysing anonymised data from 2 million UF Health medical records, incorporating a staggering 82 billion medical words. The GatorTronGPT model was then trained with a combined dataset of 277 billion words.

The application of this AI tool in healthcare holds the promise of substantially reducing the burden of routine tasks, such as paperwork and documentation – potentially reforming the way healthcare professionals work by introducing new efficiencies and assistance from AI into their daily operations.

The success of this study is attributed to the collaborative efforts of a diverse team, and is further supported by grants from the Patient-Centred Outcomes Research Institute, the National Cancer Institute, and the National Institute on Aging. This is a collaboration thatboth shows the capabilities of AI in the medical field and also lays the groundwork for future advancements in AI-assisted healthcare.