The VUMC Explores New Generative AI Tools
Nancy Deng | November 10th, 2023
In an era defined by remarkable technological advancements, perhaps one of the most promising and transformative applications of artificial intelligence is its integration into healthcare. Massive amounts of healthcare data are constantly collected by hospitals, most of which are too much for any provider or patient to interpret. To handle this data, doctors and researchers are relying more and more on the use of computers and ultimately, AI. Recognizing the potential of artificial intelligence in the hospital, VUMC has worked in partnership with Microsoft to establish a set of private, institutionally managed, and secure tools like ChatGPT that follow VUMC privacy guidelines.
The VUMC has established aiChat, a generative AI tool, which is available for VUMC personnel to use. Similar to the ChatGPT interface many students are familiar with, aiChat is a VUMC-contained interface which allows VUMC personnel to experiment in a safe environment. aiChat acts as an privatized alternative to public-facing generative AI tools like OpenAI’s ChatGPT and Google’s BARD. Due to HIPAA policies that forbid the sharing of sensitive patient data with third parties, this data is prohibited from traditional AI tools.
Benjamin Collins, a physician-researcher and informatician at the VUMC studying ethical, legal, and social issues of AI in healthcare, discusses the benefits, limitations, and future of AI in healthcare at a TEDx talk. In his talk, he mentions the biggest limitation of AI: the black box problem, or the inability to determine how an AI model makes its decisions. Collins emphasizes that “technology makes it easy to sleepwalk through decisions.” While AI has profound implications for the way data can be processed and interpreted, its use requires caution. The VUMC explicitly states that while aiChat is appropriate for internal use, there are important limitations that must be considered. Similarly to Collin’s argument, aiChat states that “users should not use information derived from the service to inform clinical care.” As advancements in AI are made, it is curious to wonder if doctors will reach the point where they can simply look up and follow the responses provided by a generative AI to treat patients.
The future of AI in healthcare does not appear to be slowing down given the exploration of VUMC-managed generative AI tools. However, despite how exciting these tools can be, it is necessary to consider the limitations of these tools in the case of sensitive information like patient data. As Collins puts it: “…to ensure you receive the benefits of AI, you need to be aware that the quality of AI in healthcare ultimately comes down to human decisions.”
Collins, Benjamin. “Navigating the AI Future of Healthcare.” Youtube, uploaded by TEDx Talks, 7 July 2023, https://youtu.be/gsAlpElHyN0?si=-xSHj2X4Yu2MrNGd.
“Generative AI at VUMC.” Vanderbilt Medical Center Department of Biomedical Informatics. https://www.vumc.org/dbmi/GenerativeAI.
“On the Horizon –VUMC Generative AI Update.” Youtube, uploaded by Vanderbilt Health, 26 July 2023, https://youtu.be/jWgYA9iBf3s?si=dCz_Z8tJoXh98agA.
“VUMC statement on the use of generative AI.” VUMC Reporter. 27 June 2023, https://news.vumc.org/2023/06/27/vumc-statement-on-the-use-of-generative-ai/?_gl=1*1t0ohno*_ga*MzUxNTY2Mzc2LjE2OTA0MjU5NzY.*_ga_CR2EC8KML5*MTY5NjkwMjY1NC4zNi4xLjE2OTY5MDM1MzEuMC4wLjA.&_ga=2.200963435.455380970.1696873081-351566376.1690425976.