Today’s medical researchers are using large language models (LLMs) and other AI tools to help speed up certain tedious research tasks. Using these tools to process massive data sets, they can move research forward more quickly — from the hypothesis phase, to real-life value for patients.
But how can we make sure AI is used responsibly in this setting, and how is it impacting the medical field on a larger scale?
Dr. Anai Kothari is an assistant professor at the Medical College of Wisconsin (MCW), where he specializes in surgical oncology. He’s also examining AI’s utility as a research tool as the inaugural director of the Bud and Sue Selig Hub for Surgical Data Science, a program created to tackle real-world problems across Wisconsin using data science and AI. He spoke with Lake Effect’s Audrey Nowakowski about AI and healthcare in Wisconsin.
“We need to have experts that understand medical context, working together with both companies, as well as research labs, to think about how we can get the best version of artificial intelligence into our clinical environments,” Kothari says.
He notes that once ChatGPT came out in December of 2022, it transformed how we think about artificial intelligence. "Pre-2022, most of how we thought about artificial intelligence was really response-based modeling... so it was really narrow in the idea that you had to have the data in hand to predict [a] very specific outcome," Kothari explains.
The human effort research traditionally took was also "painful, difficult to do, but it was the best we had." Kothari cites an example of a research who fellow previously spent a year building a model to predict how long a surgery would take.
"At 12 months, we tested 40 different model types and a lot of human work that went into that. The generative AI era came along and he reproduced that same project in about a week," he says. "That pre-[generative AI] era was a lot of manual work, making sure that we understood the labels really well, got the data, trained several different models against it, and now are in a slightly different position to maybe make that same type of discovery work just on a more accelerated timeline."
However with data being processed at a faster rate, Kothari says medical expert oversight is crucial — both for fact-checking and for protecting the privacy of people's medical data being used. Kothari’s role at MCW also includes development of governance policies around AI usage, and says the same privacy guardrails that govern all of MCW’s electronic records still apply when processing patient data with AI tools.
“The message to patients is that a lot of health systems have taken that stance — that we're going to make sure that when it comes to clinical data that’s being interacted with within our boundaries, that we do it in the same way that we have previously with other electronic systems,” he explains.
Human oversight is also very important for controlling bias in AI, Kothari notes. “We still need to keep that human in the loop, assessing and evaluating the outputs for those exact things — bias, things that might be misleading, or could actually cause harm if not carefully checked."
While AI can help accelerate the research process, there is also the question of AI's impact of people's jobs in other medical specialties. Will there be fewer pathologists or radiologists if initial work is done by AI? Kothari doesn't think so.
"Some of it is probably just not gonna happen because you do need that really nuanced expert to be in that process to make sure that the artificial intelligence tool is doing the right thing," he notes. However, when it comes to jobs like medical scribes, you've probably already had your care provider use an AI tool to take notes during your visit.
"I have many trainees who were medical scribes, and yeah, I think that job is going to go away," admits Kothari. "But maybe that's not something we need, and are there better and different ways that they can contribute in that role that would look a little bit differently?"
"This is one of the most challenging parts about the excitement in AI is what is descaling and these aspects, how do they look? And, you know, we have to come up with solutions to that," he adds.
As AI is being used more widely in healthcare, Kothari says it is very important for both care providers and patients to have a strong literacy about AI.
"My biggest caution is ... just talking about this at a depth that makes people understand what's out there, what's coming and feel educated about this wave of new technology," he says.
"Patients have the opportunity, the ability to link their health data to these tools outside the boundary of maybe their routine clinical care. [It's] super empowering. I think it's really important and useful to have that capability," Kothari adds. "These are things that can help, but it makes a huge difference if we can talk about them ultimately and understand what limitations, problems, benefits may be coming from that."
_
