Andrew Gonzalez, M.D., J.D., MPH, proposes to create a chatbot that allows clinicians to interact with patient data in real-time and in a dynamic fashion.
Transcript:
For the NAM fellowship (National Academy of Medicine, which is part of the National Academies of Sciences, Engineering, and Medicine), I propose to create a chatbot that allows clinicians to interact with patient data in real time in a dynamic fashion. It’s very simple, easy to get used to. You ask it questions, just like we’re all familiar with the local bank’s chatbot or buying tickets at the movie theater or whatever it happens to be.
That chatbot in the background has a patient’s data loaded into it. So in the same way, I would go to a resident of mine or a med student who has prepared the patient case for clinic, and ask them, “When was the patient’s first surgery?” “What’s the patient’s heart failure status?” I can interact with the chatbot in the same way.
So the challenge there is that even though we’ve all played around with large language model-enhanced chatbots, each of those bots has to be fine-tuned to a particular task or a particular domain. So that process is called domain adaptation.
So if you take the chatbot at your bank to figure out what the balance of your savings account is and start asking it questions about peripheral arterial disease, it’s not going to do a good job. So taking the baseline chat bot and domain adapting it to peripheral arterial disease is essentially the task of my National Academy of Medicine award.
