ChatGPT in Medicine
It didn't take much effort to get used to using ChatGPT. My first introduction to this LLM was through the publication "ChatGPT is nothing if not apologetic." (1). This hilarious way of describing its limits created a lasting impression on me that I would never ignore. The simple reason for that is that any dialogue I would initiate with ChatGPT would eventually end with one of those apologetic responses like: "you tell me", "sorry, I did not know" or "thanks for your patience".
Because the level of "humanness" in being apologetic is near the top of well-behaved, I would excuse ChatGPT every time and generously provide the missing information. My next step would be to shut it down, as my trust would anyway be repeatedly deceived.
This is why, the paper "Benefits, limits, and risks of ChatGPT in medicine" (2) published today in "Frontiers in Artificial Intelligence" turned into a significant subject. As I always do, I checked on the research in detail in order find out that the "experiments" represent those simple dialogues, I was talking about. The scientific assessment would consist in applying statistical assessment of correctness and "humanness" (3) in order to quantify the reliability of the answers.
The final conclusion of this publication is that ChatGPT can only augment the human capital but not assume its role.
Is it really possible to create a LLM bridge to the reputation of a respected medical university professor or general practitioner beloved by the community? And would the proposed regulatory measures (2) be sufficient to create a trustful relationship to those apologetic LLMs?