The fundamental problem that I don't really ever see touched on either by the loud "pro" nor "anti" use crowds, is that both seem to presume that the quality of care was good & high, with the medical sphere largely absent of biases that negatively impact patient care, before LLMs began to be used in said sphere.

This is why those loud people who are against LLMs in said sphere claim that the only thing that's needed is to remove it & similarly those who are in favor of it, just believe it needs more expansion.

The thing is, even without it's wrong guesses/extrapolations, which definitely harm people, with messes like the Replicability problem being widespread, and many processes that amount to just extra paperwork, being pushed without any evidence of efficacy (they were never even studied), the medical sphere is full of biases. This is why if all these LLMs did was to get fed all the existing medical care information & patterns, then give out all the most common &/or average responses, as a continually responsive machine, it wouldn't be "good", rather it'd be a mess.

#MedicalCare #AI #LLMs #QualityOfCare