Privacy and hallucinations would concern me. It seems likely to me that relying on these tools will mean physicians will "trust" the output of these systems and stop checking the notes for accuracy. Checking the notes will become a burden too.
Maybe the real solution is to lower the amount of administrative work required? Or hire people to do it?
I too feel a bit worried about AI making mistakes, but the thing is that a doctor rushing or an assistant can (and do) make mistakes too. These are all just cost vs quality of service tradeoffs. From my experience (having been in and out of hospitals a lot in the last year) it wouldn't surprise me if AI, even with occasional mistakes, would still be a net improvement in overall quality of care and outcomes.
In my impression of using AI a lot more recently, hallucinations are a non-issue if you are simply using it to transform data. AI hallucinates when it's being asked to extract data from it's own "memory" but not when it's simply given a task to perform and all the necessary input to do it.
whisper hallucinates - or is just incorrect - fairly often. I have been transcribing old tv and radio programs just to get a real feeling for how inaccurate it is, and for an example, the tv show JAG's main character is Lieutenant Harmon Rabb, Junior. I prompt "You are transcribing a show called JAG - Judge Advocate General - whose main characters are named Lieutenant Harmon Rabb jr, known as Harm or Rabb, [...]" Maybe 1 time out of 20 it will actually put "Rabb" or "Harm".
Even better is stuff like old call in radio shows, the callers are perfectly understandable, but whisper has lots of issues.
If there's something better than whisper, i'd love to try it out.
Maybe the real solution is to lower the amount of administrative work required? Or hire people to do it?