Like it said “patient states that these symptoms began when she was a little girl”. I never use “little girl” when describing myself as a child. I said “kid”. Maybe I’m being overly suspicious but it seems like a computer at a women’s health visit would be more likely to change “kid” to “little girl” than a grown ass gynecologist would.

I asked if Elon Musk or any other fascist would have access to it and he laughed and said he wouldn’t use it if it did. He claimed it’s only in their internal system so I said ok.

But don’t the providers prefer the human scribe because they also serve as a chaperone/witness?

But like if you've ever used speech recognition either on your phone or via a tool like Dragon (which is what most clinicians use) you've almost definitely used this tech before

And, yes, it would fall under the term "AI" and operates in a manner rather similar to something like an LLM, albiet at a much smaller scale and for a specialized purpose

Here's an IEEE paper on exactly what I'm referencing from *1995*
ieeexplore.ieee.org/document/479408

non-deterministic language models for voice recognition have existed at least since the 90s

Frankly, I'm surprised & **disappointed** by your eagerness to jump to conclusions and make biased inferences. Eg: "an AI scribe will change how physicians speak", but *character limits don't impact how ppl write here*. Sets how seriously I should take this.

My inference: you've had minimal input from actual providers familiar w/ the app (point #4 and 7 were dead giveaways) or who spent >10,000 hours writing notes (even #9 seems to be from a non-provider).

No thank you.

Guilting you is not a good sign. I clearly don't know all the facts, but trust your feelings and don't let someone pressure you.

You might send copies of Dr. Bender's and Dr Gooch's (elsewhere in this thread) essays to him and suggest you are trying to help him with his workload by not letting him get sucked into "AI" hype.

This part struck me because I hadn't even considered the problems trying to use notes you didn't write. Its an extra chance for misunderstanding.

And the doc said the notes are accurate earlier in the piece, but here he admits he can't be sure about that.

@Kierkegaanks @starluna @WhiteCatTamer @emilymbender

I've just passed your paper and Dr Gooch's along to the most recent Dr to ask me about using an AI scribe. Give them some heads up, as well as data when responding to management.

That's it! Thanks, and clearly you've already seen it. And I miss remembered where I saw it.

Is it this one?
https://mastodonapp.uk/@LPhilpott/116453668612462241

I just read (in a JAMA newsletter, I'll try to track it down -- it's not in my email or trash) about a Doctor who as been an early adapter. He did it "right", going over the notes in the evening to clean up the errors in transcription.

He found:
1) He could just focus on the patient, rather then the screen.
2) He got off track and was less focused, and spent more time with the patients without providing better information.
3) Most importantly, when someone came back 6 months later for a follow up, he realized that the notes were not that good. Accurate, but without insight -- they read like someone else had written them and did not help him recall what was going on.

As some others pointed out - that evidence is based on recorded encounters being transcribed by humans, or possibly by non-AI speech-to-text.

Nobody has yet studied the resulting practice effects of AI/LLM-based interpretation. Recording, transcription, and interpretation are separate realms and skill sets and “AI” is either unnecessary or unproven in all 3.