Hospitals adopt error-prone AI transcription tools despite warnings

https://arstechnica.com/ai/2024/10/hospitals-adopt-error-prone-ai-transcription-tools-despite-warnings/#gsc.tab=0

OpenAI’s Whisper tool may add fake text to medical transcripts, investigation finds.

95 points · 10 comments · view on lemmy.world

10 Comments

ChihuahuaOfDoom@lemmy.world · 16 pts · 1y

Regular transcription software is finally respectable (the early days of dragon naturally speaking were dark indeed). Who thought tossing AI in the mix was a good idea?

SpikesOtherDog@ani.social · 14 pts · 1y (2 replies)

I work in judicial tech and have heard questions of using AI transcription tools. I didn't believe AI should be used in this kind of high risk area. The ones asking if AI is a good fit for court transcripts can be forgiven because all they see is the hype, but if the ones responding greenlight a project like that there will be some incredibly embarrassing moments.

My other concern is that the court would have to run the service locally. There are situations where a victim's name or other information is redacted. That information should not be on an Open AI server and should not be regurgitated back out when the AI misbehaves.

FatCrab@lemmy.one · 3 pts · 1y (1 reply)

Don't court stenographers basically use tailored voice models and voice to text transcription already?

SpikesOtherDog@ani.social · 2 pts · 1y

I don't get too technical with the court reporter software. They have their own license and receive direct support from their vendor. What I have seen is that there is an interpreting layer between the stenographer machine and the software, literally called magic by the vendor, that is a bit like predictive text. In this situation, the stenographer is actively recording and interpreting the results.

EleventhHour@lemmy.world · 10 pts · 1y

Private hospitals care about only one thing: profit. These error-ridden tools serve that purpose.

SubArcticTundra@lemmy.ml · 10 pts · 1y (1 reply)

God I hope this isn't the AI plan that the NHS adopts

ladicius@lemmy.world · 9 pts · 1y

This is the AI plan every healthcare entity worldwide will adopt.

No joke. They are desperate for shit like this.

ShareMySims@sh.itjust.works · 5 pts · 1y

Errors and Hallucinations are definitely serious concerns, but my biggest concern would be privacy. If my GP is using AI, I no longer see my medical information as private, and that is unacceptable.

FigMcLargeHuge@sh.itjust.works · 4 pts · 1y

If anyone needs to know the state of AI transcription, just turn on closed captioning for your local tv channel. It's atrocious and I am sorry that people who need closed captioning are subjected to that.

sgibson5150@slrpnk.net · 1 pts · 1y

Years ago, I worked in a tech role at a medical transcription company. It hadn't occurred to me that AI would render their jobs irrelevant. This used to be an area where women in particular could make decent money after a bit of training, and there were opportunities for advancement into medical coding and even hospital administration.

I worked with some good people. Hope they landed on their feet.