It requires physicians to read and verify the LLM output before signing it. And not just rubber stamping.
There is a precedent for this with speech recognition, which a lot of doctors use to dictate their documentation. Speech recognition isn't completely reliable either, so the doctor needs to be diligent to review before signing. And not just assume it's all correct.
> It requires physicians to read and verify the LLM output before signing it. And not just rubber stamping.
Then what's the point? That's not going to provide a "more efficient system and greatly improve job satisfaction for physicians."
AI would only improve the situation if it can actually be trusted. You're actually making things far worse if you replace a "production" task with a "continuous monitoring for errors. Humans are terrible at the latter.
> There is a precedent for this with speech recognition, which a lot of doctors use to dictate their documentation. Speech recognition isn't completely reliable either, so the doctor needs to be diligent to review before signing. And not just assume it's all correct.
The one transcript like that I've seen had obvious speech recognition errors, when the notes got to my primary care doctor.
It requires physicians to read and verify the LLM output before signing it. And not just rubber stamping.
There is a precedent for this with speech recognition, which a lot of doctors use to dictate their documentation. Speech recognition isn't completely reliable either, so the doctor needs to be diligent to review before signing. And not just assume it's all correct.