The health minister says a doctor using an artificial intelligence scribe tool is able to see, on average, one additional patient per shift.
Simeon Brown has announced every emergency department in the country now has access to the tool, which records consultations and generates draft clinical notes, referral letters and follow-up summaries.
Association of Salaried Medical Specialists (ASMS) vice-president Dr Sylvia Boys said she was concerned about how secure the artificial intelligence scribe tool was.
It could also misunderstand what was said, Boys added, especially when it came to an examination.
"You have to verbalise what you're finding at the time, and that difference between patient speak with the patient in front of you and the medical diagnosis, AI can sometimes misinterpret what is going on."
It also could not differentiate between patients when a clinician was dealing with multiple, Boys said.
"Within the ED environment, we also have multiple interruptions, and they have to step out of the room, be talked to about other patients, and so separating out what is going on with one patient and what is going on with another - with an IT system that is listening to both - can be troublesome as well."
This is one area where I think AI can make a real contribution; I get that there are major privacy concerns etc...but step back from the detail for a sec and imagine what this could be.
A caveat: I am an AI skeptic, through and through. A lot of the places AI is being shoehorned into are idiotic.
Saying the above, I would love something like this for my sparkies and fitters; rather than them making notes about what they did during the day, they just wear a headset and quickly make a voice note about they are doing, which is auto transcribed and loaded against the work order, when they go to the stores the headset picks up a beacon and the assistant asks what they are getting and what job it is for. So many issues are caused at stores by stuff not being booked out correctly. At the end of the day; they review and correct for 10-15 minutes to ensure accuracy; rather than sitting around trying to remember what they did 8 hrs ago, after 4 breakdowns and six interruptions.
It is unreasonable to have every person with a personal assistant; but basically every job would benefit from having an assistant.
Preferably this would all happen on device; and the reviewed text is all that is sent. Trust would be the biggest hurdle; especially for the teams I work with, a lot of tech skeptics....but they all use facebook!!!!
So the idea of rolling this out firstly in medical contexts is insane, but I agree there is big potential here.
Is "AI" (as in agentic LLM) really necessary here thought?
Imagine exactly your setup, but instead of some LLM you instead have locations of jobs in your system before you start, press a button on your headset or whatever while on site and say things to be logged against the job. Speech to text can generate the text, GPS can apply it to the right job, and key words can be used if you need to do something that isn't just logging notes.
Surely we can go a long way to what you describe before trying to plug an LLM in?