Technology

The description sounds more like an AI receptionist than an AI nurse. It would be helpful if patients could ask follow-up questions to the automated phone call before an appointment. Some clinics don't have the manpower for that, and especially not in all the languages that the local population might speak.

I'd be interested in seeing how good the model actually is, and how it determines when to pass it along to a human

The concern is with making sure the AI model is only used where it makes sense. Those who are looking to cut costs will try and use it everywhere, and that needs to be kept in check

I am sure this "nurse" will do everything in it's power to keep you from talking to an expensive real human. It will have a very soothing voice while extracting as much value from you as possible after it examines the insurance you have.

I assume that this is using a highly-curated, custom model, and not some off-the-shelf GPT that just anybody can use, so it probably won't be suggesting that patients eat glue or anything crazy.

From what I can tell, it sounds like this is actually a fairly valid use for a chatbot, handling a lot of the tedious tasks that nurses are charged with. Most of what it seems to be doing, any untrained receptionist could also do (like scheduling appointments or reading dosage instructions), so this would free up nurses for actually important tasks like administering medications and triaging patients. It doesn't seem like it's going to be issuing prescriptions or anything where real judgement would be necessary.

As long as hospital staff are realistic about what tasks the chatbot should handle, this actually seems like a pretty decent place to implement a (properly-tuned) LLM.

Why don't I just Google. It's the same data after all.