Zach Lipton

AI scribing is just that.

It is the idea that an AI agent that is, we call the genre here ambient listening, but that is sort of in the background listening because it's like the doctor has a device, usually their phone,

that has a secure application on it and they say, they consent the patient.

Hey, this is my AI note taker.

Would you be okay if I recorded this conversation?

So, you know, the AI could take the notes and I could focus on, focus on you.

Patient, you know, 99 point, you know, something percent of the time says yes, they hit record.

and it is uh on the basis of the raw audio together with the information that we are able to the relevant context that we could pull from the ehr the ai agent is able to transcribe the audio structure that content collate it with what we already know from the problem list from prior notes and synthesize it all into a kind of comprehensive first draft of the documentation so that

The doctor kind of gets to, like, lift up in the stack from being, like, you know, on the front lines to writing copy to being more of an editor.

You know, when you have a very straightforward classification task, like I have a test for a blood test for HIV.

I want to say how accurate is it?

The straightforward definition of accuracy, which is out of N cases, what fraction are classified correctly?

Then, you know, you have other metrics that you can let you say, what is a recall?

Out of every case that exists, what fraction did I call positive?

Or what is the precision?

Out of everyone that I called positive, what fraction were actually positive?

And these three numbers kind of tell you everything you need to know.

They identify something called the confusion matrix.

And it tells you everything you need to know about the performance of the classifier.

I should caveat that, that even then things can get more complicated in interesting ways because you could say, oh, you tried it on this population, but do those numbers hold on this different