So I've got to sit down and sign these, sign these, co-sign these notes, read them over, et cetera, write notes. And the documentation time that it takes to do all this and now spend an hour, another hour or so. You know, over time, we've gotten used to, I think we've gotten used to documenting we into the night, taking our documentation home, et cetera.

But is that the way it should be? Well, you know, thankfully, you know, we've come to the point where we want to document medicine or is it documenting instead of medicine? So that's kind of what we're where we are at this point. When artificial intelligence c has come in, it helps us analyze all this information that's coming, as I mentioned. You know, the access to the information that it

Or enhances the communication, it checks on the quality, and even helps us to interpret some of the data that's coming in. And at that time, we get to sit down and analyze it. Now, when it was just the electronic medical record, as you see there, we had to do all this stuff. It wasn't, we were the ones pulling the data right.

and so forth. So you know when I was first and went to medical school and first started my residency, we're still on paper, you know, on most of it. And believe it or not, that was so much faster. Like, you know, now granted our notes were a little different. I I I will I will admit that you know patient five was one of my notes. But

But like the level of detail that that is required now is onerous. And so you find that for every one hour that you spend with a patient, you're spending about two hours of documenting. That's that's what came out in the analysis of internal medicine several years ago. and people have been advancing, as I mentioned, more and more how much time they're spending at home afterwards. So

You know, with each era we went in order to advance medicine, we went from, you know, using our memory and saying, okay, we have we know everything, we're gonna use our information and help patients. Then we went over to evidence-based medicine, which was more data. And then finally in the early 2000s and on, we went to electronic medical records, which became even more onerous.

So, as I mentioned, the goal now is to move things in a way that that data can be analyzed so that it can allow us to spend more time with patients, which is where we should be. So, what is AI going to do for us from a standpoint? It's going to summarize charts, it's going to look at labs, look at the data, reconcile medications, make sure there's no

contraindications, et cetera. And at the end of the day, we look at this, it gives us that information, so we can then interpret it appropriately and return the physician to the patient.

So data analyzed and we're sitting and thinking instead of what you saw before where we have all this other work.

So there are four pillars that we'd like to look at when it comes to medical information coming in and how we spend our time. There's the electronic medical record or EHR. We then allow that to be analyzed and organized. It's then presented to the physician. And with that information coming in, we then can spend time explaining it to the patient.

and spending time doing that whole sit and talk situation in the in the the room instead of rushing in and rushing out. Now, what is important about this whole movement of artificial intelligence coming into any aspects of our life, but particularly in medicine,

is the theory of or the concept of human in the loop or human on the loop. And what does that mean? So we call the loop and it comes to when it comes to technology, etc., that when we move from one aspect to the other, so you have the information coming in, it's analyzed, and then there's an output

But we want a human always in that type of loop because at some point you don't want it to be where we call it autonomous. You know, we have the the our war fighters now want autonomy, electronic autonomy. even Trump was fighting fighting anthropic in February. I don't know if you guys remember that. So that was a situation where

They're allowing wanting to allow robots to find the enemy, figure out a a way to attack the enemy, and then proceed without any sort of human in the loop. That's, in my opinion, crazy talk. That's ultimately going to give us nothing but trouble. So we want a human in the loop at every aspect, so that the the AI

will then analyze all the data that we that you see here. Well blood pressures, heart rate, lactic acid, et cetera, and then deliver our a sort of a early warning signs to say, okay, this is something to be concerned about. And that's when the physician comes in and says and looks at that, looks at the patient, exams the patient, sees if the data fits the patient and if he needs to do something, and then we move forward with doing it.

What we don't want is all that same anal analytics to go on. There's alert, et cetera, and then the AI or the data says to do something without a person, i.e. a physician, saying, hey, let me evaluate the patient, let me see. There's certain nuances that you'll never see.

But where the AI will never see good, whereas a physician will always be able to understand. So I said all this now, but I'm sure people are wondering, well, you know, can you trust this thing? Can you trust the computer to analyze your patient, to help to move you forward, even if at the end of the day just to

give you better time. Well, you know, those studies have been done too. the New England Journal of Medicine just published last year how just just implementing AI scribe well with the documentation can give back an average of 30 minutes of of time per patient. And that's significant, right?

Because now you have time to do so many things, so many more things. And it showed a decreased level of burnout, which is a real thing in our profession. Now, talking about analytics, well, you know, looking at all the things that it can do, does it analyze things better than we do? Well, I mentioned earlier the whole concept really is that now we're at a situation where

the the computer can take on information from millions of individuals and then analyze just to see small nuances, small changes that can then be impacting on our single patient. So you know there's a meta-analysis which over six million p individuals came up with a scoring mechanism for sepsis.

Which was a allowed us to come up to about ninety-three percent accuracy rate. That's tremendously significant.

Continuing on with the trust thought process is, okay, well, what is the computer doing with? So we call that in data explainability. Can you explain what you're doing through the process? That actually is a very big key process. You don't want this black box scenario where I give you information.

I don't know what you do with it, but you give me an answer. That's the wrong thing because at the end of the day, just like when I'm in surgery and I go and I, before surgery, I explain to my patient what I'm actually going to do to them or the family member what I'm going to do to their loved one. That's the same concept that you want to be able to have with computers. You want to be able to explain, okay, it's going to take these numbers and compare it to this and then give us this result.

If you come up with a system where you have a black box situation, I would strongly suggest back away from it because you always want to be able to understand. You as an individual, as a physician, want to understand, but you also want to be able to at least give some sort of inkling to your families what's happening in this scenario.

So that's where you want to stay. Now the other aspect of this is we always talk about error rates, you know, because that's always concerned. How how bad is it?

or hallucinations. So at this point with what we have in many of these programs so forth we're down to like a one to three percent error rate when it comes to scribes and and when it comes to implementing things into the EHR. Well you have to have a concept of what it was before. Well I'll tell you when we write things down or type things down

before it was anywhere best scenario 7% and someplace up to 20% errors in in the documentation. So actually we've gone and moved way like a relative, the relative the reduction in in error rate was tremendous in comparison to what we had before. So we're definitely going in the right direction.

And then it's hallucinations down to you know low rates now 1.4. What we're talking saying with hallucinations is basically sometimes the computer will just literally come up with something and inject it into your into your medical record. And it could be something small, that's why you have to kind of review it, etc., make sure that it's it's that it's gone it's agreeing with what what the processes of your plans and how you

But that rate is hugely slow low. I mean, it used to be up at 7, 8, 10% before. So here you can start to see that there's some level of trust that you can start to have with your with your patients. The other thing that everybody talks about, which is important in our situation, is bias. And you know, how does bias creep in? Now we know that humans are the ones that are programming computers, so therefore.

the bias that humans have, i.e., the bias that we have in our medical system system itself is gonna be creep is gonna creep in. So we have to now build build I guess the best term would be adjuncts so that we can identify the biases and then highlight them. So you know every time we we put together health with our

So we'll see, we'll be able to attribute that. There was one recent study that showed that they were able to identify biases that were in and reduce those biases so that removed the patient population that was biased against, i.e. African Americans at the in this particular study, that we were able to reduce that by.

nineteen percent, the twenty percent range. So that was a significant thing because now you with all these individuals, over two hundred mil million people were s were were were evaluated on this population base, you can see the significance in that number

Change. So at the end of the day, can we have a partnership with AI? Of course, this is kind of gonna come. We have to come, but we have to know how to handle this, how to handle these changes. Allow the the organization of the data, the insights, reducing the risks and saving time. We do that part while the AI does that part while we

interpret, decide, and we figure out how to care for their patients. So wrapping up, will AI ever replace us? No, because we need to do the part that I just mentioned, but also we can kind of connect with grieving families. We we directly talk to our patients and about their fears and their c and so forth. We also understand the cultural nuances that

So you never have to worry that AI is gonna replace it. It's just that we're gonna be able to shift the way we interpret the way we treat, take care of our patients. It's EHR, AI, and then trust. As I mentioned before, there's gonna be a significant increase, as I mentioned in the beginning, in adoption, which is great, but maybe about

In 2022, the the studies showed only about twenty to thirty percent of of physicians were even starting to use now it's gonna moved up to eighty-one percent. Many people are now understanding that the adoption of AI can now benefit their practices and and and even

Three out of four physicians say that that AI benefits their patients. So these are kind of the things that we want you to look at doing. You know, as I mentioned, obviously read everything. those many of you have mentioned that too, but the bottom line is you don't want to blindly just allow AI to write your notes without you reviewing, etc.

Make sure that you it's explainable. The decisions that are being made are that you can deliver that information, not just for yourself but to your patients. Any type of discrepancies are are that are flagged, please take a look at it and review and analyze those. what the the blind spots of it, so where where is potential bias? Evaluate that and and make sure we highlight that.

And then always, as I mentioned, stay in the loop. We gotta have our finger on the trigger, not blindly allow the machine to do what it thinks it should do. So never also clinical judgment. AI should just augment physics physician intelligence. That's all I got.