Episode Transcript
[00:00:00] Sam, There's a moment sometime in the next year when you'll walk out of a doctor's office and you won't know what just happened.
[00:00:39] You'll know what the doctor said, you'll know what the diagnosis was, and you'll know what the bill said.
[00:00:47] But you won't know during that 15 minutes in the exam room if an AI system was listening to you and if it caught every word that you spoke.
[00:00:59] Maybe it transcribed it, maybe it summarized it and quietly slipped a structured note into your chart.
[00:01:07] You won't know that another model, maybe from a different company, a different vendor, potentially read that note, flagged a pattern in it that maybe your physician didn't see.
[00:01:20] You won't know that maybe two weeks later, when your insurance company processed the follow up that your doctor recommended, a third AI evaluated whether that follow up was medically necessary or not. And maybe it decided in less than a second that it wasn't.
[00:01:44] So I'm Dr. Alan Badot and this is AI Today with Nell Media. And thank you for being here tonight. We're going to look at what happens when you walk into an exam room and we're going to go into this together.
[00:02:01] We're going to find out who's actually in there with you and we're going to think about maybe, maybe you're not alone, right?
[00:02:12] So if you have read the highlights, the headlines and the highlights recently, you'd think that AI in healthcare is either a miracle or it is a giant scandal. The truth is, you know, it's, it's quieter than, you know, either of those actually. And it's, the reality is it's already here and it's been here for a while.
[00:02:36] Let me start with really the quietest part of that and you know, the part that you would really never notice if you didn't tell somebody about it or I wasn't telling you about it.
[00:02:53] It's called the Ambient AI Scribe.
[00:02:57] And you know, really it's an algorithm that just sits in a room and it's usually some software that's running on a doctor's laptop or on their phone and it listens to the entire conversation and it's, you know, between the physician and the patient. And when the visit ends, it produces a clinical note and that goes into your chart and it can go into your ehr, you know, the doctor's EHR system they have, or they're still using paper could go in there, you know, as well. And, but in any case, it's, you know, after that visit ends and it captures all that stuff. It may make a diagnosis, it may do a plan of care, it may do some medication changes, maybe some follow up instructions, but all of it is formatted, it's filed and saves the doctor a lot of time. And that they don't have to, they don't have to do any sort of typing, they don't have to do any of that stuff.
[00:04:00] Now it's pretty amazing for the doctors because it clears up a lot of their time and lets them focus on more important things, right, Your care and that kind of stuff.
[00:04:11] But I read that, you know, a study published earlier in the year by JAMA, and JAMA is the journal of the American Medical association, almost forgot that it looked at five, you know, academic medical centers that are using these AI scribes. And the results, you know, if you ever sat across from a doctor who's, you know, eyes really never left the screen or, you know, it sounds kind of science fiction, you know, that was before, you know, you'd walk in the room and they'd immediately start typing, asking you questions. They may not even make eye contact with you. But that's really changed with this scribe capability. But the total time spent in electronic health records EHRs, you know, it was down by almost 15%.
[00:05:08] That's, that's pretty good. And I think that equates to about, you know, a 10 minute difference in the amount of time that they would have to spend.
[00:05:21] That's, that's amazing. Documentation alone, you know, was down as well. And you know, one of the things I think was that was the most striking of this study was that physicians using these tools were able to see roughly, you know, a half more patient during a shift and you know, maybe even another full patient during a shift, depending on what your specialty was. Of course. Now half a patient, okay, may not be revolutionary, but when you cut across an entire practice of providers or a health care system, that becomes significant very quickly. And you know, looking at that entire hospital system, you're talking about tens of thousands of additional appointments a year that they can, that they can make.
[00:06:18] And that's, you know, in a country where the average wait time to see a primary care physician is now weeks, math matters, right? Becomes important.
[00:06:34] Now it's not just speed though, right, because we got to think about quality of care at the same time. But the study tracked physician burnout and that's one of the epidemics that's really killing the US is that there's not a lot of physicians coming in to replace the ones that have burnt out that are retiring early.
[00:06:52] And that's not, that's not unusual the problem.
[00:06:58] And you know, charting really became an after work event for physicians. You know, almost.
[00:07:08] I heard, I heard some would call it pajama time because they would do it right before they would go to bed. But you know, they would have to take it home and do it when their kids are sleeping.
[00:07:18] So that's really the first revolution in that.
[00:07:22] AI is giving doctors back some of their own time.
[00:07:27] And you know, just like everything else, when a physician is happy, when they're, you know, fully rested, those kind of things, quality of care is seriously going to be improved now. The second revolution though is really happening at scale and nobody outside of the industry can really see what's really happening.
[00:07:53] It's really happening in drug discovery.
[00:07:56] It's amazing.
[00:07:58] Today you look at the different capabilities but Profluent and Eli Lilly announced a partnership that's valued at almost two and a half billion dollars. And the work is going to be AI driven gene editing.
[00:08:19] Specifically we're talking about using machine learning to design therapeutic proteins at scale.
[00:08:27] And those are proteins that, geez, just five years ago it would have required armies of biochemists and quite honestly decades of lab work to identify.
[00:08:40] That's crazy.
[00:08:43] And that's just in five years what we've been able to do.
[00:08:47] So that's a big deal, right? But you know, it's really a much larger pattern that is starting to really take over.
[00:09:00] You know, we talk about crossovers and we talk about convergence and you know, in other industries you often don't hear that in the medical field with AI, but we're starting to see that convergence.
[00:09:16] You know, Eli Lilly also has, you know, an arrangement with in Silico medicine and that one also is valued at, you know, two and a half, $2.7 billion.
[00:09:33] Tempus AI, another company looking at building some of the largest libraries of multimodal what I would call clinical data in the world. They've expanded their multi year collaboration with Merck. I mean these things are every day you're seeing them and those just keep going and going and we're developing more and more and we're rewriting some of these things that it in real time almost.
[00:10:04] So the question is though, are algorithms becoming as critical as laboratories?
[00:10:13] So that's not my quote, that's not even my question. That's a quote that I got from a biotech analyst describing really this moment.
[00:10:22] And I think it really does land precisely where we want it to.
[00:10:29] We have spent hundreds of years equipping medical researchers with better instruments, whether it's microscopes, sequencers, mass spectrometers, all those things. Now we are equipping them with something that previous instruments couldn't provide and that's pattern recognition at a planetary scale.
[00:10:54] That's pretty phenomenal.
[00:10:56] And then we look at the third revolution and then we're thinking, oh, geez, you know, that one really just arrived almost last week.
[00:11:07] And you know, on April 23rd, OpenAI released a clinician specific version of ChatGPT and it's free, it's built specifically for medical workflows.
[00:11:22] Now alongside they also release something called, you know, Health Bench Professional and that's a, you know, a standardized benchmark for measuring AI performance in real clinical conversations, diagnosis, triage, care, coordination, those sort of things.
[00:11:41] Now, if you ask yourself, and from my perspective, I was more excited about the benchmarking than the tool itself.
[00:11:52] For the first time, though, a hospital, a state medical board, a regulator, whoever that is, they finally have a way that they can compare AI tools before they touch a patient and whether that's measuring the hallucinations that still occur and will continue to occur with those types of tools in their rates to audit, guideline adherence to, you know, check before you trust type things. All of the pii, all those things phi same thing.
[00:12:27] Those are not small things.
[00:12:30] And I think, you know, in the end it's going to turn out to be one of the most consequential developments of, you know, the entire next six months because it's always, it's always something, right?
[00:12:45] So I'd say we, you know, we really, we've given doctors some of their time back. That's great. We have AI accelerating the discovery of treatments that didn't exist five years ago. Amazing.
[00:13:02] We finally started to figure out a way that we can measure AI and put it in an exam room and see if it's any good. That's great, right?
[00:13:10] Great revolution.
[00:13:13] But that's the part that the headlines are getting, right?
[00:13:18] When we come back, we're going to talk about some of the parts that the headlines are missing, the diagnosis that AI is catching that maybe doctors aren't seeing or some of the patterns that, you know, your data is giving off that maybe aren't seen by the AI, but a doctor has been able to see him. So we're going to look at all of these cases when we come back. So stay with us. We'll be back after a few messages from our sponsors.
[00:14:23] Welcome back. Before the break, I told you about the AI that's helping you Know, doctors do the work that they were already doing before, but now they're doing it significantly faster. Right.
[00:14:37] And, you know, this segment, though, it's going to be about something entirely, you know, different.
[00:14:46] We are going to talk about the AI that's doing work some of the doctors, they just cannot fundamentally do.
[00:14:55] And I want to. I want to get some folks to understand that, you know, there are some nuances. There are just some facts about human cognition that we just don't talk about enough.
[00:15:13] And so, you know, for instance, you know, a radiologist looking at chest CT scan, you know, reading, you know, and on average, I think somewhere between 300 and 500 images, you know, per slice and, you know, per scan, I mean, and, you know, the scan takes roughly six to eight minutes, right?
[00:15:44] So that means they might read 50 of them in a day. And that's, you know, maybe they get sent some from others and they're in their own practice or whatever that is. But it could even be more than that.
[00:15:57] But by the end of the day, just the math says that, you know, they have looked at roughly 15,000 individual images.
[00:16:12] That is crazy.
[00:16:16] Human attention does not scale that way. And this is not, you know, anything against the radiologists, of course, but that is, you know, why there's so much specialty training. Why, you know, it's one of the most desired practices that you can go into coming out of med school. There's, you know, it's challenging and, you know, it's very well reimbursed, but, man, that just is. It's crazy when you start to look at what's happening on the screen, you know, next to me. Let's go to the next one and we'll take a look.
[00:17:00] There are just some nuances in realities that we have to.
[00:17:05] We have to think about. And this is really a simplified visualization of how a medical, you know, AI process is for a single patient. So on the. On the left side, you see imaging data, whether that's ct, mri, ultrasound, you know, above it, you're thinking clinical notes from years and past visits and everything else that's going on. And then you've got.
[00:17:33] Over to the right, you've got your lab values, you've got your vital signs, your medication history, all of that stuff. And then below, you've got your genomic data in some cases, or family history, even, you know, social detriments, whatever that is, pathology reports, biomarker panels. Go, go, go, go, go.
[00:17:54] Then you can start to see that you have a classic case of multimodal fusion. You've got data streams that are converging into, you know, these unified embedding spaces and pattern detections and, you know, pattern detection waves that can, can activate and start. And a human doctor looks at one of these streams at a time, sometimes two, maybe more, right?
[00:18:24] But they read the imaging, they check the chart, they read the labs, they form a hypothesis, and they test it.
[00:18:35] AI doesn't do any of those sequentially, Right?
[00:18:39] It looks at all of them at once.
[00:18:43] And what it's looking for, you know, what the most advanced systems are doing right now are, you know, they're generally, generally good at finding those types of patterns.
[00:18:53] You know, is the, you know, is the pattern. It may not show up in a single stream, but it only starts to emerge as you start to put some of these streams together.
[00:19:04] Now, that pattern, though, maybe a doctor wouldn't be able to see it, maybe they can.
[00:19:12] But there's an advantage of looking at all of the data at once.
[00:19:18] Now, let me give you, I'll give you a concrete example. Okay? So the FDA, they have, since 2016, granted what they call a breakthrough device designation. And what that really means is that, you know, at least for most of them, they're AI powered. It allows them to, you know, quickly get through the FDA requirements that are, you know, really the drivers behind what it's allowed to be used for, how it can be used, how it's shifted, and what it's trying to do is get these tools into doctor's hands significantly faster.
[00:19:57] And, you know, that's, that's great. It's. It's very important that we, we continue that, of course.
[00:20:04] But in some cases, the FDA appears to be reserving its breakthrough designation for AI systems that can do things physicians simply can't. And that makes sense, right? Because, you know, the last thing we need to do is compound a provider situation that just continues to get worse and worse, as you know, over the next five to 10 years.
[00:20:35] You know, things like detecting multiple diseases and processes simultaneously from a single scan.
[00:20:43] AI can do that.
[00:20:45] Predicting some cardiovascular events from retinal photographs. Right. Spotting pancreatic cancer and imaging that was originally captured for entirely unrelated reasons.
[00:21:00] You know, going back to the radiologist piece, you know, how often are they finding something that you really weren't there for in the first place?
[00:21:12] Because they're looking through everything.
[00:21:16] And, you know, when we talk about the pancreatic cancer piece, that's, that's one I want to, I want to focus on if you've had family that if you've lost because of that, you understand how, how cruel and ruthless pancreatic cancer is.
[00:21:35] It's one of the deadliest ones that we have. And usually by the time the symptoms appear, it's too late.
[00:21:43] But pancreatic tumors, they leave some subtle radiographic signatures, and those can exist in scans that were taken months ago, sometimes years.
[00:21:55] And really that's before any of the symptoms even emerge. The scans were always there.
[00:22:01] Nobody was looking for some of those signatures, though, because nobody knew where to look.
[00:22:08] It's not. I'm.
[00:22:10] Don't confuse it. I'm not pointing fingers at anybody by any stretch of the imagination. We're just talking about the complexities of so much data that's out there. We've seen it in the Department of Defense, right? When you take multiple sensors and you put all that data together, now we're starting to see it in other areas of, you know, our daily lives.
[00:22:36] AI can know where to look. It's learned it from being shown, in some cases, the future, because by being trained on patients whose later diagnoses allowed the model to walk backward through their earlier scans and identify what was already visible, then you start to be able to see the brilliant value of using a tool like that.
[00:23:03] Now let's look. We'll take a look at the next part of our screen.
[00:23:08] It gives you an idea on.
[00:23:11] Of course, you're seeing a sterilized version of how AI attention maps can, of course, differ from, you know, what the human eye visual scan path would look like.
[00:23:25] But, you know, if you think about it, the white traces are kind of the radiologist's eyes as it's moving across an image, structured train patterns, looking for certain things, looking for markers that they've learned, you know, where, as the heat map over to the right is where AI is concentrating, they overlap, but they're not the same.
[00:23:50] There are regions the AI weights heavily that the human eye barely can register.
[00:23:57] And occasionally, not often, but consequentially, it's in those regions that that diagnosis lives.
[00:24:09] And this is what I mean by sight beyond sight. You've heard me say that before. But it, you know, the AI is not seeing through walls. It's not seeing into the future. It's not seeing, but always is present in the data. Okay?
[00:24:25] It's what, you know, a human reader might have had a hard time looking at. And now, like everything else, we are trying to put them together and elevate what their capabilities are.
[00:24:41] So there's a company called Tempus AI that I just want to mention really quick because it Kind of illustrates the scale at which this is happening. You know, Tempus said, I think they've got a approximately 38, 39, a little less than 40 million, whatever that is, research records, and over 7.
[00:25:02] I think the last number I saw was 7 billion clinical notes.
[00:25:07] That's a lot of bad handwriting. But they recently ran a pilot where, you know, their AI processed 60,000 patient records in a few days.
[00:25:19] And generally, though, if you would, had humans looking at that, we're talking about months to complete.
[00:25:25] And it's not a story, though. Don't think I'm telling a story about replacing humans. I get asked this all the time. We talk about it all the time the time. It's a story about asking questions that were impossible to ask before.
[00:25:41] Like when you can structure 38 million patient records and search them with natural language in seconds, you can identify patterns. You can look at entire generations of, you know, characteristics that you can bring to the surface.
[00:25:59] And that's in an afternoon.
[00:26:02] And if you don't think that is impressive, then, well, you're a tough one to convince.
[00:26:12] But here is what I want you to hold on to at home, okay? I want to. I want to break this down. And the medical AI that you're most likely going to encounter in the coming years is it's not going to be a chatbot that answers your questions, just won't, especially about rashes and those kind of things, because that's a different story entirely.
[00:26:35] It's going to be that system that can read your scan before the doctor did.
[00:26:39] It's going to be that one that has flagged a pattern, the one that is, in some quiet sense, the first set of eyes on your case.
[00:26:51] And in many cases, I would say in probably even the cases that matter most, those eyes are going to be something a human reader either missed or just can't see.
[00:27:04] And that is a breakthrough.
[00:27:07] That is the revolution that we're being told about. Except this time it's actually here.
[00:27:14] But we know every resolution or revolution has two faces, right?
[00:27:21] Resolution, I guess, too, because the same data, you know, the fusion of the data, the techniques that we're doing it that can create a pancreatic or that can catch a pancreatic tumor early.
[00:27:36] That same architecture, that same multimodal capability, that same ability to read your entire medical existence once and, you know, make it through that entire, you know, your entire story, well, it can do some other things, too.
[00:27:56] And the story of how an algorithm decides whether the treatment your doctor just recommended is one your insurance company is going to be willing to pay for.
[00:28:12] And that's where we're headed when we get back. Stay with us.
[00:28:38] Foreign.
[00:28:48] Welcome back. I'm your host, Dr. Alan Bideau. And welcome back to AI Today, before the break, I told you about seeing, you know, doctors and you're not sure if it's AI or what's, you know, how things are shaken out. And you know, we talked a little bit about how AI can see some things that just, you know, humans can't.
[00:29:13] You know, I, I mentioned that I wanted to really hammer on a certain aspect of it and it's really going to be about that mysterious algorithm that is deciding whether you should get a certain treatment or, or not.
[00:29:39] So earlier in the month, you know, Stat News published a special investigation into United Health's group's, you know, their AI strategy. And the numbers in that piece are, of course, depending on your vantage point, either inspiring or unsettling.
[00:29:59] And you know, UnitedHealth has hired, of course, hundreds of AI engineers by some count. Really, it's really almost, I've heard numbers as high as 22,000 people.
[00:30:14] And that's to track and build and deploy AI across its claims processing capabilities, looking at prior auths, looking at care management options, everything in between.
[00:30:29] Now, UnitedHealth is projected, and this is all public information, nearly $1 billion in AI related soft cost savings in 2026 alone.
[00:30:44] HCA Healthcare, you know, they expect roughly 400 million in similar savings and a lot of that can be driven by some of their automating, their revenue management strategies.
[00:30:58] And these are, these are, these are huge numbers and they reflect a very real efficiency that insurance companies are taking advantage of.
[00:31:12] Now we'll go to the next slide. So just a, you know, an idea I just wanted to show you so you could see what a claim submission process, you know, looks like. But I want to put the idea on the table so you can, you can think about these.
[00:31:31] And you know, that's when an insurance company says it saved a billion dollars in a year by deploying AI.
[00:31:43] What exactly did the AI do?
[00:31:48] It made decisions, we know that, faster than humans could and probably about claims, about prior authorizations, about whether a recommended procedure is medically necessary and about whether a diagnostic test was justified or not and about whether a hospital admission could be approved.
[00:32:14] And I'm, in this case, I'm not blaming anybody for any of this. Right. And I have written algorithms and I've built agentic solutions to facilitate some of these advancements, you know, but let's, let's just look at it and we'll look at it from a clean perspective and we'll figure out, you know, the good parts about it and then the bad parts about it.
[00:32:42] So oftentimes you don't see what the decisions are until you hear about them. Right? You don't see what's going on behind the screen. You see the result.
[00:32:51] The result comes in an explanation of benefits. And that letter that arrives in the mail a couple weeks later and it says, oh, you know, you're, you're either covered or you're not covered.
[00:33:03] And the result is the either a prior authorization denial that your physician's office then has to spend a few hours appealing with, or, you know, the result is the moment when your specialist tells you that imaging study you need has been denied and you have to decide whether or not you're going to fight it or you're going to pay for it out of pocket.
[00:33:31] What you don't see is the algorithm.
[00:33:34] You don't see the model reading your case.
[00:33:37] You don't see the training data that it was built on. You don't see any criteria that applied. You don't see any threshold that has determined whether that particular request crossed the line from approved to denial.
[00:33:53] And so you see the problem is you, you don't see the math.
[00:33:58] You only see the answer.
[00:34:01] Now, again, before we go any further, I just want to, I want to be clear, all right?
[00:34:10] Not every AI and healthcare claims processing is sinister.
[00:34:15] Not all bad. Some of these systems are only there to flag fraud, for example, to catch providers potentially that are doing or billing for services that they did not perform or, you know, things like that. And it can drain billions of dollars out of the health care system every single year.
[00:34:38] And some of those just exist to enforce coverage rules. And it has nothing to do with a medical judgment or denial. It's legitimate, contractual, you know, operations that it is trying to, trying to enforce.
[00:34:55] But here's what we do know, okay?
[00:34:58] We, and we know it because Blue Cross, Blue Shield has now publicly said it and Blue Cross has, you know, suggested that AI enabled medical coding practices can contribute to more than $2 billion of additional claim spending nationwide.
[00:35:22] $2 billion in additional spending, not savings, additional spending.
[00:35:29] So what the heck is going on?
[00:35:33] So if we take a look, I got another slide for you. Let's take a look real quick because this is important.
[00:35:42] That number tells you something very important, right? Because, you know, AI and healthcare finance, whatever that is, it's not unilaterally cutting costs, all right? Same way with jobs. It's not costing jobs unilaterally. And it's not cutting costs unilaterally either.
[00:35:59] It's shifting where the money goes.
[00:36:02] Some of it goes to the insurer's bottom line. Some of it goes to the providers, some of it goes to the AI companies themselves. Right.
[00:36:11] Some of it gets denied to patients who couldn't afford to do an appeal.
[00:36:18] The system on net may not be cheaper.
[00:36:23] It may simply be more opaque.
[00:36:30] And that's what I want you all to remember tonight.
[00:36:33] Opacity.
[00:36:35] I want you to remember that word.
[00:36:41] So look at, you know, what is going on with this. And you're, you're. It's pretty easy to see that, you know, as those claims are going back and forth, it's being processed by some sort of model and it's going to be against coverage for coverage, whatever that looks like.
[00:36:59] And then sometimes it's doing an awful lot of work.
[00:37:05] And, you know, there's now, though, you know, these decision trees and things that are offering, you know, some potential exits for us as we, as we continue to move forward.
[00:37:26] There's a wave in a lot of state legislatures. And, you know, I want to be fairly precise about this. And so I'm gonna, I'm gonna do a little cheating that.
[00:37:37] So there's, there's seven bills in five states and they've got a common goal. Alabama, Minnesota, Wisconsin, Michigan and Massachusetts. And the bills are different in their specifics. Right. But they're, they're in essence trying to do the same thing in that they want to mandate human review of AI assisted insurance denials.
[00:37:59] They want to have a ban from AI making that final decision, decision on coverage on its own.
[00:38:07] And so take a moment and think about what that legislation implies.
[00:38:13] It says currently AI is making the final coverage decision, doing it on its own. You know how we feel about that not going to change based on these, these kind of studies. Human in the loop. Real human in the loop, making sure that there's some sort of, you know, regulatory promise that your decision, your final decision is going to be looked at by humans. California's gone farther, right, in January, which means, yeah, it's effective right now, actually. They require that, you know, healthcare chatbots disclose their AI nature and it bans those that don't have suicide prevention protocols, which is fantastic.
[00:39:02] I think it's an early model of where we're going to go.
[00:39:05] And, you know, from my perspective, you hear about this all the time, transparency.
[00:39:10] The floor, that's a basic, has to be a basic disclosure. The floor, again, has to be a Basic.
[00:39:20] I can't, can't, can't shy away from that.
[00:39:24] But when AI denies your claim, what interest is that AI serving? We have to ask that, and we've got to take that into consideration.
[00:39:40] So the question no longer is whether AI belongs in healthcare. It's here, we know it, it's been here for years.
[00:39:48] The question is whether, sitting at home while you're watching this, you know AI today, whether you know which AI is acting on your behalf or which one is acting on someone else's behalf.
[00:40:02] So when we come back, we're going to talk about what to ask, what to look for, what rights are yours, what some of the most consequential readers of your medical record may be, and it may not be a human. So stay with us. We'll be right back.
[00:40:51] Welcome back to AI Today. I'm your host, Dr. Alan Bideau. And this week we have been talking about AI and how it's being used in the medical field to assist doctors, maybe assist you with detecting things early or some of the negative effects of assisting insurance companies and denial of coverage and those sort of things. And I want to spend this last segment, though, really trying to give you something that's practical.
[00:41:26] It's not a manifesto, okay? It's not a prediction. It's a short list of things that you can carry with you as you go into your next medical appointment, you know, the next time you have your conversation with your insurance company on why your procedure was denied.
[00:41:48] And it's really just four, four basic items, okay? It's, that's it.
[00:41:52] And the first one, though, is just a basic one. Ask whether or not AI is being used in your care.
[00:42:02] That's sounds very simple though, right? But unfortunately, it's not.
[00:42:09] Most of the time in most healthcare systems, the answer, your physician is, you know, gonna be partial. They're not. They may not know. It's not that they're hiding something.
[00:42:22] It's because the AI is often deployed at different organizational layers that your doctor may not even know, may not even see.
[00:42:32] You know, like ambient scribes, for example, probably image analysis, probably increasingly likely though, right?
[00:42:43] You know, the whole clinical decision support woven into the EHR, almost certain coverage determinations on the AI, you know, with, with AI, 100%, probably.
[00:43:00] But you know, your physician, the office is dealing with those sort of technologies with AI more and more.
[00:43:10] And it's happening more than you know, you even know.
[00:43:14] The question, though itself is the point. It's asking, you know, asking it changes the conversation.
[00:43:24] It signals that you are a Patient who wants to understand the layers of decision making, every one of them that touch your career.
[00:43:33] Nothing wrong with that.
[00:43:35] And in a system that historically has not encouraged that question, the question itself is really, it's a small act of agency if you think about it.
[00:43:53] Number two, when you receive a coverage denial, you need to ask whether the AI or whether an AI even was used in the determination.
[00:44:07] And you want to ask for the criteria. How did it make that decision?
[00:44:12] Most insurance companies are not currently required to volunteer this information, but many are required to provide it on request.
[00:44:24] The seven state bills that I talked about in the last segment with the laws in California already passed though. Right. But pushing this information towards a mandatory disclosure is happening more and more.
[00:44:37] In the meantime, though, asking activates your rights, and you've got to make sure that you're protecting those.
[00:44:50] If the denial was AI assisted and you appeal, your appeal in most of these frameworks has to receive a human review.
[00:45:01] And that's the difference between a denial that stands and a denial that could get overturned. The appeals process is where AI errors most frequently get caught.
[00:45:16] The vast majority of patients, they don't appeal.
[00:45:20] And maybe it's the single most consequential statistic in healthcare, AI today.
[00:45:29] Number three, be skeptical of AI, only mental health and chatbot services.
[00:45:39] Okay, I've done a lot of research in this and, you know, we're just not there yet.
[00:45:47] Now, California's new law that they have, it was driven in part by reports of mental health chatbots that failed catastrophically.
[00:46:01] Like, for instance, it didn't recognize a suicidal situation or even the ideation.
[00:46:07] Now other states, they're starting to catch up with this. And again, the technology itself is not uniformly safe, especially to augment a human therapist work.
[00:46:22] If any place that you want a human in the loop, that's the place.
[00:46:29] There are just even some dangerous, you know, AI technologies that are out there. And one can pretend to be a substitute for another. And the difference isn't always obvious, and especially from a marketing perspective.
[00:46:45] And now it's only further compounded with agentic and all these other things.
[00:46:52] You know, if you or someone that you love is in distress and AI is not a sufficient first responder, the chatbot may be helpful, but the clinician on the other side, you know, the other end of that crisis line, that's essential.
[00:47:11] Number four, and this is what I want you all to remember the most out of this show.
[00:47:22] The doctor who treats you is not, at least in 2026, the only intelligence that touches your care.
[00:47:34] There is A system.
[00:47:36] And it's a system of multiple algorithms deployed by multiple vendors, owned by multiple parties, and some who have your interest at heart, and some have other interests.
[00:47:49] That's just how it is.
[00:47:52] But. But you are entitled to know who is reading your file.
[00:47:58] You are entitled to know on whose behalf you are entitled to ask.
[00:48:05] And increasingly, you are entitled to a human answer when an AI algorithm has made a decision against you.
[00:48:17] So remember that at the start of the episode, we were talking about, you know, things going on or that we, we see next year. And when you walk out of a doctor's office, you won't know what exactly just happened.
[00:48:35] Now, after talking through this, I want to.
[00:48:39] I want to. I think I might want to event amend that. And now I want to say, you won't know unless you ask.
[00:48:51] That's the difference.
[00:48:55] That is where the agency lives. The technology is moving faster than our laws. We know that. We've talked about that on this show. And faster than our habits. Right?
[00:49:10] But not faster than a question.
[00:49:13] Ask clearly at the right moment.
[00:49:17] The doctor who treats you and the algorithm that judges you are not the same actor anymore. They're just not.
[00:49:26] We hope that the algorithm is going to support the doctor, and it should, if it's done the right way.
[00:49:34] But you have a right to know which is which.
[00:49:39] The doctor that treats you and the algorithm that judges you, they don't necessarily converge.
[00:49:49] We want them to. Some systems they do, and some they're doing it the right way. Others they don't. You could have an AI that is recommending a treatment, see something earlier, looking at all your medical history.
[00:50:05] Then you have another AI that's denying claim or denying a procedure.
[00:50:12] That's the world that we live in.
[00:50:16] But don't be afraid to ask, because that is the most important thing that you can do.
[00:50:24] I'm Dr. Alan Badot.
[00:50:26] This is AI Today. Thank you again for watching. We'll see you next week.
