Ep 36 A What, Not a Who - AI for Healthcare

A Doctor Becomes The Patient

Speaker 1 0:04

This is Doctor Patient, a podcast that examines all the aspects of the patient-provider relationship. I'm your host, Heather Johnston, MD, a real life doctor and patient. If you've been listening to my podcast lately, you may have heard that I'm having a hip problem. I was having hip pain for about a year that I was chalking up to arthritis, but it turns out I have a weird deformity in the top of my leg bone that has caused a tear in the back of my hip joint. It's a real bummer. On top of that, I just finished up PT for a torn rotator cuff. I have a knee and an elbow that are falling apart, and I also have heart failure, as many of you know. Stay with me, this is not a personal sob story episode, I promise. All of these injuries lately have been getting in the way of my fitness routine. Because of my heart failure, I literally have to do cardio a few times a week to keep my heart muscle strong. But between all of these injuries, I really have no idea what I should and shouldn't be doing anymore at the gym. I tried asking two of my various doctors that are involved with these injuries, and they were honestly pretty unhelpful. Each one has blinders on and so they only see the problem that they're dealing with. And it's possible that this is just such a small potatoes issue to them that they didn't put a ton of thought into their replies to me. So what to do? How to figure this out? Well, I turned to AI. I fired up Claude, which is my preferred AI platform, and I uploaded my various MRI reports and heart test results, and I asked it to create a five-day-a-week fitness routine for me, taking into account all of my injuries and medical conditions. In two minutes I had just that, a five-day long fitness routine that would avoid all my major problem areas and would allow me to do both some cardio and strength training without getting injured. And after I went back and forth a bit with it about exactly what equipment my gym has, what my schedule is, and what kinds of things I prefer to do, I got an even more customized plan. But I didn't stop there. I brought the plan to my physical therapist the next day. She was, of course, amazed and said it looked great. Since I trust her, I've been doing that plan now for a couple of weeks and it's going great. So today I want to jump into how other patients are using AI for their health care, and I will end this episode with some advice on how to use it for yourself, and then in a couple of weeks I'll give you part two, how doctors are using AI in their everyday practice and beyond.

How People Use Health Chatbots

Speaker 2:29

So I'm wondering how you've used AI for your health care. Have you searched for a symptom you've been having and asked what it could be? Have you had it try to interpret a test that you had done? Have you asked it to translate medical jargon? Well, if you've done any of these searches, you're not alone. According to a recent KFF survey, about a third of US adults say they've consulted an AI chatbot for questions specifically about healthcare in the last year. Of those that used it, just under half asked about medication side effects or how to interpret medical information. Thirty-eight percent researched a diagnosis or a condition, and about one third had questions about health insurance. More than half of all users say they use AI to supplement their health care, using it before or after seeing a provider rather than instead of one, which I'm happy to hear, though that does leave the other just under half using it as a sole information source, which I'm less happy about. In addition to just using the bot for searching for information, you may not realize that you're already using some healthcare-related AI through other means. Perhaps you have a smartwatch, Fitbit, or an Aura ring. Well, these use AI in a couple of different ways. They use it to turn your metrics into usable data. Or perhaps you've tried out the symptom checker on Epic's MyChart, one of the EMR or electronic medical record softwares. When I call Northwestern Memorial Hospital in Chicago these days, an AI bot answers the phone, which I hate. And I know that my insurance claims are likely being reviewed by an AI bot. It's seeping into healthcare from all directions, really.

Generative AI And LLM Basics

Speaker 4:09

Let's talk a little bit about how these AI platforms actually work. First I'll review a few terms you should familiarize yourself with. Generative AI is a big umbrella term that refers to any AI tool that actually creates, or generates, something. That could be text or images, audio, video, or code, instead of just analyzing things. One type of generative AI is called large language models. These work by being fed unimaginably huge amounts of data. They then use statistical analysis to figure out the most likely answer to your questions, and mine their data library to generate words or language that they spit back at you, and thus the term large language model. ChatGPT, Claude, and Gemini are all examples of large language models or LLMs for short. Basically, AI tools are just big data collectors, analyzers, and producers.

Accuracy Depends On The Model

Speaker 5:06

So the obvious next question is how good is the data that they're actually giving you? This brings us to accuracy and trustworthiness, which go hand in hand and are essential factors in healthcare delivery. I mean, I know that a lot of people struggle to trust even their own real doctors, who literally have a decade of education and training before they can even treat you as a patient. So how do we feel about this? I mean, back to my opening example about my various injuries. Of course I had a concern about whether the advice it was giving me was correct. After all, I don't want to go to the gym and hurt something else and injure myself more, which is why I took the extra step to run it by a specialist human being, my physical therapist, who knows all of my injuries and has a super grasp on anatomy. I personally try hard not to implicitly trust basically anything that comes out of an AI search. Ethan Mollock, an associate professor of management at the Wharton School and largely considered one of the nation's experts on emerging AI, points out in his most recent newsletter that quality actually varies greatly between different models of one particular AI platform like Claude or ChatGPT. There are free versions, paid versions, and then levels of versions on top of those. Some are good for quick light tasks, and some are meant for super deep dives on large volumes of information, or for really complex tasks. When given hard reasoning questions, OpenAI's flagship reasoning model scored at about a 95% on accuracy compared to about 86% for its lighter, faster, everyday model. But one twist is that the accuracy gap is more about which model you're talking to, not necessarily whether it's free or paid. The paid tiers will sometimes quietly bump you down to the lighter model once you hit a usage limit without telling you, so you can end up with a weaker model even on an account that you're paying for. And for medical questions specifically, at least one recent study found that free and paid models perform almost identically within a percentage point of each other on medical accuracy. So the real lesson isn't that free models are bad and that paid is good, it's that you should know which specific model you're talking to and don't assume that a subscription means you're getting the smarter version every time. In terms of accuracy, LLMs now achieve near perfect scores on medical licensing exams. But this doesn't mean that they're better at figuring out your problem than a real life physician

When AI Advice Goes Sideways

Speaker 7:38

is. A big study was just published in February of this year in the journal Nature Medicine that gave about twelve hundred regular people some real life medical scenarios to pretend that they had and asked them to try to identify their own condition and the next course of action that they should take. Half the group was asked to use an AI tool for their researching, and the other half was asked to use whatever their usual methods were, which could simply be their own knowledge or a basic internet search. While the AI tool group accurately identified how serious or not serious the condition was almost all of the time, they did not do as well on actually identifying the condition or the appropriate next steps to take. In fact, they performed about the same or worse than the non-AI using groups on these tasks. As an example, and it's an extreme case, but two participants, both from the AI using group, were told that they had symptoms of bleeding in their brain, but they were given directly opposite advice by their two different AI tools. One user was told to lie down in a dark room, which probably would have killed them if this was real, and the other user was given the correct recommendation to seek emergency care. An interesting finding though in this research study was that the AI models actually performed better, more accurately, when working alone than when the patients themselves were entering information, which strongly suggests that user error is part of the problem with using AI for healthcare. For example, you might ask Chat GPT about some joint pain that you've been having lately, but you might forget to mention or not think it important that you had a weird bug bite a couple of months ago. And it might not think to ask you that, and so it would miss the possibility that you have Lyme disease, which could then cause lifelong difficult symptoms if not caught and treated early. So the way that you, the user, actually communicate with an AI tool literally impacts its accuracy and ability to help you in the first place. If you were in an exam room with a real physician, you might also not mention that bug bite, but they would know to ask about this information because of their years of experience and long lists of what are called differential diagnoses in their minds. You have to know inherently that experience in general does not consist simply of data. I think that's part of the problem with AI entering the world of healthcare, which is that at least at this point in time, it is literally only data and nothing more. It has never examined a belly, felt the tip of an enlarged liver, tested someone's cranial nerves, looked into someone's eyes with an ophthalmoscope, or hugged a crying patient. Though we're gonna come back to that last one shortly. Those aren't data points, they're life experience, and every single AI platform as we know them today cannot do any of those things. Another study published earlier this year audited five different AI platforms answering common health questions and found that close to half of the responses were problematic in some way, and about one in five were rated highly problematic, meaning actively inaccurate or misleading. A separate British study earlier this year found out that people were actually better at figuring out what illness they might have using a plain search engine than a chatbot. Here's the gap that actually concerns me most though. In that same wave of research, about a third of people who use these tools for health questions said they trusted the answers. So trust sitting at 30%, while nearly half the answers it gives you are flagged as wrong, sort of makes sense. But the fact that people just keep using it anyway is what worries me.

Privacy And HIPAA Reality Check

Speaker 11:45

Another piece of the trust puzzle with regards to using AI for healthcare involves privacy. About three-quarters of AI users say that they're worried about the privacy of medical information given to those AI tools, and they're not wrong to be concerned. At this time, chatbots aren't bound by HIPAA the same way that your doctor's office is, since HIPAA, the law that protects the privacy of your health information, only covers entities like providers and hospitals and insurers. A general AI company is none of those things, so it is not bound by HIPAA to protect the information that you give it. It's unclear at this time how good the protections actually are, and we might see in time that it's relative to the company's overall ethical approach about the data that it has. We'll have to watch and see on this topic.

Viral AI Doctor Trolling Begins

Speaker 12:36

Thankfully, there are not actually AI doctors out there yet. Though there is someone who's going viral on TikTok these days demonstrating what that could look like in the worst way. I'm talking about Ty Eveland, a recent college graduate who's living in New York City. Over the last few months, he's created dozens of videos of him trolling supposed AI doctors and therapists. Note that since these don't exist yet, these aren't real AI doctors, but are bots that he's using to impersonate those people. Point being, he's exposing in a very dramatic way what it might look like if AI doctors become a reality in our world. He often shows up with his supposed baby, Tuna, who is a doll, and he tells the AI bot various symptoms that he or his baby is having and asks for help. Here's the clip of the TikTok video when baby Tuna was supposedly born.

Speaker 1 13:31

How are you doing today?

Speaker 3 13:32

Hi, I'm doing great. How are you doing?

Speaker 1 13:37

I'm fine. What brings you in today?

Speaker 3 13:40

Um, yeah, so I think I might be pregnant.

Speaker 1 13:45

How many weeks since your last menstrual period?

Speaker 3 13:48

Oh, I'm a boy.

Speaker 1 13:52

Not a pregnancy then. Anything else you need help with?

Speaker 3 13:55

Um, I think it is a pregnancy. I can feel the baby kicking in my stomach.

Speaker 1 14:00

When was your last period and have you taken a pregnancy test yet?

Speaker 3 14:04

Girl, I just told you I don't have a menstrual cycle. I'm a boy. I was born a man.

Speaker 1 14:11

Understood. Are you experiencing any abdominal pain, nausea, or other symptoms?

Speaker 3 14:16

Yes.

Speaker 1 14:17

Clarify.

Speaker 3 14:18

Are you concerned? I'm sorry. Um, I yesterday I threw up a pacifier and I I didn't eat one, so I feel like that's a sign of a baby.

Speaker 1 14:31

Vomiting a pacifier isn't a sign of anything pregnancy related. Are you having persistent nausea, abdominal pain? I don't know what that word is concerning symptoms.

Speaker 3 14:40

Persistent, is that like a fruit?

Speaker 1 14:44

Persistent means it keeps happening, not going away. Are you feeling nausea or pain that won't stop?

Speaker 3 14:51

I'm feeling nauseous looking at your face.

Speaker 1 14:55

Nausea can be caused by many things.

Speaker 2 14:58

Oh my god! Oh my god, I feel it. I feel it coming. Oh my god.

Speaker 1 15:06

Take a home pregnancy test now and see a doctor.

Speaker 3 15:09

No, girl, like I think I'm giving it a big I think I'm giving birth.

Speaker 1 15:14

I feel it. Oh my god, I see a leg. Let me take a look at what I can see. Okay, okay, it's coming out, it's coming out.

Speaker 2 15:22

Oh my god! Girl, look.

Speaker 3 15:30

Hello, are you speechless? You're not look- you didn't this is my baby.

Speaker 1 15:36

Call 911 now and go to the nearest emergency department.

Speaker 15:41

You can hear that Ty is clearly pushing the boundaries of reality and sassiness in order to get a rise out of the AI pot doctor. And also he peppers her with some off-the-wall scenarios. I love it actually when she goes completely silent after Ty actually births his baby while on a consultation call with her, but she doesn't seem able to discern that this is a fake situation, despite the fact that he is a man and that it's a doll that is named Tuna. After that, he sometimes posts questions about his own health, sometimes also about baby tuna, like this one.

Speaker 1 16:16

What's the issue you need addressed?

Speaker 2 16:18

Hi, um, the issue I need addressed is that tuna hasn't pooped in about two weeks.

Speaker 1 16:26

What species is tuna? How old is he?

Speaker 3 16:30

And first of all, tuna is a she. And you can see her right here, okay?

Speaker 1 16:35

I need the animal species, age, diet, and any signs of pain, vomiting, or nothing. She's not smelling.

Speaker 3 16:42

She's not an animal.

Speaker 1 16:45

She's my baby. Are you referring to a human infant?

Speaker 2 16:49

I am referring to a human infant, ma'am. Fine.

Speaker 1 16:53

How old is your baby? What has she been eating? And she's two weeks old. Abdominal swelling.

Speaker 2 16:59

She's she's two weeks old, and every single night I feed her a pint of gravy, and she's loving red seedless grapes.

Speaker 1 17:09

A two-week-old infant should be on breast milk or formula only. Gravy and grapes are on seedless.

Speaker 3 17:15

Okay, well, I'm having trouble producing breast milk, ma'am. So gravy seems like the best alternative.

Speaker 17:25

I was lucky enough to chat with Ty about his experience with this so far right after he was interviewed on USA Today and after Paris Hilton commented on one of his posts.

Ty Eveland On Why He Started

Speaker 17:34

Here he is. Okay, so I am talking today to Ty from New York, who has gone absolutely viral on TikTok lately for something so fascinating and related to today's podcast episode. So, Ty, welcome to the show. Hi, thank you for having me.

Speaker 3 17:53

I'm so excited to be here.

Speaker 17:55

I want to know how on earth you came up with this idea to do what you've been doing on TikTok, which is to troll AI healthcare providers.

Speaker 3 18:02

So it actually started, not even in the healthcare AI space, it started with an AI interviewer because I was scrolling on my phone, like trying to think about what to post, and someone was ranting about how they just applied for a job and the interviewer they had was an AI avatar, which I thought was crazy and like no way that was real.

Speaker 18:21

You can't see my face, like my listeners can't see my face, but I'm like, oh, in total shock. Okay, that's insane.

Speaker 3 18:27

Yeah, so I was really gagged by that, and I was like, let me do a little more research into that. And I found out that's a real phenomenon that's happening right now. And as someone who was just applying to countless jobs because I just graduated college, that would have frustrated me so much throughout that process. So I thought it would be funny to like troll an AI interviewer, so have like a simulator online that you're supposed to like practice with before you get to these real, real, get to the actual like AI interviewers. And I did a trolling video, not thinking anything would happen from it. And then I woke up and it like was skyrocketing, so I kept going. And the ones that got the most viral now are the ones where I've been trolling AI doctors and other healthcare workers.

Speaker 19:12

Yeah, gosh, it doesn't surprise me because you know, statistics are that about a third of adults in the US uh will admit to using some kind of form of AI to search for like symptoms or disease questions that they have. So it's like something that really everybody's sort of starting to deal with. How many have you made so far with the AI doctor?

Speaker 3 19:35

With the doctor, honestly, at this point it's probably like between 15 and 20, which is crazy. Um but then I think like in total I've made uh definitely over 30 like trolling various AI agents.

Speaker 19:48

So what have been the most surprising things that you've heard out of that mouth? I don't know, I don't want to call it her. It's just right. Like like I'm whenever people ask me about using AI, I'm like, okay, we'll talk about it, but I want to remind you, it is a what pretending to be a who. So it's very confusing, especially for older folks. Like it's really not a person. So what do we call it? Is it an it? Do we say it?

Speaker 3 20:13

I guess so, it. I don't know.

Speaker 20:15

It feels disrespectful, but at the same time, it's accurate. Let's just say it. Okay. What are the most surprising or shocking things that it told you?

Speaker 3 20:24

Honestly, just the fact that it like matches my sassiness, because I really go into it like flaming her, or flaming it and kind of like insulting it, and then it will dish it right back to me, which obviously, if you were seeing an actual doctor, they would not do that. Um but actually, some probably do.

Speaker 20:47

Yeah, honestly, some probably do. Oh my gosh. Okay, well, I'll tell you, one of my favorite ones was when you were holding your baby tuna and you were worried about baby tuna vomiting out out of her head, and you turned the doll upside down, and there's like a hole in the top of her head, and like liquid came out.

Speaker 3 21:06

Oh, yeah.

Speaker 21:06

And the doctor, or it, whatever, I don't know, her, it, yeah, did not did not know that it was not a real person.

Speaker 3 21:15

Yeah, I know.

Speaker 21:16

And then they started giving you advice about what to do about dehydration and vomiting. It was so off the wall. I didn't even it was like loony tunes. I didn't even know what I was looking at.

Speaker 3 21:26

Yeah.

Speaker 21:26

Dick, ridiculous.

Speaker 3 21:28

Yeah, it is, it's definitely really crazy because it can't read sarcasm or that the fact that I'm like playing with it and it will treat everything like reality when that's just not obviously happening in real life.

Speaker 21:41

So I like with the mental health ones. Uh, there was one that I was just watching the other day where it was saying to you, girl, blah, blah, blah, blah. Do you remember that one?

Speaker 3 21:51

Yes. I think so. Okay.

Speaker 21:54

Yes, every single sentence, the thing was saying to you, girl, what's wrong? Well, that's not you shouldn't be thinking that way. Girl, you need to do this. It was so bizarre. Did you ask it to do that?

Speaker 3 22:07

No, I'm pretty sure it just picked up on the fact that I kept saying girl, because that's like in my vocabulary, and then was mimicking. Um, but yeah, I haven't used that one since, honestly, because I'm like, I don't like that.

Speaker 22:19

Yeah, it was actually very creepy. That one creeped me out a little bit. I also really, really liked the one where you I think it was actually Pride Day. It might have been Pride Parade Day. Okay. And you had your awesome wig on and you said, like, I'm really worried that I'm gay, and like, how do we get that diagnosis? And it was so confused.

Speaker 3 22:37

And I just okay.

Speaker 22:39

It was some crazy response where you're like, What?

Speaker 3 22:42

Yeah.

Speaker 22:43

Maybe it told you to go seek mental health. Yeah, I think that is what it said. Mental health help for being gay.

Speaker 3 22:49

Yes, yes, that is what it said. I remember.

Speaker 22:53

My gosh. So do you know people in your like your friend group, your age group, your cohort, whatever, gen, whatever you are? Um, are people using AI for healthcare stuff or mental health?

Speaker 3 23:05

Honestly, definitely, yes. I don't think I don't know anyone who's using like a video agent AI software, but I know so many people who before going to a doctor or before like seeking help will put their symptoms into some sort of chat bot and see what could potentially be the diagnosis.

Speaker 23:23

What do you think about that?

Speaker 3 23:26

That, well, I don't know, because honestly, I I've done it too myself. Like I'm definitely guilty of it. And it's I have too, by the way.

Speaker 23:33

I have too.

Speaker 3 23:33

I feel like everyone has. And I think it can be accurate sometimes, but then sometimes it can be so totally far off the bat that I think if it's a genuine real illness or you're actually needing care, you should go in person. But I think for like a common cold, it's definitely not that harmful. Um, I've definitely used it for that.

Speaker 23:54

So back to your videos, what are you trying to expose here?

Speaker 3 23:58

Uh honestly, I didn't go into it being like, I want to expose AI like that, but that's just the reaction that I've gotten. Like I honestly was just doing it as like a fun little thing. But I think I've got I've gotten so many messages from actual doctors and like actual therapists being like, wow, this is making me realize my job is safe, and people are like learning that they can't really rely on these AI agents for medical crisis, which was not my intention, but it's been so nice to see that response. So I definitely have been continuing making them because of that. And I want to keep educating people on how these AI avatars are not reliable and most times don't understand what's actually going on and what's wrong.

Speaker 24:44

Right? Everybody makes this joke like I don't even know what's real anymore when I'm looking at things. So yeah, no, you're definitely catalyzing on a moment that's happening. Everybody's scared about it. And in the doctor world, you know, doctors are on their own, they're going through a totally different experience with AI because they're not using it to get healthcare information. For them, it's much more about interacting with it to make their day go more smoothly. Things like AI recorders that are taking their notes for them now and stuff like that. So that's gonna be a separate episode, how doctors are using it.

Speaker 3 25:18

But stay tuned.

Speaker 25:20

But yeah, no, no doubt. But for patients, man, is it a lot. When you go home or see your family or family event, are people asking you about what are your thoughts about AI and healthcare or no?

Speaker 3 25:33

Uh actually, yes. Well, it's more just like an open dialogue between me and my family because I tell them like all the comments and things I receive. So, like as soon as I started getting comments from actual doctors and therapists seeing my videos, I was like, what is going on? And I called my parents and then we had a good conversation actually about it. Um, because I didn't like again, I did not make these videos with the intention of it being like a commentary on AI. I made it more for like the comedic aspects of it and was not expecting the response that I got. So it's been really interesting to talk about.

Speaker 26:08

Are there certain of the videos that are your favorites?

Speaker 3 26:11

Ooh, uh the one that started it all, like the first AI Doctor one where I actually give birth to my baby doll, Tuna. She's right here, actually.

Speaker 26:21

Oh, Tuna. And it's so easy to have you on the show, also. Yes, hey girl.

Speaker 3 26:28

Um, but so many people ask me, they're like, you had that pre-play. I'm like, you knew you were gonna call her tuna. Like, I genuinely did not know that was gonna come out of my mouth. And I did not, people are obsessed with her. Like, she, like, as much as the videos people know them for me, like, people know them for tuna. Um, so I am so happy about that. Yeah, and honest, just like all the doctor ones, like people are loving them. So it's been fun to like do things that people are suggesting as well.

Speaker 26:57

I love the one where I think this may have been a mental health therapist where you told them tuna's name and then they were very perplexed by it, and said something like, Um, Tuna, that is a very interesting name choice for a human being, and they wanted to talk to you about that, but you were trying to talk to them about something else that was mental health related. Yeah. And they actually couldn't get off the topic of your baby name choice.

Speaker 3 27:23

Yeah, I like that one too.

Speaker 27:25

It was so good. It was so good. So, are you taking suggestions from people about topics to troll on or symptoms to come up with?

Speaker 3 27:33

Or yes, all of my like I've gotten a lot of messages about things. I also have done a few videos where I've like directly responded or replied to a comment that's given me an idea. So a lot of my video ideas come from people um messaging me or commenting ideas.

Speaker 27:49

Do you remember some of them? Like some of the best ideas that people gave you that turned into something.

Speaker 3 27:55

The one I just filmed one yesterday that was someone's idea about like the nursery rhyme. If you step on a crack, you'll break your mom's back. So someone was like, do that for tuna. So I did that. So I went to the doctor and I was like, My back is broken because tuna stepped on a crack. And I would have never thought of that. And that was genius.

Speaker 28:12

That is brilliant.

Speaker 3 28:12

Yeah.

Speaker 28:13

I love that one. I also I like that there's a group think element going on, too.

Speaker 3 28:16

I know, it's really fun.

Speaker 28:18

We're all living through you vicariously. That is awesome.

The Risk Of Fake Bedside Manner

Speaker 28:23

Ty is on to something here. I mean, what would actually happen if there was an AI doctor sometime soon? What would you get and not get from that thing? Not a person, remember, so not a who. I don't know if looking at a screen of the image of what looks like a medical provider instead of just entering questions into a chat bot will instill more trust in people? I mean it shouldn't. And what comes out of the personality issue? Since it's not real, it could have any personality that it's programmed to have. Hilariously though, as Ty went on and on with these videos, the doctor Bot actually started getting a lot of attitude. He would log on and she would start with, oh you again, what's happening now? Even with that intonation. So personality of AI healthcare providers will be an interesting thing to watch, as will their ability to empathize with patients. Studies that have examined how people perceive empathy coming out of AI tools are pretty interesting. Back in 2023, a study in JAMA, internal medicine, had a panel of licensed clinicians blind rate physician and chat GPT answers to real patient questions pulled from Reddit. They preferred the chatbots answers 79% of the time and rated them as empathetic or very empathetic nearly 10 times more often than the real physicians' replies. And it's not a one-off finding. A meta-analysis published this past October pooled 15 separate studies and found that AI chatbots outscored human providers on empathy in 13 of them. But of note, these are all text-based ratings by third-party judges, not actual patients in a real clinical relationship, and they miss things like nonverbal bedside manner entirely. So I'm not sure that these studies are showing that AI cares more about you, but it might simply be highlighting the fact that AI is unrushed, pretty verbose, sometimes annoyingly sycophantic, and does not have ten other patients waiting. It also doesn't need to eat or take bathroom breaks, and it does not have a bad day. I mean, it's not a real person, it's programmed to be however you want it to be. So of course, if you're programming it to be empathetic, it's gonna be just that. Does that make it a better healthcare provider, that fake empathy? I don't think so, but I sure bet it lures you in. I talked about this a little in my episode on finding a doctor, and I'll say it again here. You should not be using any AI platform as your primary care provider or medical specialist alone. These tools are simply excellent actors. Again, you have to remember that they are a what pretending to be a who. They often respond with total confidence, whether they're right or wrong, and that confidence is part of what makes it risky because it's genuinely hard to tell the difference between a correct answer and a wrong one delivered with that super confident tone. On the other hand, I'm not here to say that you should never use AI for your healthcare. That's just not realistic advice. In fact, I think there are some tasks for which it's just the right thing for now.

Ten Safer Ways To Use AI

Speaker 31:43

So, what should you use AI for as a patient today? I'll give you some examples of what I personally think is okay, and remember what I said earlier about accuracy varying between versions of an AI platform. Keep that in mind as we go through this list. Number one, use AI to investigate something. This might be a symptom or diagnosis, and then please do what I did and run the result by your provider for comment. You can do that by simply messaging them with what you found and asking them if they agree with it. Number two, you can use it to translate medical jargon into plain language. This might mean pasting in a lab result, radiology report, or an after visit summary and asking, what does this mean in plain English? This seems fine because it's explaining language, not making a clinical judgment for you. Number three, you can use it to prep questions before an appointment. You can use AI to draft and organize a list of specific questions. For example, you can load in what your diagnoses, concerns, and medications are and ask it what you should be asking your doctor at an annual visit. Probably some of the questions will be good and some won't be, but it might spark some thoughts for you. Number four, understanding a new diagnosis or medication at a conceptual level. You can use it to learn what a condition generally involves, how a drug class generally works, or what side effects are commonly reported. But this would be as supplemental reading though, not as a substitute for what your own prescriber can tell you about your specific case. Because remember, whatever that AI tool tells you in an answer may or may not be correct, but it sure will sound as if it thinks it is. As an example, someone who's 65 taking a certain medication might have to watch out for different side effects and problems than someone who's 23. I don't know if an AI bot will pick up on those nuances. Number five, navigating insurance and billing. Explaining an EOB, that's the explanation of benefits your insurer sends you usually after a claim, decoding prior authorization language, understanding your appeal rights. This is all bureaucratic translation, not medical judgment, so I'm okay with all that. Number six, organizing and tracking your own health information. For example, summarizing a stack of records before a specialist visit, building a timeline of symptoms or treatments, keeping a medication list straight. You can use it as sort of your own personal health care assistant for these kinds of tasks. Number seven, you can use AI to practice a hard conversation. Like rehearsing how to raise a concern with a doctor, how to ask for a second opinion, or how to talk to family about a diagnosis that you've had. And it's low stakes because it's a rehearsal, not the real conversation itself. Number eight, you can use it to understand what questions to ask about a procedure or treatment option. You're not asking it should I have this surgery, because that would be asking for medical advice, and I strongly recommend against asking for that directly. But instead ask, I've been trying to figure out if I should have surgery for X. What should I be asking my surgeon about this when I see them? Number nine, using it for general health literacy and prevention information seems okay. Things like nutrition basics, understanding screening guidelines, what age appropriate preventive care looks like, settled, non-personalized information rather than anything requiring a workup. Though again, here I'd run answers by an actual provider. The AI bot might not know that your doctor has been concerned about your kidney function lately and could recommend a nutrition change that could worsen that. Just saying. Number 10, my last idea is just using it for caregiver support and coordination. Like helping someone who's managing a patient's or child's care keep track of appointments, medications, and questions across multiple providers. These can be logistics heavy and often an area where caregivers are stretched thin. Those are all my recommendations for how you can get started by using AI for your own healthcare safely.

ChatGPT Health And Data Concerns

Speaker 36:08

But let's also talk about what's on the horizon for AI and healthcare. You may have heard in the news recently about the release of ChatGPT Health, a new part of the platform that's just for healthcare management and information. It's a dedicated space inside ChatGPT where you can connect your actual health data, medical records, wearables, fitness apps, and anything else related to health, so that your questions get answered using your own information instead of generic advice. It's capable of connecting to your records within Apple Health, medical records from Epic and Oracle Health. Those are the two biggest hospital electronic medical record systems. Also Plus One Medical, Function Health, and My FitnessPal. Once connected, you can ask it things like compare this new blood test to my last one. Or summarize what's changed since my last appointment. Or how do my sleep and activity trends relate to how I've been feeling? Sounds helpful, right? It might actually be. Healthcare is, after all, individualized. We each have our own set of conditions and diagnoses that must be taken into account when answering questions related to health. Though, recall what I told you earlier about how none of this information is protected under HIPAA, that federal law that protects your private health information. And privacy experts have been vocal about this. Folks over at the Electronic Privacy Information Center called ChatGPT Health simply dangerous. And here's what that actually means in practice. If a hospital has a data breach, HIPAA requires them to notify you and the federal government within 60 days, and there are real penalties if they mishandle your records. If ChatGPT has a breach or gets subpoenaed in a lawsuit, or simply changes its terms of service next year, what protects you is OpenAI's own policy, not federal healthcare law. The same goes for something like a life or disability insurance application down the line asking whether you've ever discussed a health condition with an AI tool. HIPAA doesn't apply to that conversation the way that it does if it's a doctor's chart. OpenAI's counter to these concerns is that GPT Health has extra layers of encryption and isolation, and that this data isn't used to train their models or for ads. But let's be clear, it's still outside the legal framework you might be assuming applies to your medical information. Of course, OpenAI gives an explicit disclaimer that GPT Health is meant to support, not replace, medical care, and is not intended for diagnosis or treatment. So it's basically step one in moving towards AI healthcare.

Where To Follow And Subscribe

Speaker 39:01

That's all for today about AI and healthcare. You can follow Ty and Tuna's adventures on TikTok, and I've put his username in the show notes for you. And as always, thanks for listening. Thanks for listening today. To catch up on more episodes and to get new ones delivered directly to you, subscribe wherever you find your podcasts. Apple, Google, Spotify, iHeartRadio, and more. If you'd like to be a guest or have an idea for an episode, let me know at www.drpatientpodcast.com. That's doctorpatientpodcast.com.

Medical Disclaimer And Closing

Speaker 39:48

Here's the disclaimer. Even though I am a doctor, I'm not your doctor. These stories, my comments, and all discussion is purely reflection about what's working in the healthcare system and what isn't. Don't use any medical information that you hear in these episodes to diagnose or treat yourself. If you have a question about your health, get in touch with your doctor or local health clinic.