AI in Medicine Part I

Welcome to this episode of Bioethics for the People, the most popular podcast on the planet. according to Grandma Nancy. I'm joined by my co-host, Dr. Tyler Gibbs Well if you weren't here recording right now. Probably be golfing. All right, good morning Tyler. Good morning. So it is a new year, a new top. Uh yeah, I am excited. I I love the new year. I love it when the year rolls over and people start thinking about

resolutions and start thinking about goals and stuff like that. I think it's like a really refreshing time. I don't like that it's in the middle of cold winter in the middle of Michigan, but I like New Year's. New topics. New Year, new us, new me. New me, full of water. This was one of my New Year's resolutions, which you can't see'cause of the podcast. This is a sixty four It also it looks it looks like it's m issued by the military as well. It has cute little like motivating Phrases on it.

Oh, for f at each level that you drink down it's like almost there. Good job. What is it? Rise and shine. Let's get started. Remember your goal. Keep it up. Oh man. What's the last one? The last one is All in a Day's Work. Oh that's very affirming. So affirming. Yeah. I don't think that that would motivate me. I think I would be annoyed by it mostly. Uh no I love a good motivator slogan. But mostly I just like carting this thing around and everybody going, Wow

So Tyler, I have a question for you. Sure. How do you decide? I had Oh, that's a great question. I think for me I was actually having this conversation with um one of my students actually yes just yesterday. They were starting a new elective with me. And I told them that I think probably seventy or eighty percent of what I end up writing and publishing comes from

clinical ethics consultation questions or questions that students are like, hey Dr. Gibb, have you ever thought about blah blah blah? I'm like, no I haven't. Let's get into it. So most of it is from either clinical ethics cases or students. Um asking insightful questions. Yeah. on my question. Yeah. I it's good for them because uh if they ask me the question then I try really hard to include them in the research and writing process and the publication process.

So I do publish quite often. Uh if you look at my C V I have zero publications where I'm a sole author. It's always with other folks. So usually students are residents. Um okay, so that's sixty to seventy percent. I agree, except maybe uh But mostly my students are undergraduates. Um so they do ask really good questions, but not

that before because they're just really new to the topic. Yeah. Um but a lot of times yeah it's clinical ethics or um I hear about cases in the news sometimes um like where discrimination has happened or that that happened to them in the clinical setting. So a lot of it's that. Or sometimes requests. I get a lot of requests to be to especially for like book chapters, like hey we're writing a book And so that can be rewarding too because it's like, oh

Yeah. Th there is a small subset of publications that come uh come about based purely on uh other goals so um sometimes I get asked uh what really want to go to a conference for example and sometimes I finagle my way um in that way. How about how about you? Have you ever done that? Since you mention it, yes I have. Uh every now and again you get invited somewhere really cool, like say overseas. And on the condition, and they'll pay for it, on the condition that you deliver. Whatever that was.

Sometimes it's like right up your alley. And sometimes it's a little bit less, but for some reason they wanted to invite you anyway. And that can be awesome, but it does take a little bit of extra research. So that's a nice segment.

Our new topic for this season is going to be AI in medicine, which seems to be a really hot topic right now. And so I got invited to a conference on human dignity and AI. Okay. And they said, Would you talk about medicine? Because in their minds, advances in AI that could improve people's health. That's great. I I don't know that I've ever been invited

specifically with in that way. It's usually I wanna go to the conference and I try to uh craft my proposal or the research that I'm proposing to do in order to justify going. Uh I had a really good uh presentation, uh a panel that I was on in uh at the University of Edinburgh in Scotland. that um I had to bone up on uh Adam Smith. Uh Adam Smith is usually known as a uh you know, a political philosopher in economics, but he does have this really uh kind of long book about moral philosophy. So

Uh we finagled Adam Smith into our proposal. Uh the invisible hand of Healthcare in the marketplace? Kind of something like that. It was it was a little bit of a stretch, but uh it was a it was a good trip to Scotland with some really close colleagues, good friends. So sometimes it ends up being

Okay, maybe I wouldn't have done this otherwise, but I got this cool trip. Mm-hmm. But in this case it was sort of for me, I got invited to Oxford and it was like, well, this is something that I'm going to do. Yeah. The way we know it. going on. Yeah. AI it's everywhere. And I I mean it's one of those things that it's not it's not gonna go away. I don't I don't think. And so people who are resistant to it.

Like saying like trying to ignore its presence and its influence, I think are uh it's kind of wrongheaded, I think. I think everybody who is engaged either in education, academia, research, healthcare, I think all of those areas that we kind of live in. I think AI is transformative in a lot of ways. So I'm excited about this season and and this topic and talking about AI with people who know way more about it than me. I'm really good at using Chat GPT to write haikus and limerick.

um abysmal. Alright, well okay, so I've just been doing some really basic research on the ways in which AI is already Some of those those ways. So don't ask me super technical questions about exactly how it works. In fact, um I don't know that there might be only a handful of people who really understand how AI works. There's a But so this is just kind of like my preliminary research talking about the multiple ways in which we're basically So are you ready?

It's kind of like a survey of the field. Survey. Survey of the landscape. Okay. Okay. Yeah, let's do it. So place number one. AI. Okay. Um, I think they are very well intentioned and and not I mean, I think for I I wanna be careful'cause I know how many clinicians listen to this. But um

I think for a lot of diagnoses, doctors are really, really good. Um I but I think that sometimes healthcare providers, whether it's a you know, a physician's assistant or a nurse practitioner or a physician of any specialty, I think that there's a lot of overstating, over belief, uh Maybe false confidence in diagnostic accuracy. Mm-hmm. So room for improvement I would say. Room for improvement. Sure, it could be. And I think no physician would argue.

And they can never use help with leveraging data. Right. So, of course, physicians already do this, they look to the data. So, in some ways, this isn't brand new, but the promise of AI is that it will leverage big data to improve diagnostic assessment. This we're talking about large language models. So they're basically combing all the data from medical charts and figuring out like what.

Mm-hmm. And and it's really good, you've probably heard about this, but it's already being used to read medical images. So like for like radiology, MRIs, X-rays, stuff like that. Yeah. Cancer or And there's some stuff saying that it's it's better at um seeing things earlier, which of course for cancer care is super important. The faster you can catch cancer, the usually the good. ECG abnormality, so if somebody has a cardiovascular

So again, earlier detection of those things. Some fertility specialists are using AI to choose embryos with the highest chances of achieving successful pregnancy. And and and this is all using massive data, like you said, right? So it's it's using basically the experience of all healthcare providers over like whatever records we have to improve the The precise diagnosis of this patient is sitting in front of us. Yeah. So yeah. So it's just gathering large amounts of data.

About what people are agreeing to, but for now, let's just say the the outcomes of that seem to be positive. So this could be a really Yeah. I I mean it's gonna be really hard to argue against better diagnosis. Right. I mean, nobody's gonna say, Oh, I I I wish I weren't diagnosed sooner or I wish I weren't diagnosed with more accuracy. Yeah, although and we'll get to this in just a second, people don't love the idea of their physicians relying.

To help diagnose or make treatment plans. So there's people still are pretty wary of this. So let's get to that in just a second. But let's say: okay, so these are the kinds of diagnostics. Right now AI's not replacing physicians to do diagnoses, but as usually Mm-hmm. And some have suggested this would be really good for pivotal points in a physician's schedule. tend to work for a really long time and there are times in which they miss No physicians listening to this have everyone.

But so maybe it's an even more effective or more useful tool later in a shift, for example. If you've been sitting in the basement of a radiology office and reading scans for eight or ten hours. Yeah. I can see that. So there's this idea that like, okay, when you're faltering, when your human mind, your simple, simple human mind gets tired, unlike a machine, this might be a good point to say, Okay, I'm gonna run

Model to see if it's giving me the same thing that I'm suspecting. Yeah. So like a kind of decision fatigue aid. And maybe that's like a graph. Care. Mm-hmm. Yeah. It I it that seems less problematic, I think, for a lot of people. I think if my doctor was at the end of his Or her or their shift and they were using a an AI m tool to help improve their accuracy. I'd be o I'd I think I'd be okay with that. So here's what I found. There is some promising results from reading diagnostics.

sort of providing additional information. It tends to be that when physicians are incorporating these diet Yeah, some some confirmation bias. Yeah, so we'll see. So it seems like physicians are not totally on board with using this tool.

Uh-huh. Okay. Okay. So AI can also be used for treatment planning. Again, the same kind of idea is that okay, once you have a diagnosis, you need to figure out the best kind of treatment plan for And again, by combing huge amounts of data to see what has worked for other people that seem to have a similar profile as the patient in front of you, you have

Mm-hmm. Mm-hmm. So like more tailored to this individual. Exactly. I can see this being useful particularly in really complicated treatment plans like multiphase cancer stuff, for example. I would think man if it improved the success. for that. But Americans on the whole believe that the use of So what do you how do you feel? Interesting. Well, let's try to understand that. So the concern is that it will worsen the doctor patient relationship in

Huh. In I guess I'm I'm trying to s I'm struggling to try to think of which ways it would decrease that relationship. May maybe less trust or maybe the doctor would feel less

obligated to really build that relationship and and get into the the entire life story or medical history of the patient if they feel like they can just plug it into an AI and get the same outcome maybe? Yeah I think that's a so this was a From 2023, it said 60% of patients are uncomfortable with the idea of their healthcare provider relying on AI, and that they

But I think that's a good suspicion, is you know, the way that my doctor can diagnose and treat me is by knowing something about me. So understanding my history and that kind of exchange forms a relationship. Right. So that that kind of bonding relationship might be severed if in this you you can imagine a scenario. And that might feel nothing.

Yeah, and that kind of questions the the role of the physician anyway. So I I don't view my physician as somebody I just go to for treatment plans or for diagnoses, right? They provide advice, they provide encouragement, they Um sometimes it's uh they help me think through things sometimes about my own health. And so I there's a lot more than just the diagnosis treatment.

So maybe that's what it that's getting at. That's right. Yeah. We don't go to robots. Like I mean a lot of us go to the internet first before we go to our physician. Right. Me never. I would never try to self-diagnose. But we usually don't Um even if we thought which To talk to somebody, some professional who's gonna form a relationship with us and be there with us through. Um so yeah, I can see why I don't think that's a ridiculous fear amongst patients to think that this could Yeah.

If AI is better or more accurate in diagnosis and treatment planning. Okay, so that's topic number one, maybe the least controversial. Um like there's still ethical issues here, but for the most part, I think Yeah. Here's another one. And these might get progressively Okay.

Have you heard of this? Um no, not specifically in terms of I think I understand kind of what it means, but why don't you just explain it to me? Yeah. Um so a lot of healthcare systems are now using AI to Or their mortality risk. Okay. So so basically the tool will prompt the doctor to talk about different things that may be coming or different issues that that might develop.

Right. So for example, I don't know about your experience, but my experience is that physicians are not great at predicting mortality. Like physicians, at least in the literature,

overestimate, but that it's really hard and and physicians will say it's really hard to prognosticate mortality. So a AI algorithm could conceivably and is already in some healthcare systems combing records to comb records to show a patient profile where it says a a patient who meets you know your patient's profile is likely to die in the next twelve months.

Yeah. I mean that that seems super helpful. Um I particularly if you have uh maybe uh not comorbidities, but a a number of different health issues that in isolation are hard to connect the dots about.

how that impacts overall mortality or development of other types of illnesses. I can see that being super helpful. Yeah, in in reporting on this, the f it tends to be the physicians who already are having Find this kind of nudging helpful because they might some of them said, Hey, there were patients who I thought I would not have flagged myself.

Yeah. That can be super I I can see that being really helpful, p particularly when we're talking about advanced care planning or maybe getting someone's ducks in a row towards the end of their life, you know, maybe repairing re revisiting relationships or financial questions. I I can see how that could be very useful. Yeah, all sorts of I think you and I would agree, all sorts of really important work can happen.

month of life. And if you don't know that you're dying, you don't start doing that work, right? So lots of good work can happen, but you gotta know that you're dying, which is why physicians should tell That they could die within a year. You know, and no one's ever perfect. You always meet patients who are like, oh, they told me I only had a year left to live.

So you know they weren't very good at that. So you have to kind of couch it in we don't know for sure, but um but yeah but my guess is that the physicians who already are good at could use this, the ones who never want to talk about that are the ones who could use this kind of nudging the most. Will they adopt it? Hard to say. Mm-hmm. Yeah. But at least it's you know, maybe the prompt will get them thinking about A slightly more controversial use in

your profile meets the criteria for somebody who's really likely Okay. So just imagine being told by your physician, hey Tyler, because of your lifestyle and because of your health choices, our age Here's some preventive Mm. Yeah, that feels different. I yeah. Um but not just HIV or or other type of infectious disease. Um I would imagine as well. I guess well I guess th that's that's my question. What what other things besides HIV does it prognosticate risk?

Death, HIV. There's gotta be more The two I found that are more about like disease um are were cancers and HIV. So those were the two that seem to be like the one You can imagine for all sorts of, you know, COVID or bird flu or anything coming down the pike, it could be really helpful for. But there's, especially with um contractive. You can imagine that. Yeah. Yeah. So the tool will be used to nudge the doctor to begin or have these or initiate these conversations that they otherwise for

uh a variety of reasons, good and bad, would be reluctant to start that conversation. Is that is that what you mean? Good. So these are already difficult conversations to have, both end-of-life conversations, you're at high risk for cancer, you're at high risk for HIV. These are, I mean, nobody wants

Nobody's gleeful to have a conversation like that. These are tough conversations. So this is just a kind of nudge to say maybe you should start this conversation. I don't know how it nudges. In my mind, it's like, doc, you're a Patient X. Yeah. I bet you can get it to do that. I I ch so on on my uh account for Chat GPT I changed the voice so it's much more uh It's a kind of a more maternal directive voice than it is. Uh kind of more than Siri, for example. It's a completely different uh

Voice. So yeah, maybe if you uh have those nudges come in uh a very sweet, encouraging, empathetic voice, that maybe that would improve it as well. Yeah. So there Ethical challenges that might arise in this kind of prompting. So maybe less so on the end-of-life conversation, but in the

Of course, like what does it mean that a cer a patient profile is prompting a kind of conversation about a stigmatized disease like HIV? That that's tough. And then are we Because you know, you meet this kind of demographic Also, there's conversation around like what if it's one thing if there's like mitigation strategies that you can employ that are really successful, right? So you know, there definitely are

You know, there's really nothing you can do to prevent this, and there's almost nothing you can do to treat it. Um like pain I'm thinking like pancreatic cancer is really hard to treat. So if I told you you're high I mean if there's nothing that I can do about my life style or my choices or obviously my Pisposition to cancel.

I mean there's nothing you can do. Then it's just merely added worry or added concern or added uh stress to my life that I can't do anything about. Yeah. I mean maybe you'd go in for like cancer screenings all the time. Not everybody dies. It's a the mortality rate's really high. So maybe if you were constantly There are some things in here. of work we can do and there's only so much treatment we can do.

Like it would make you so anxious. So there would be some conversation around that. Like what what's really actually helpful to know about risk? Yeah. What's the benefit of having that more more information isn't always beneficial. Yeah, and then can our healthcare system take people? cancers that they're at high risk for uh maybe a different question about the system but yeah agreed okay so maybe there's like s a little bit more Here's one that maybe you're not going to be able to do Predicting

Hmm. Okay. So so trying to predict what di clinical decisions the patient or their proxy assumed the a surrogate uh are actually going to make. Mm-hmm. Yeah, especially in the case that the patient And now we're trying to guess what they would have wanted. So either we meaning the surrogate or we mean the

Right. Which is very difficult. Even for people who know the patient very, very well, the r the accuracy of decisions made by a surrogate or by a proxy, even a very close family member, are not Like you said, even if we know so ideally you've had the conversation about this exact scenario with

with a caretaker and they're just then sort of enacting that. Less ideal is something like, oh, we've never had the conversation, but I know the kind of person that they are and I can make Even less ideal is like they never told me I have no idea what I Right. Mm-hmm. Yeah. And very common, particularly in sud end of life situations facing an unexpected death. The the number of people who have had that conversation in that level of detail that is clinically useful is very low and very

And even worse, uh sometimes we have patients who are unrepresented. There's nobody to speak on their behalf. And at least in Texas then we can have a m a physician As long as that physician isn't treating them. But what does that person know about this? Mm. Yeah. Yeah, all kinds of problems with with that form of decision making. Um but also not a lot of other options, a lot of other frameworks and how to make decisions for somebody that you don't have any very, very little information.

Yeah. So how could you how do you imagine you could use AI for this? I mean I guess you would put in uh so if you could predict what I guess I don't know, is it trying to establish what a reasonable patient would choose in in that particular scenario based upon massive amounts of other decisions in similar situations? Maybe. So yeah. So one kind of way of Like what do patients typically choose who meet this patient's profile? And that could and then you have to decide.

Like is it important? Is their age important? Is their race important? Is their gender important? Is there I mean all that feels a little bit icky. Yeah. And of all those things that you just mentioned, I think Personally, and this is probably just my delusion, but I don't think that those are the primary deciding characteristics of a lot of decisions that I make.

I think, particularly like in in a healthcare setting, but also uh how how individuals make medical decisions is kind of this also kind of black box super interesting. But I I don't that that feels I I don't like that one as much. You'll feel way better about this then. What if what if instead I comb your social media? Nope. And all the things you've ever put on the internet and your Compile that into Decision maker, a Tyler decision maker.

Uh I don't know that accuracy is the the the main thing that I care about right there. Um yeah that feels worse because then it's a bigger question about all of AI, not just in healthcare but what type of data is getting put into into the soup, right? And I don't know that any of those things that you mentioned would pr in my mind provide more inf insight into what type of person or decisions or values

Particularly not my search my browser history. I won't ask what's in your browser history, but del it's a lot about golf swings golf swing. Yeah, so there's like So one is like, would that be accurate? So it's just a kind of data question, an empirical question. Would your browser history and your social media be the kind of data that we could pull that would?

You know, it seems like that those are really different kinds of um selves. You know, like my internet self maybe does not well represent my real. That's one kind of question, which is actually potentially an empirical question, right? So we could really test this. We could do all that work, we could have the output, and then I could just ask you, independent of that, what would you choose?

Yeah. So m whether or not that's super accurate, uh there's there's claims in which we think th some people think it could be in ways that are really subtle and you don't want to acknowledge. Like I the idea that my browser history just feels kind of like icky even if it's true. But then it's is that what we ought to be doing? Like is there is that the kind of data that even if it is accurately predicted That's a different kind of

This data that isn't that we should be using. And then there's the question of like what is the purpose? If you do have a surrogate, is the purpose of a surrogate to make perfect decisions exactly as you would have made them? Which is a way we talk about surrogate decision making in healthcare, right? Substituted decision making, where we we do sort of

Yeah. I mean that raises the question about what type of data is being put it so if it's being put into the soup. So if it's social media activity, if it's browser history, that stuff is can all be useful, I guess. But a lot of my decision making is based upon, like you said, the relationship or the impact that that decision has on other relationships. Whether it's uh relationships with colleagues or relationships with my wife or with my kids or

you know, extended family or other people that I care about. Yeah, and all of that is like this really murky, moving, almost like intuitive gestalt about how I make decisions. Which I I I can't imagine that all being captured in even the best of an AI way. There's more research on this being done.

Okay. Moving on. The next domain is something about administrative efficiency. Uh so counting coppers. Exactly. The average primary People typically meet with their primary care physician. I know they're scheduled for usually 15 minutes, so seven and a half. I don't know. Well, actually, so uh that might be true in the hospital, but if you're just

They will budget about 30 minutes. So you're the the average patient will meet with their primary care physician for their you know annual checkup or whatever for about 30 minutes. The physician who does So they're spending more time on their computers inputting information than they are in the patient interaction.

Yeah. Yes. Um like that's not what most physicians will say this is not why I went to healthcare. So that could sit in front of a computer after hours every day and input information. Yeah. Um and it it leads to burnout Mm-hmm. Yeah. Uh job satisfaction is one of the primary motivators of uh

Burnout particularly in healthcare. I mean this is why we see a lot of the people who I interview who are applying to medical school have been scribes. And so this is this is like AI being the scribe. Is that how it works? You're getting into so we I just identified the problem, right? Like there's way too much time being spent manually by

So sometimes that's offloaded to scribes, but maybe there's an AI scribe that we could use. There are a bunch of companies right now that are using AI or training AI to listen in on the physician page. So that 30 minutes, it is transcribing the conversation, summarizing it, and inputting the relevant data into the electronic health record all out.

So it's almost like a like a court reporter sitting there and and uh not only just like capturing exactly what is said, but then digesting it and trying to identify what is most relevant. Uh so that like filtering process A little bit uncomfortable, I would say. Well, yeah, I mean you'd have to really trust it. So you'd have to work with it quite a bit. But the idea is it would highlight all the pertinent information, put it directly It would queue up prescriptions and

So it would say in the interaction, you mentioned this drug or this medication. Do you want me to order it? And here's how and then it would also set up the insurance claim. Yeah. Interesting. So uh the insurance part is is fascinating to me. So it would go through and basically pull out um like billing codes from the conversations and saying, Can you also bill for this? Can you also bill for that?

Mm-hmm. Yep, yeah. So it would go through those I C D codes, it would do billing. And there was a New York Times article a few months back about a physician who was really despondent about getting rejected. Wanted to enact with his patients. And so he had AI write the letter to the insurance company, and he went from about 20% approval. Oh my goodness. I it I wonder if that's just a matter of using the right key words or the key identifiers or framing it in the right way in order to

meet the whatever reviewing but who's to say that it's not AI on the other side uh on the insurance company side evaluating all of these things, right? It is. Yeah. So it that's happening as well. Physicians are using this to write the letters and then to read the letters. So I was talking to one AI programmer who's doing some of the scribing programming and he said, That's fine.

Like that doesn't bother me because he and he this I'd never thought about it in this way before. He said when I think about like what I'm trying to program, I think what is toil for No, sake of toil and what is actually important. And we should offload all the toil. Like we should not be having physicians do stuff like fighting with And if AI can do that, you can see that. Which I thought was...

Yeah, I can see that. And I I know a I mean a lot of my colleagues who are in, you know, clinical practice spend a lot of time arguing and petitioning and working with insurance companies to get stuff covered that they think is important. Yeah, so if the AIs wanna fight it out, let the AIs fight it out. And if it helps improve um, you know, acceptance of the plan, then great, all the better. Yeah. The hope is that AIs

clinicians can focus on their patient care. Of course, the history of medicine tells me that if you had an extra four hours in your day, you would love to spend that on patient care, like direct patient care. And Yeah, yeah. So maybe the efficiency will be gobbled up in just working more and more and more. So Yeah, I I can s I can see pros and cons of both of those, but

I mean it definitely takes out the the human element of it, but how much different is it really than electronic medical records or using scribes or using kind of boilerplate documents that get reused? I think th in some ways uh these just seem like Yeah, that's exactly what they are. So you're spending so much time inputting information.

Oh my goodness. That's so much time, right? So you do want to regain that efficiency. And the electronic health record itself was meant to be a system that gained efficiency, but it Um so we're actually spending more time documenting now with the electronic health record than we did.

Down in the chart. So yeah. You know, maybe the electronic medical record was just a like a proto tool, uh, an early tool that that tried to be more efficient, but maybe AI is the next step to actually increase real efficiency.

Yeah. So it turns out that Some c like the bigger companies that are owned by the mega tech companies that are doing this work are just like getting all the information about like some of the programmers can actually see the interactions between the physicians and patients, which feels like a huge Yeah. Yeah, particularly if you're talking about like I mean not to name names, but like Google Health for example.

Tons and tons and tons of data. Um it would be shocking to me if that type of data was not being used to at least train or develop some of these AIs for specific. Um and then questions about if if I'm very concerned about my privacy, can I opt out of that? Or is that just kind of part and parcel of being seen by that provider as

Right, right. So you say, Okay, I'm gonna use this AI scribe or maybe you don't say that, right? So that's a good question too, is like do you let people know that you're do you let your patients know you're using You know, what if they find out and you didn't tell them and they feel like their privacy's been violated? Um can the companies really keep this information safe? Um can we imagine that there's you know somebody could get access to people's sensors?

with folks that that makes people very nervous, right? Yeah. Yep. For I mean for very good reasons historically, right? I mean it's not like not like there's never been a scandal or a problem about health information being used against somebody in very

Okay, here's the last one. Okay. And then I think in our next episode we can talk about all the ethical Yeah. The last is triage and allocation. Okay. This is a a a topic close to my heart. I I love talking about a good resource allocation uh schema. Yeah, so remember COVID? I do. Remember how hard it was? to both like triage and figure out like who got what essential resources. Mm-hmm. Um, when we knew so little about COVID

Yep. And not just COVID as a disease and a disease process, but also the all of the medications and vaccines and treatments that were being tried, you know, who had access to those, like Remdesivir, for example, was a medication that thought to be a game changer, didn't actually end up being that that much of a game changer, but there was a lot of focus on how do we allocate these novel experimental n therapies and medications uh ethically.

Appropriately. Totally. Or and you know, like even vents that like ventilators that we assumed would really help people. Mm-hmm. Yeah. So the idea would be to use AI to predict which of the patients would benefit most from these um particular resources or medications or devices that are in short. Yep. I can see some issues with that. Oh yeah? Yeah.

Uh particularly like uh who gets to choose who is more uh deserving of this particular intervention. I mean this is something that we've been going back to we've been talking about in in kind of big world bioethics. Since what the nineteen fifties, once ventilators were invented, and organ transplantation allocation stuff, all of these.

I don't know that AI is gonna fix that because we don't really I don't think we even really know what questions are the most relevant questions to ask in these allocations. And I know that you've written about this a lot. Yeah, I mean the only thing that AI could do is perhaps by combing big data better predict mortality rates based on the data. Yeah, it's not gonna tell us about all the important things.

Mm-hmm. Yeah. And that presumes like we can even agree on what the values are or what questions to ask or how do we evaluate, particularly in in contexts of patients who have uh disabilities. I mean p part of the big controversy when we were talking about early in COVID about the allocation, how do we allocate medications and access to ICUs and stuff, the some of the early Frameworks that were proposed had a very kind of anti-disability, ableist um bias towards them.

was kind of baked into it. Yeah, it wasn't subtle. It was like patients with disabilities will not be considered right. Sometimes it was like patients with an IQ lower than, which is like wow. Or sometimes it was slightly more coded. Patients with a loss of reserve of energy. Like what in the world does that mean? Yeah. Yeah.

to still struggle with all of that kind of stuff because that would be the inputs, right? Those would be part of the inputs. Yeah. But apparently AI is already being used with like what's called patient flow management. So predicting like what kinds of patients need to be moved to the floor or an inpatient rehabilitation kind of situation or can go back to the nursing home. So it's like trying to do discharge planning and moving people up and down based on severity by predicting

Yeah. I mean I can see that a little bit in like the the flow uh the the process, the patient flow like through a clinical s you know, setting. That makes sense. I think that there's a lot of inefficient get built into not not just like how somebody goes from point A to point B within the clinic, but also like even the design of The architecture and the way the offices are set up and all that stuff.

And uh and triage management in the emergency room. So this is already kind of a thing that people have to try to figure out quickly, like who is so if you come in with a gunshot wound Mm-hmm. Yeah, but but also if somebody is absolutely screaming their their their face off in the in the emergency department and they get treated very quickly and they're no longer disruptive, that can aid in the the care of other people as well, right? It decreases the disruption.

to stress on other patients. And so yeah, some of those like interpersonal types of uh nuances I think might be Well definitely will be lost if AI is the one that's controlling all of that. Because AIs don't get annoyed. They don't get you know they don't have uh a a rough morning with their kids trying to get in and out of the car, right? And and it doesn't impact them for the rest of the day.

Exactly. Yeah. So for better or worse, right? So y we always think of that as like the benefit of AI is that it doesn't get annoyed. But maybe uh because it isn't annoyed, it doesn't read the room very well. So it actually might be a That same way. It doesn't understand social interactions or it can't account for those in the same ways. Yeah. Yeah. But but you should definitely if you have. Yeah, Tyler. Yes. I've I've heard that that's the best strategy. So

Yeah. Th this feels like just the tip of the iceberg though. Uh like I mean, these these issues uh with AI. in healthcare. I think any any time this this tool is being contemplated and being used, there's a lot of ethical questions I think that that are involved in that. Um but also I think AI is just going to be permeated

throughout so much of our society and so much of the systems that we in are involved with. Um so I I think there's a lot to talk about. There is. In the next episode we'll talk about kind of the biggest We've sort of touched on some of Yep. Yeah, so this is gonna be an ongoing conversation and we're gonna bring in some guests who are very well versed in this and kind of down in the trenches about it to help us

walk through it, help us understand it, kind of deconstruct it a little bit. But um yeah, this is gonna be an exciting series of episodes. And yeah, keep listening for more exciting breaking AI controversies. Awesome. Thanks for tuning in to this episode of Bioethics for the People. We can't do it all. So a huge shout out to Christopher Wright for coming up. For designing our own. If you're into what we're doing, Or wherever you live. Yeah. and snag some merch.