Skip to content
PodSubs
ShowsPeopleGuestsJobsNewsEvents
More
BlogPodCred
Menu
ShowsPeopleGuestsJobsNewsEventsBlogPodCredNetworksSign in
Join nowSign in
  • Shows
  • People
  • Jobs
  • News
  • More
    • Guests
    • Events
    • Blog
    • PodCred
    • Networks
    Join nowSign in

C B m E & U

C B m E & U artwork

Can AI make better doctors? with Walter Sinnott-Armstrong

Sep 16, 202544 minAudio

Send us Fan Mail If you were a doctor, woken in the middle of the night to decide which patient should get the only available liver, would you be comfortable making that decision yourself? What if an AI device could also do it, or even do it better? In this episode, Prof. Walter Sinnott-Armstrong (Chauncey Stillman Professor of Practical Ethics at Duke University) gives us the reasons why AI can help people make the kinds of decisions they would make if they were the best version of themselves. Walter gives us examples of AI-judges making bail decisions and AI-doctors making medical decisions, argues that dinosaurs were killed by excessive volcanism than by an asteroid, and why you should be a vegan when you are within a 50km radius of Peter Singer. Remember people, be savvy!

Audio and video play straight from the show's own hosting. PodSubs stores and serves no media.

Transcript

From the show. Tap a time to play from there.

SPEAKER_01The center presents a new podcast, C BME and You, a podcast devoted to promoting awareness and debate around controversial ethical issues in biomedicine.

SPEAKER_05Welcome back to CB Me and You, a podcast about biomedical ethics. Today we go once more into the breach, the breach being the ethics of using AI and medicine. How can AI be used to help doctors make medical decisions? What are the ethical challenges? How can we overcome them? And if our guest had an AI named after him, would it be called Wally? Now you'll see what that is about in a second, because joining us today is Walter Synod Armstrong, a Chauncey Stillman, distinguished professor of practical ethics at Duke University.

SPEAKER_05Welcome, Walter.

SPEAKER_02Thank you so much for having me. I'm having a wonderful time in Singapore. Good to hear.

SPEAKER_00Excellent. So our first question to get to know you just a little bit better is coffee or coffee?

SPEAKER_02So I tried some coffee with cream and sugar. And no, I did not like that because I don't like cream and sugar in my coffee.

SPEAKER_00Condensed milk?

SPEAKER_02So condensed milk, yes, not cream. And now I have it black. So I'm gonna try it for the first time here and let you know.

SPEAKER_05We have a live tasting.

SPEAKER_02You can hear a pin drop, which is a little bit more than a little bit.

SPEAKER_05Oh, yeah. Yeah.

SPEAKER_00Excellent. So did did James tell you about how they roast the beans with caramel sugar and butter?

SPEAKER_05No. Uh that stuff is uh Sinead's expertise.

SPEAKER_00It's one of my favorite things about coffee. Excellent. Uh, second question what country do you want to travel to next?

SPEAKER_02So I've been dying to go to India. Uh a friend of mine invited me, and then sadly we couldn't go because his wife needed an operation, and and so it got canceled. But within the next year or two, I'm gonna go to India. Uh, and you might think most people go to India, they go to see the Taj Mahal or something like that. I want to go see the Deccan Traps. The Deccan Traps are a giant lava field, uh, which I could tell you why I think this, but which I believe is actually the source of the volcanic eruptions that killed the dinosaurs and this stuff about the meteor and the Yucatan.

SPEAKER_02It finished them off, but there were already two-thirds of them died before the meteor ever hit. And I'll cite the science if you want, but we're not here to talk about that, are we?

SPEAKER_05You know, I'm so tempted because I'm I'm a huge dinosaur nerd myself. Oh. I go around the office sometimes wearing a dinosaur t-shirt. Um and and that's that is a controversial take. I do know that that happens to be a really controversial take.

SPEAKER_02Yeah, I have a full-size T-Rex skull in the front of my house.

SPEAKER_05I'm extremely jealous. Maybe one day when I become a distinguished professor of ethics, I'll have a you'll be able to.

SPEAKER_00I'll tell you about my collection later. How does a full-size T-Rex skull fit in the front room of your house? Is it this is the house?

SPEAKER_02It's a big house.

SPEAKER_05Nobody's gonna fit a full-size T-Rex house in the HDB.

SPEAKER_00Yeah, yeah, but still aren't T-Rexes huge. Like, isn't the T-Rex skull like the size of this table?

SPEAKER_02No, the the skull is like maybe four, four and a half feet long.

SPEAKER_00And it's a ri it's a real one.

SPEAKER_02No, it's a it's a cast directly off of Stan, the most complete male T-Rex ever found. But he's not really male. They don't know if it's male or female, but it was discovered by a guy named Stan.

SPEAKER_03Okay.

SPEAKER_05So Stan, the paleontologist, is like, ah, must be a dude. I found it.

SPEAKER_02Oh no. Sue is the other one, but it was discovered by Sue. So a guy named Sue, and now they always refer to that T-Rex as a female, but they don't know.

SPEAKER_05And Sue is the most complete uh T-Rex, if I'm not wrong.

SPEAKER_00Okay, I'm really sorry to do this to both of you to move on from dinosaurs. James literally go all day. James, hear me out. Last question. Uh, what is your go-to karaoke series?

SPEAKER_02And there's absolutely no question about that. We do karaoke a lot at my house. I actually have a uh dedicated karaoke room in my house.

SPEAKER_05And he has a T-Rex call and a dedicated karaoke.

SPEAKER_02This is like an American house. Uh and uh whenever we do karaoke, we start with Bohemian Rhapsody and we end with We Are the Champions.

SPEAKER_00Can we can you invite us over one time? Sure. I think James. Come on over and zoom. I want to see this karaoke room.

SPEAKER_05We just have to go halfway across the world.

SPEAKER_00Who knew that adult houses? And now I say that, that came out really bad. Grown-up homes could be so fun. Excellent.

SPEAKER_05So there are plenty of ways in which AI, at least people are talking about plenty of ways in which AI can be used to help make medical decisions. Let's zoom in on one, just as a starting example. So, one of the ways that you've talked about um is the use of AI to help allocate organs. Um, and this is something that you talk about in your book, Moral AI and how we get there. There's the plug. Right? So, do you want to take us through how exactly we can uh AI can help doctors make decisions in in these sorts of cases where doctors have to make very quick decisions about how to allocate organs to um uh to different people who need it?

SPEAKER_02Sure. I think we have to start by recognizing I know a surprising fact that doctors are humans. And they make mistakes like all humans do. And if you look at the decisions they make in allocating organs, they're just mistakes all over the place. Like the the allocation of organs is different on the weekends than it is during the week. They don't take organs from you know black women because they think, oh, that's not gonna work.

SPEAKER_02And it's amazing how many organs get wasted and how the organs get distributed in ways that are not fair. Uh and so our thought is that we can study how they make these decisions and get rid of the mistakes by correcting in various ways. So we set up situations where you have patient A and patient B, and you have, say, eight characteristics of patient A.

SPEAKER_02They both need a kidney and there's only one available. So they each have eight characteristics. You know, say patient A is 31 years old and patient B is 45 years old, and patient A has um has one child, but patient B has three children that depend on them, you know, or maybe an elderly parent and you know, criminal record, drinking history, uh, other kind of health, you know, they keep themselves healthy by eating healthy or exercise. And you put those different features down, uh, all kind of relevant, and you don't say whether they're male or female or Asian or Western or what.

SPEAKER_02Uh, and so you're leaving out the things you think people should not be making this decision on. Uh and then you ask them to do 50, 100 of these. In some cases, we had them do 600 of these, uh, and you study the patterns and how much weight they put on things. Uh, and then you can build an AI system using machine learning that will predict what they'll say about another case that they haven't seen yet, and we're getting it right 92% of the time.

SPEAKER_02But that's just the start. Because of course, that's just normal human judgments. Well, how are we gonna get rid of the mistakes that I mentioned earlier? Well, we can study information. How does information change the way you distribute kidneys? So now we're looking, for example, at uh information about diabetes. You say, suppose A has diabetes and and B doesn't, who are you gonna give it to? And a lot of people say, well, I'm gonna give it to B because the diabetes might mean there's gonna be a later health problem and the kidney will be wasted.

SPEAKER_02And then you give them a little information. More African Americans have diabetes than European Americans, and so if you institute that as a policy, then um African Americans will have less access to kidneys. How does that affect the way you distribute the kidneys? And so now we can we can study how information, becoming more informed about the situation, affects the way you distribute. We can also look, uh, for example, at uh biases of various sorts uh by in by putting things in that are correlated with race and seeing how it affects your judgments.

SPEAKER_02We can actually even give profiles that fit actual cases of blacks and whites and see whether the blacks are going to have less uh opportunity to get a kidney transplant when they need one. And the third we correct for is stability. We're finding that in certain types of cases where the Bradley Terry scores are equal, but that doesn't matter. That's a technical detail, uh, that uh 20% of the time people give a different answer to the same scenario. So you got to correct for all that. You got to get rid of those errors, and then the doctors can use the AI system to give them some guidance.

SPEAKER_02You can say, if you were better informed, more impartial, and more stable, this is the answer you would give.

SPEAKER_05Yeah, great. I just want to go back to the statistic you gave. You we we have this AI system, we could train it in a way that uh I think you said 90 that gets it right 92% of the time.

SPEAKER_02Yeah.

SPEAKER_05Yeah. How are we measuring that? Like who's deciding um whether an AI system has got it right in one case? Is there a panel of expert uh ethicists who are deciding, all right, this was the right decision in this case?

SPEAKER_02No, we're not doing that at all because we don't want to impose our own values in this way. We're simply predicting what people would say. And they might say the wrong thing before you've corrected, right? Uh so you're simply predicting. So you tray if they do 100 scenarios, you train the machine learning algorithm on 70 of them, and then you predict the remaining 30. And you randomly sample 70, 30, you know, different subsets, and then you get an average accuracy. That's where we get the 92%.

SPEAKER_02But all you're doing is predicting what they would say. You're not telling them whether it's right or wrong.

SPEAKER_05So what's the difference then between getting an AI to make those decisions and just a person to uh who can just get predict what they will say 100% of the time?

SPEAKER_02Presumably Because the person is gonna see the patient and know whether they're black or white or male or female. Yeah. Because the person is gonna be tired because they haven't had enough copy. Uh and because the person, you know, is going to uh get confused or forget about certain information that's relevant. And the AI doesn't do those things. It doesn't get tired, it doesn't get confused, it doesn't, you know, it doesn't forget information. It can check everything in the blink of an eye.

SPEAKER_05So what you're saying is that the 100% uh in this case, the person who gets it right 100% of the time would be the same person if they were in some kind of ideal situation, if they weren't tired, if they weren't affected by information that uh they that they would think is irrelevant if they uh you know if they weren't drunk.

SPEAKER_02Not a hundred percent. Yeah. You're never gonna get it right a hundred percent of the time. But I would rather trust the judgments that I make when I'm informed and impartial and stable than the judgments I make in the other cases. Think about language. Everyone makes grammatical errors when they speak. But then you point them out and they go, Oh yeah, you're right. I wasn't supposed to put it that way. Well, you know the rules of grammar, but you don't always perform properly.

SPEAKER_02And what we want to capture is the rules what you would follow, you know, what you would do, the judgments you would reach, the decisions you would make if you were in these more ideal circumstances. Not always gonna be right. It's gonna be right more of the time. We're just trying to reduce the number of errors. Nothing's perfect.

SPEAKER_00One of the questions I wanted to ask you is about uh the difference between bias presenting in humans versus bias presenting in AI. There's been a whole lot of uh hoo-ha in the literature about bias in AI, and everyone wants to talk about it. And anytime you raise the idea of maybe using an AI tool to solve one decision or uh in a different uh decision-making process, there's always someone that goes, but what about the bias? And so, given that AI is trained on human decisions or human data that we've generated ourselves over however many research and clinical trials, is is the bias in AI uh different conceptually or morally to human uh human bias, or is it just simply making human bias apparent?

SPEAKER_00And why do we seem to care so much more about the about AI bias than the fact that our doctors are already doing these decisions?

SPEAKER_02So there are a lot of questions there. I'll start with the last one. Why do we care more? I think it's because AI is new and we're scared. Uh, and because AI is young and so it makes a lot more mistakes than it will five years from now. And and so you should be scared and careful and correct for those biases. But yes, it does have advantage. Remember, in our system, you're basing it on human judgment, sure, but you're basing it on human judgments of people that did not know whether patient A or B was male or female or black or white or Asian or Hispanic, whatever.

SPEAKER_02And so those biases that we run into in everyday life uh are not gonna be present in the data because you didn't know. There are there are um problems still with that, and we can talk about those later if you want. But humans are gonna be much more subject uh to those biases. Uh the other advantage of AI is you can fix it. If you measure how much bias there is and you analyze it to figure out where the bias is coming from, you can fix it.

SPEAKER_02Try fixing a doctor who's confident that this is who ought to get the kidney. Try fixing a judge who thinks I know what the law is and this person's a criminal. It's really hard to fix humans. We've been trying for millennia, and it is difficult. But AI, you can actually correct pretty quickly. A good example is uh self-driving cars. You know, a human has an accident, and a few people hear about it, but that's the end of it.

SPEAKER_02AI has an accident, self-driving car. The new data goes out to every car in the fleet. And they and they're not gonna do that again. So AI can learn very, very quickly. At this point, it's young, at this point it makes lots of mistakes, not always being used properly, but there's a potential to be a lot better uh than humans because it's easier to just retrain it. It never complains when you tell it it was wrong.

SPEAKER_00I saw that quote once, or I saw a quote once saying something about that uh AI doesn't have an agenda or it doesn't have a pride problem that when you point out an issue in it, it's like, no, I didn't do that. Yeah. Or that's not wrong. Uh and ever since I learned that, I realized that actually, yeah, buy AI is never gonna sit there and be like, oh no, no. I'm I'm completely justified in making this biased decision or I didn't say that.

SPEAKER_05That's not what I said. I guess one thing I am wondering about is that um there's this idea that we can make an AI that is less prone to certain mistakes that uh human beings make. But what about the about AIs that we know that do replicate a whole bunch of mistakes that human beings make, right? Uh we know based on um our experiences with Chat GPT and a whole bunch of other chatbots that they um make a bunch of mistakes that we make.

SPEAKER_05Uh uh, they have biases, biases that we have, they make errors in reasoning that humans commonly make, uh fallacies that humans commonly make. They make that because they're trained on what we say and what we do. So is there any way of uh getting rid of those mistakes?

SPEAKER_02Um You cannot get rid of them all. Yeah. But you should not let the perfect be the enemy of the better. Uh so we can get rid of some of them by studying the patterns. If you look at the um so uh Kleinfeld did a one very interesting study, 150,000 bail decisions, and you can see how much preference there was among the judges for uh whites over blacks getting getting bailed, getting released before the trial.

SPEAKER_02Uh and then you can take the same data that the judges were looking at and train an AI and see how much bias it has. Now it's still gonna have some bias, but it's gonna have less. And you're gonna be able to analyze how did it look at that data, what was producing those biases, and then correct for it if you want to. You might not want to correct for all bias because don't forget that you might have to sacrifice accuracy to get rid of bias.

SPEAKER_02After all, it's very easy to get rid of all bias. Let everybody out. Or let nobody out. Or flip coins all day. Yeah, or flip coins, or you know, but you don't want to do that because then you're gonna have criminals out on the streets hurting people. And so you have to worry about the potential victims, too. And so with an AI, you can actually measure the accuracy, you can measure the bias, and then you can make it's gonna be a social decision, how you weigh how much accuracy you're willing to give up to reduce the bias.

SPEAKER_00So isn't it true then that we can measure fairness in mathematical terms when we plug it into an algorithm?

SPEAKER_02Sure. I mean, you can you can look at you know which of the people were black, which of the people were white, what was the rate of uh getting pretrial release or bail in the black group and the white group. You can compare them. You can also look at their future histories to see whether, in fact, the people that were kept uh in jail were more dangerous, as evidenced by the fact that they committed more crimes later on. Uh, and so you can study those trends.

SPEAKER_02Now, again, it's not gonna be perfect. Statistics never get you 100% certainty. Uh, but I'd rather see those decisions made on statistics for 150,000 cases, than some judge that says, I've been a judge for years now, and I've seen hundreds of cases, and I know people, you know, what? Like you've got seen a very bad sample that's not representative at all, and you've only seen a few hundred, maybe a few thousand, and the AI is, you know, looking over a hundred times as many, or at least a, you know, at least ten times as many.

SPEAKER_00In the context of um like organ donation and transplant, if we can bring it back to healthcare decisions here, um, mainly because one of the one of the problems with uh trying to correct for fairness in organ donation and transplants is that it is true that there are proxies for things like race and gender. And those proxies don't lie. If uh a certain race has on average would you like to finish the song?

SPEAKER_00Thanks for that, Shakira. A surprise guest. And so uh, you know, for example, a a certain race might actually statistically have worse health. And so it is true that implanting this organ into a black person will do less good uh because that person will already have a shortened lifespan than if it was translated into a white person. And so what I'm asking is not simply accuracy in prediction, because we can predict which organ will do better in which body.

SPEAKER_00But then over if we continue to use that algorithm, what's gonna end up happening is that the algorithm will prefer to transfer organs to white people. And so what I'm asking if we can correct for fairness, can we add in like a little nodule that says every fourth transplant to a white person should be to a black person regardless, just so that we're not systemically favoring white people in these sorts of decisions?

SPEAKER_02So again, it's gonna be a social decision. You could do that.

SPEAKER_00You can.

SPEAKER_02But if you did, you would be sacrificing health. So if you look at you know quality adjusted life years after the transplant, you would be sacrificing some of that for the sake of fairness. Now, you might want to do that. That's gonna be a social decision, um, but you ought to recognize what you're doing. So black people in the U.S. uh have higher rates of diabetes, and someone with diabetes is less likely to be a successful transplant. Black people in the U.S.

SPEAKER_02uh have less support at home because often the relatives are, especially poor blacks, or are working, you know, two or three jobs. They're not gonna be home to help the person after the transplant. Is over. Black people in the U.S. have less healthy diets because they live in food deserts where the local grocery stores are not carrying fresh fruits and vegetables. Not always, of course, but on average, more often. And as a result, if you take into consideration those things in who gets a kidney, blacks are going to get them less often.

SPEAKER_02But if you don't take those into consideration, you're going to be wasting kidneys because you're going to be giving kidneys to people who are not going to survive after the transplant. So what's the solution? The solution is get rid of food deserts. Give these people, you know, good social support. You know, those are the solutions. You know, bring down the price of insulin so the diabetics can control their disease better. But you can't expect the AI that's distributing kidneys to solve those giant social problems.

SPEAKER_02The way to get rid of that bias, I think, is to get rid of those social problems.

SPEAKER_00No, I I completely agree with you. And I hope that's a good idea. It didn't come off like you cannot do both. I think, though, that there would be room for justifying not using AI algorithms to be like the sole decision maker in these sorts of decisions, or having maybe a human in the loop, I think is a phrase commonly tossed around in AI ethics, or uh ensuring that some human administrator is overseeing the systemic use of a particular AI tool and decisions like organ transplant just to make sure that it doesn't result in the, I guess, overuse or the reinforcing of oppression.

SPEAKER_00I do think personally for me, it is okay sometimes to sacrifice, you know, qualities or quality-adjusted life years in favor of equality or equity. But I think that's room for debate. And as you said, it's a social value. It's up to the community in which it's deployed in order to determine that value.

SPEAKER_02A lot of people want humans in the loop. What's common to those people? They trust humans. If you're pessimistic about humans, I don't know that I want a human in the loop because they're just gonna mess it up. When they tried to put humans in the loop when AI was distributing bail, the humans mess it up. They took people that the AI said were were dangerous, and uh they let them out on the street, and then those people committed crimes. You know, so putting the humans in the loop is great if the humans are you know careful and more likely to get it right than the AI.

SPEAKER_02But if the humans are just displaying their biases, why do I want this person to get bail even though the the algorithm said they're they're dangerous? Well, because it's my cousin. Well, whoa, now you don't want the human in the loop.

SPEAKER_00Yeah. Now that we've talked about this a little bit more, I can see that maybe my claim might be more limited to cases where, you know, it's it's a one quality life year difference between a white person and a black person in favor of the white person. And in that circumstance, you might say, okay, well, yes, it's one year, but it is quite, it could be arbitrary. Prognostics aren't amazing. And so we could, in that situation, favor quality or equity. But you know, if the difference is five to ten years, then maybe we might be better off justified following the AI prediction.

SPEAKER_02Absolutely. And those close calls, those are the ones that we find human judgments are not stable on, though. So, you know, okay, it's a close call. How are you gonna go? Well, if you put a human in the loop, on Monday they're gonna give it to patient A, and on Tuesday, they're gonna give it to patient B, and on Wednesday they're gonna give it to patient A. On Thursday, they're gonna give it to patient B. Now, that makes it sound arbitrary, but maybe that's fair because now it's arbitrary. You know? Uh I don't know exactly how to react.

SPEAKER_02We're we're studying, you know, how that distribution process goes, but it is gonna be a social decision, what you do about it.

SPEAKER_05Yeah, I think that whole thing raises a really interesting set of questions about how we decide what is important in the first place. And this might be just something that we as a community have to decide. Because right now it sounds like AI is gonna be really good at helping us make decisions that are consistent with a certain set of values that we've decided to have. But if that set of values is the wrong set of values in the first place, the AI is just gonna be really good at teaching us how to uh helping us make the wrong decisions every time, uh, but really consistent about making those wrong decisions.

SPEAKER_05So is there more that can be said here about how to make sure uh make sure that AI is making not just the most consistent decisions, but uh consistent decisions that are consistent with the right set of values?

SPEAKER_02So I just want to repeat, there's no way to make sure there's no certainty, there's no perfection, give it up, you're never gonna find that in this world.

SPEAKER_05Giving us the best chances.

SPEAKER_02The best chance is to correct for ignorance and partiality and instability. That's why I like this idealized observer approach to ethics, that you don't ask people, what do you think about this, but what would you think if you were better informed and so on. So, for example, a lot of people say uh that well, let's go to the legal context instead of the medical context for a minute. If you, you know, that you shouldn't, you know, be uh denying bail to more blacks than whites.

SPEAKER_02But if it is just a fact of our society that more blacks are going to commit crimes, and if you release them in the name of fairness and they go out and hurt people, and if most black crime is black on black crime, then now you're taking these innocent blacks who live in the community with these criminal predators, and they are now getting hurt, and they are and now let's make sure that we know all those facts when we decide what we're gonna do about this.

SPEAKER_02And that I I I should quickly say that is not true for every black community. Of course, of course, we're we're talking on average here.

SPEAKER_05How do we know though? How do we know what the ideal observer is gonna choose? I don't even know what the ideal version of me is gonna do uh tomorrow, you know.

SPEAKER_02So my hero in this case was Thurgood Marshall. He he said in a case in the early 70s on capital punishment, uh, that if people knew more about capital punishment, they would be opposed to it. The standard he was applying was a standard of, you know, contemporary standards of decency. Uh and so he wanted to know. And he said if they knew more. And then some people at Duke uh did studies and found that sure enough, the more people know about capital punishment, the more opposed to it.

SPEAKER_02Right? And so you can do correlation studies like that. You can also do interventions, give them more information and see how that affects their views. So they develop methods like that for idealizing your judgment of capital punishment by giving you more information. We can do that on kidneys too. Give them more information about you know the rates of diabetes in blacks, about the success rates of operations when people have, you know, um less support at home, and give them the information and you can see how much they know and see if it correlates with how much weight they put on the different factors.

SPEAKER_02And you can also do interventions, give them more information and see how that affects in those particular cases the weight they put on. And then you extrapolate to what they would say if they uh knew a lot more than they do now. Of course, nobody knows everything, and the extrapolation is gonna be risky, but you know, you do the best you can. That's how you know what they would do in the more ideal circumstances. You do careful control studies.

SPEAKER_05So it's it sounds good when it comes to specific questions, but when it comes to deeper value questions, it's not as clear. Um, so I I'm gonna cite this this Phil Papers survey that was done uh in 2020, I believe, and they asked uh as many philosophers as they could, you know, uh experts on philosophy and ethics and all that, uh, what moral theory do you subscribe to? And if I recall, it's it's almost an even split. You know, about 30% uh classify themselves as deontologists, another 30% as consequentialists, and another 30% as like virtue ethicists.

SPEAKER_05So, in that kind of case, it's like, all right, that gives us no clue uh about what uh moral theory a person would choose if they were m experts in moral theory.

SPEAKER_02So I'm gonna cite another shocking fact. Philosophers are weird. Everyday people don't subscribe to consequentialism or Kantianism or virtue theory. They don't know what they think. Right? And so they're working with their intuitions and a few rules that they learn from their parents or their church, you know, and something like that. Uh so uh you are gonna get differences among those people, but you'd be surprised you don't find that many differences on some cases.

SPEAKER_02There are an awful lot of cases where you know every kind of commonsensical person out there, uh almost everybody ends up with the same um conclusion. I mean, we got way over 95% of people going race should not affect whether you get a kidney. Gender should not affect whether you get a kidney, sexual orientation should not affect whether you get a kidney, and so on. And so, you know, they're all saying that, and then you know, you have to correct for when they don't meet their own standards, but they're not gonna be like tied to a theory uh the way philosophers are.

SPEAKER_00One of the things I kind of want to go back to is the uh discussion around how we can use AI to predict when people might commit crime in their community. And you made the point of saying, well, you know, if we try to uh correct for fairness, to be like, okay, well, there is a disproportionate number of black people who are being sentenced or uh to avoid committing crime in the community. And the reason we we we want to correct for that is because you know they might actually hurt others. Um and so someone else will be paying the price for us trying to be fair in this system.

SPEAKER_00And I was just thinking about it, just seems so unfair that these AI systems will continue to reinforce certain norms. Um, and so the reason I was thinking about this is because I'm aware of, you know, the statistics around committing crime is generally those, it's not the rich, white, privileged people in society, at least getting caught on things like violent crimes. It's people who have systemically been disadvantaged over and over again or who get particularly get caught in the juvenile system at a young age. And, you know, if you get put into prison at a young age, it's just so likely that you're gonna get caught in a cycle of crime and you'll quite likely never make it out.

SPEAKER_00And then I was thinking, but you're so right. Uh why should other people pay the price for us trying to correct for fairness? And I think I've realized that there is a bit of a tension here between what I would see as morality and empathy. And I I think I'm just struggling to equate that. I just I I wish we didn't have to use these these systems because these people have already committed crimes, you know, or or in in it, particularly in America and Australia, it's may not be true actually that they've they've convicted been convicted of crimes or actually done those crimes because we know about bias in uh police officers picking up people in the streets, for example.

SPEAKER_00And so I think I'm just struggling to reconcile my understanding of like the socioeconomic system in which we understand people and who commits crimes and trying to protect others by using AI prediction. And I was just wondering if you had any comforting words for a little for little old me trying to work my way through this this this, I guess, uh juxtaposition I found myself in.

SPEAKER_02Well, I mean, the simple answer is to make the uh prison system less criminogenic so that you can send someone to prison and they won't come out a worse criminal. Right? And it's amazing what little things you can do to stop that. So there was one um the Donaldson Correctional Facility in Alabama at one point was having a trouble with inmate violence. And it's inmate, inmate on inmate violence that often turns people into criminals because they're getting raped and attacked, and by the time they get out, they're just so hardened and so on.

SPEAKER_02So, what's the solution? I mean, you wouldn't believe how simple it is. Cable television.

SPEAKER_00I thought it was gonna be cats.

SPEAKER_02No, oh no, but cats help too. Oh, there's a big program in the US to give pets, and it's amazing. You give like a triple murderer, you know, a pet kitten, and they'll take care of that kitten, and they won't hurt anybody, they'll just sit there petting it. Don't touch their cat or they will kill you. But but if you, you know, have the kitten. And so animals, uh, I had a student who did a paper on this, it was wonderful. Uh, but then also cable television. They'll just watch television all day and then they don't fight. Why do they fight? Because they're bored out of their minds because they got nothing else to do.

SPEAKER_02Uh but in Alabama, what happened was they put cable television in the prison, and then the local people complained because they said, I can't afford cable, and you're giving cable to the prisoners. That's not fair.

SPEAKER_00Cable TV is a basic human right.

SPEAKER_05Yeah, how much is a Netflix subscription?

SPEAKER_02Just give them a yeah, you know, I mean, uh Thurgood Marshall is my hero for many reasons. Uh, but one time in one thing they said, you know, there's this you gotta pay to vote. And the person said it's only a dollar and a quarter. And he said, You don't understand how much a dollar and a quarter is to a poor person. Like, you know, what seems like a small cable subscription to somebody in, you know, in our situation, or we're very lucky, um, is you know, it's just unaffordable. It's like if you pay for the cable, you're not gonna be able to buy food for the table.

SPEAKER_02You know, it's the cable or the table.

SPEAKER_00We have our podcast title, everyone.

SPEAKER_05It's incredibly catchy line.

SPEAKER_02So, I mean, that's so that but that's an easy thing to say. Let's correct it. That's gonna take years and years. What do you do now when you face it? That's a really tough decision. And a lot of these really tough moral dilemmas that we face are due to us creating a situation where there's just no good answer.

SPEAKER_00So it's our own fault that we have to make these tough decisions. All right.

SPEAKER_02That is the conclusion. One thing I would say is uh America's got a lot of problems now. Don't get me started. But the idea of having 50 states where every state tries something different and then you can see what works, it's not so bad. You know? Who would have thought different states will try different things and they and they learn from each other and see what works. That's pretty cool. You know, you could actually do experiments in the criminal justice system to see what the effects are of the different different programs.

SPEAKER_00Can I ask one more question? Of course. Which is it uh something you said triggered it, which is uh we've got these really, really tough decisions that we just have to make now. And my brain sort of just went, I think I would rather a person. And I know we've talked a lot about accuracy, and uh judges are just, you know, they're not great. They might be trying their best, but they're not great. And I think the thing that triggered it in me was I feel like at least I can say that someone was trying, or maybe there is like an element of that human relationship that we're kind of all in this together, we're trying to do our best.

SPEAKER_00Someone, someone tried. But with AI, no, no one's trying. There's no one there. It's just, you know, it's the cloud, it's an algorithm.

SPEAKER_02So let's just get the scenario straight. I don't want to make fun of you.

SPEAKER_00No, no, I'm not saying it's because it's lies or rational. There was just a little tug on my heart.

SPEAKER_02I'm so glad the judge is trying their best. They got it wrong and they discriminated and didn't know what they were talking about, but they were trying, you know, and all these other people outside the courtroom that didn't even get represented are suffering from crimes that are the result of this judge letting out a dangerous person. I mean, I'm you know, I agree. It's nice that people are trying. It's so I'm a contrast of this. It's better that they try than that they don't even try.

SPEAKER_02But it's better that they try and succeed than that they try and fail. And the problem is too often they try, but they fail. And the people out on the streets suffer as a result.

SPEAKER_05I'm gonna try to boost uh Sinead's intuition by that. Uh and I think maybe you won't what?

SPEAKER_00I said, what? No, this is the first time everyone on the boat cast.

SPEAKER_05Radical, excuse me. I just radical about face for this podcast.

SPEAKER_00What?

SPEAKER_05Teaming up against a guest, it sounds bad that way. Um I I know you're you're you you've told me you're a consequentialist, so you might not be swayed by this opinion. But um I think I have an intuition that there's something about living a dignified life. You know, somebody gives me dignity by thinking about me, especially when there's a serious decision to be made about me. I want somebody to have to think about my my situation, they take it seriously, weigh all the moral considerations, and that feels more dignified to me than somebody just pressing a button on a computer and but the computer spitting out something.

SPEAKER_05Uh all else being equal, if they were both as accurate, I would rather somebody take the effort to think about me. And I think it's got something, an intuition about dignity and how having other people put in the effort to think about you uh are related to each other in some way. You're gonna have a hard life.

SPEAKER_03Because I know this.

SPEAKER_02If you don't have to have T-Rex skulls in my home. No, no, what I mean is if you don't have dignity in yourself and self-respect, and you're depending on other people, you know, thinking that way about you, you're gonna run into a lot of people that don't do that. I mean, I'm I think our, you know, our personal dignity depends on our status as individuals. It doesn't depend on whether other people, you know, show us certain attitudes or not.

SPEAKER_00I disagree. There's a and um there's a you know, the school of thought of feminist bioethics, which talks a lot about autonomy and your sense of self this self as being constituted in your the relationships and community that you find yourself in. And so when other people systematically don't show you self-worth and don't respect you and don't trust you, you can and do start to internalize that and lose your own sense of self. Yes. It's not to say that it's bad that we depend on others for self-respect or for dignity. It just says that that's just the way we are.

SPEAKER_00But I do agree with you when you said we we run into people who don't do that. And we should not be relying on every single person everywhere for that sense of self. But with what James said, when someone is making a very important decision about my future, there is an intuition that I would kind of want a person to do it. But after talking with you, I think I that's just being a sort of a bit of a narrow-minded humanistic view. And I think maybe I just need to start accepting technology for what it is, which is sometimes it's just better.

SPEAKER_02Yeah, sometimes it's better at making these complicated things.

SPEAKER_00I'm just gonna have to get okay with that.

SPEAKER_02Whether AI should be in personal relationships is another thing. We could talk for a long time about whether you want to have an AI therapist, right? Because I mean, you know, the downside is it's just a machine, not a human. But the upside is hey, you can call them at 11 p.m. when you're really feeling down.

SPEAKER_00That is an excellent point. James, 2 a.m. you finish your video games, need a therapist.

SPEAKER_05Yeah, that's exactly what I want to do when I'm tired and hungry after a long like session of a long session of bald as gay. Yeah, all I need to do is a good session of introspection and digging up my problems.

SPEAKER_00I think the world would be a better place if you did.

SPEAKER_05The whole world would benefit. I'm that influential. Yeah. Uh-huh. Uh-huh.

SPEAKER_02Yeah. But I agree with you that, you know, uh community builds self-respect. It has that causal effect. I didn't mean to deny any of that. But I want to say I mean, a particular case that I'm you know particularly interested in for personal reasons is uh a trans person who everybody in their community thinks less of, I still think they have dignity.

SPEAKER_04Oh, yes. Yes.

SPEAKER_02And the fact that other people are not showing you respect doesn't mean that you don't deserve respect.

SPEAKER_00Yes, that is a really, I think, a really great way to put it. But we've run out of time. We know that you have lunch with Julian at 12:30. This has been a really fascinating discussion.

SPEAKER_02Yeah, thank you. Well, thanks for having me.

SPEAKER_00You're welcome back anytime.

SPEAKER_02Oh, I hope I'll come back soon.

SPEAKER_00Excellent, thank you.

SPEAKER_01Thanks for listening to our new podcast. Subscribe online and tune in every first of the month. In the meantime, listen to CBME and you and be savvy.

Open the transcript file

Hosts and Guests

Nobody is credited on this episode yet.

Credit the people on this episode

Be the first to rate this episode

More episodes

All episodes
  1. Can Nihilism Improve Your Mental Wellbeing?

    Nov 2, 202645 minAudio

    Send us Fan Mail We spend much of our lives searching for meaning. Some of us look for it at work. Some in our families and friends. Some look for it at an overpriced spiritual retreat in Bali where some dude named...

    Credit the people on this episode

  2. Insurance Discrimination: Is it Fair? with Harisan Nasir

    Sep 24, 202649 minAudio

    Send us Fan Mail In some insurance scheme, you might be required to pay more because you are more prone to disability. In another, you may have to pay more because you're a woman (and are statistically likely to live...

    Credit the people on this episode

  3. Genomic Newborn Screening: Where less really is sometimes more with Ainsley Newson

    Aug 25, 202655 minAudio

    Send us Fan Mail Did you know that genomic sequencing can now screen newborns for hundreds of rare conditions right at birth? That looking too closely at our DNA might actually cause more anxiety than answers? And is...

    Credit the people on this episode

  4. The Hidden Side of Birth: Injury, Trauma, and PTSD with Angela Ballantyne

    Jul 2, 202648 minAudio

    Send us Fan Mail Did you know that over 80% of people who give birth experience an injury? That anywhere between 30 and 50% of people who give birth find it traumatic? And that up to 15% of people experience PTSD from...

    Credit the people on this episode

  5. Brains in Jars with Insoo Hyun

    Jun 1, 202656 minAudio

    Send us Fan Mail Could a blob of cells in a petri dish one day think? Feel? Have rights? In this episode, Professor Insoo Hyun (Harvard Medical School, one of the world's leading experts on the ethics of stem cell and...

    Credit the people on this episode