Trump's Next Online Speech Cop + Doctors vs ChatGPT + Hard Fork Crimes Division

“I feel very strongly that the First Amendment is under the most direct threat that any of us will ever really experience.”

Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.

• 0:00Speaker 00

  Support for this podcast comes from Dell Technologies. It's time to put AI to work. Is your business ready? The challenge isn't getting started with AI. It's scaling beyond pilots while maintaining control of your data, governance, infrastructure, and costs. Dell AI Factory provides the integrated foundation you need to scale AI securely and efficiently with one trusted partner, Dell, AI on your terms. From idea to impact. Learn more at dell.com/yourwaytoAI. That's dell.com/yourwaytoAI.

• 0:35Kevin Roose?

  Casey, what's going on?

• 0:37Casey Newton?

  Well, I have really changed my feelings about Blue Sky in the past week.

• 0:44Kevin Roose?

  Yeah?

• 0:44Casey Newton?

  You know, before last week, I have to admit, while I did use it and I did occasionally see stuff on there that I thought was like really funny or interesting. The feed was so political that I had just sort of written it off as not for me. But then something really powerful happened, Kevin.

• 1:04Kevin Roose?

  What's that?

• 1:04Casey Newton?

  Which is that my following doubled in four days.

• 1:09Kevin Roose?

  My too, I got like a weird number of followers.

• 1:11Casey Newton?

  I mean, something is happening.

• 1:14Kevin Roose?

  Yeah.

• 1:14Casey Newton?

  There is something in the water.

• 1:15Kevin Roose?

  Yeah.

• 1:16Casey Newton?

  And what I think is funny about it, Kevin, is that my own experience reminded me how truly small-minded and petty we are about these social networks because we will speak in such lofty terms about, well, you know, here is kind of the vibe of the site. And well, this one, you know, suppresses political content and, you know, oh, I hate all the ads over there. At the end of the day, it is how many followers do I have? And how many little internet points did I get for making my little quip? And wherever the number is the highest, that is where you will find me. And that is all to say, please follow me on Blue Sky.

• 1:53Kevin Roose?

  I'm Kevin Drew, a tech columnist at The New York Times.

• 1:55Casey Newton?

  I'm Casey Noon from Platformer.

• 1:57Kevin Roose?

  And this is Hard For.

• 1:58Casey Newton?

  This week, the future of the internet could look very different next year. The Verge's Neelai Patel joins us to talk about President-elect Trump's pick for the head of the FCC. Then, a new study found that ChatGPT outperforms doctors in diagnosing subdiseases. One of the study's authors, Dr. Adam Rodman, is here to discuss the future of medicine. And finally, court's back in session, Kevin. Aside from Hard For, Crime Division.

• 2:22Kevin Roose?

  I rest my case.

• 2:23Casey Newton?

  No, you don't rest your case at the beginning, except the end.

• 2:26Kevin Roose?

  Sorry. Adjourned?

• 2:28Casey Newton?

  No.

• 2:29Kevin Roose?

  No.

• 2:29Casey Newton?

  Stop it.

• 2:31Kevin Roose?

  Where's my gavel?

• 2:32Casey Newton?

  You're ruining the show.

• 2:39Casey Newton?

  Well, Kevin, let's get started this week with a Carr crash.

• 2:41Kevin Roose?

  Oh, yeah?

• 2:42Casey Newton?

  Yeah, Brendan Carr has crashed into the news as the next potential chairman of the Federal Communications Commission.

• 2:49Kevin Roose?

  Yes, so obviously, in the post-election period, President-elect Trump has been announcing many of his picks to lead top agencies. And the one that really stuck out to me that I thought was relevant to the topic of our show was his pick to lead the Federal Communications Commission, the FCC, who's a man named Brendan Carr.

• 3:09Casey Newton?

  Yes, Brendan Carr is a Republican. He's been on the FCC since 2017. And the FCC has five members on it. And collectively, they do control broadcast media in this country. They have a lot of legal authority to do that. They have less legal authority over the future of the internet. But Brendan Carr is somebody who has a lot to say about how he thinks the internet should work.

• 3:33Kevin Roose?

  Yeah, he's a real activist in some of these debates about internet regulation. He's been very vocal about going after big tech, as he calls it, for their political bias and what he views as anti-conservative censorship. He's constantly picking fights with people online and sort of defending his vision of an internet free of left-wing censorship. And he also starts every day, literally every single day, by posting On X, good morning and God bless America.

• 4:02Casey Newton?

  And, you know, up until recently, I think, you know, if you're somebody who does not agree with Brendan Carr on these issues, it has been easy to dismiss him as a crank. But within a couple of months, he is going to be a person potentially shaping internet policy in this country. And whenever internet policy issues are in the news, I want to know what Neelai Patel thinks. Neelai was my old boss at The Verge. He's a co-founder of The Verge. He is also a formidable podcaster and one of our greatest adversaries in the realm of podcasting as the host of Decoder and The Vergecast. And Neelai has been writing recently about what the arrival of Brendan Carr as chairman of the FCC could mean for not just the internet, but for speech in America in general.

• 4:47Kevin Roose?

  Yeah, Neelai thinks that we are headed into a truly scary and dark timeline with the appointment of Brendan Carr at the FCC. And as someone who has not followed Brendan Carr's career super closely, I'm very curious to understand why he thinks this man poses such a threat to the future of the internet.

• 5:04Casey Newton?

  Yeah, we're going to hear about that, but we're also going to ask him for maybe some more empowering thoughts that we can bring into this next chapter in American history. So with that, let's bring in Neelai Patel. Welcome to Hard Fork, Neelai. It's great to have you here. And I want to get started with some basic background information. Who is Brendan Carr and why does Trump see him as a, quote, warrior for free speech?

• 5:33Speaker 03

  Yeah, Brendan, he is an extremely online commissioner of the FCC. He was appointed by Trump. His view is that the FCC should spend a lot of time regulating not only the traditional purview of the FCC, which is wireless spectrum and broadcast television, but also big tech companies. And he's got a lot of ideas about how he might get that power and then how he might use that power. But really what you have is a guy who likes going on Fox News and Twitter and railing about how there's a censorship cartel in big tech and it should be crushed. And I think Donald Trump likes that a lot.

• 6:14Casey Newton?

  So most people probably don't think of the FCC as a very powerful agency in their daily life, but Neil, there was a time in recent American history where it did play a larger role. So tell us a bit about the recent history of the FCC.

• 6:26Speaker 03

  So, most Americans, I think in 2024, never think about the FCC. It's not our mind, especially the last five years. The FCC has really receded. It has been sort of a neutered agency. No one trusts it anymore to do the things that people want it to do. That has not always been the case. So if you go back 20 years, the FCC is a cultural force in America. And this really hit its peak with Janet Jackson at the Super Bowl, right? Justin Timberlake.

• 6:52Casey Newton?

  Nipplegate.

• 6:53Speaker 03

  Yeah, he rips off the corset, the bodice. America is forced to endure like half a second of a nipple and the world goes crazy. And if you remember, the George W. Bush administration like hated nipples, right? John Ashcroft is George W. Bush's Attorney General. He famously covers up the Statue of Lady Justice at the DOJ because God forbid lawyers see nipples. It's just like a very weird time in America. And all of this is based on the dominance of broadcast media. Most Americans at this time get most of their media from broadcast television and radio stations over the airwaves, right? You hang up an antenna, you get NBC or CBS or ABC, you hang up an antenna on your car, you get whatever local radio station.

• 7:35Speaker 03

  And that spectrum is owned by the government and it's licensed out to these broadcasters in the public interest. And that's really where the FCC's authority comes from. And there's like a long string of Supreme Court cases that basically add up to the spectrum belongs to the people, the government gets to make rules based on a spectrum, and Americans do not want nipples on their public airwaves. So like, we're gonna, we're gonna freak out about this. So this is like the high point of the FCC as a cultural force. And what happens during all of this is the iPhone comes out and YouTube is introduced and podcasting is introduced. And Americans, by and large, switch to cable television in huge numbers. They stop watching broadcast TV. They stop consuming this content. So the FCC itself... Says, we have to get out of this business. We have to be out of the speech policing business.

• 8:18Speaker 03

  Michael Powell, who's Colin Powell's son, is chairman of the FCC. And he's like, we got to stop this. These broadcast providers, they're not competing with each other. They're not the dominant force. They're competing with cable television.

• 8:30Kevin Roose?

  And cable television is not regulated the same way as broadcast television because it doesn't go over the air on this sort of publicly owned spectrum.

• 8:38Speaker 03

  Right. It's on Comcast wires, not the public's airwaves. You know, he's a Republican FCC chairman. He's like, we're making these really weird rules for these companies. We have to get out of this business and we need to get into the business of broadband deployment. And this has largely been what the FCC has been focused on since 2011, 2012. That's where you get the big net neutrality fights.

• 8:59Kevin Roose?

  And net neutrality, just for people who are not steeped in this, the history and the context here is basically the, the rules that say, what? If you're Comcast, if you're, you know, an internet service provider, you cannot dictate what goes over your pipes, right? You can't impose, you know, censorship or speech regulation at the level of the internet service provider. You are just supposed to be like the dumb pipes. That, that is essentially net neutrality, correct?

• 9:23Speaker 03

  Yes. And it's really predicated on, I see that you're in a massive competition for internet content, which turned out to not be the case, right? You, you only ended up with a few, a few giant platforms. You ended up with YouTube and TikTok and Meta, like those are your choices. So something very weird happened along the way of the internet where we recreated the dominance of broadcast television, just a handful of giant companies that control most of the media in the country, without any of the legal foundation for how the government might get involved in that content. So this is like the, this is the stage we're in, right? We unwound this previous dominant broadcast media regime where we had pretty overt speech policing.

• 10:05Speaker 03

  All the way to, well, if you want to see a nipple, the internet will provide you a nipple at any time and like no one cares anymore. But you still have a lot of people who are very interested in how the platforms moderate and the political biases of these platforms. You have a very active right wing, which is insistent that any moderation at all that disfavors them is a moral catastrophe that should be stopped with a full weight of the government. And this is where you get Brendan Carr, who up until recently was a pretty normal, if somewhat overly online deregulatory force. Right? That's, that's his worldview until a couple of years ago, when the big push for, well, we should start yelling

• 10:45Speaker 03

  at Mark Zuckerberg more to make sure the algorithm favors conservative viewpoints, or at least doesn't overtly favor liberal viewpoints. Brendan takes this up. And in the sweep of this history, you can see that what he wants is to be an old school chairman of the FCC. Where if you're mad about nipples on Instagram, you can write him a letter and he will have the power to fine or otherwise penalize Meta. And that is just the wheel that is turning right now.

• 11:14Casey Newton?

  And so I think that gives us a good sense of like what, what this person's worldview is and what he might do if he had that sort of power. I guess the next logical question that Neelai is, does he have this power? Is any of this authority in the FCC? And if it's not, you know, what do you expect him to do about it?

• 11:30Speaker 03

  Yeah, I hope, I hope I did a good job of laying out what feels like a logical pendulum swing.

• 11:35Casey Newton?

  Yeah.

• 11:36Speaker 03

  But actually legally, none of this makes any goddamn sense. Like in a very real way, he does not have this power. And he was the author of the Project 2025 chapter on the FCC and what you might do with it and how you might use it. Notably, Project 2025 does not say we should dismantle the FCC. Like it says, we should dismantle every other agency. Brendan's chapter of Project 2025 says the FCC should get even more power.

• 12:01Kevin Roose?

  It should be even more involved in things. The chapter that he wrote, I think most people associate Project 2025. This is a sort of document, this sort of roadmap for a second Trump administration that was put together by, you know, some conservative think tanks and groups. And I think most people associate it with advocating for rollbacks on, you know, abortion and other social and cultural issues. But it actually does have this sort of interesting part about the FCC and how Brendan Carr specifically wants to regulate the internet. And a lot of what's in this chapter is, you know, boring sort of normal FCC chair stuff about spectrum auctions and rural broadband access and stuff. But he starts with this thing about sort of reining in big tech. So, Eli, what is Brendan Carr's idea, his big idea for how to rein in big tech?

• 12:46Speaker 03

  For, I want to be very clear. My personal opinion of Brendan Carr is this man isn't capable of having big ideas.

• 12:53Speaker 03

  I do not have a high opinion of Brendan Carr. But his idea is the same idea that everyone else has, was, which is, we should mess with Section 230 until the platforms do what we want.

• 13:03Kevin Roose?

  And Section 230, for people who are not experts, is the part of the federal law that basically shields online platforms from legal liability over user-generated content, right? So if you post something illegal on Instagram, the government can go after the poster, but it can't go after the platform.

• 13:22Speaker 03

  I appreciate that you think there are hard fork listeners at this point who don't know what Section 230 is.

• 13:27Kevin Roose?

  There might be a few.

• 13:29Speaker 03

  Section 230, the stakes of messing with 230 right now are, do you want YouTube to exist? Those are the stakes of 230, right? Do you want any user-generated platform to exist at scale? Because if you make Google liable for the content on YouTube, there will quickly not be content on YouTube. You will actually turn them back into cable companies. So these are existential stakes.

• 13:49Kevin Roose?

  And Brendan Carr does not propose getting rid of Section 230, interestingly, as some conservatives have done. He doesn't say we should repeal the whole thing. What he says instead is we should sort of limit these court-added extras that judges have piled on top of Section 230 to sort of extend the shield granted by the original law.

• 14:08Speaker 03

  Right, and that's the part that is wholly nonsensical. Like, it is a fantasy. So, first of all, Section 230 is a law. Congress wrote it, it's famously 26 words long, and it has gone to court numerous times, in numerous ways. And the courts have uniformly upheld the idea that these 26 words are there to keep platforms from being liable for what their users post, over and over and over again. There's not a bunch of court-added additions to this. That just doesn't exist in law. Second of all, even if there was, you get around that by Congress doing more stuff. You don't get around that by being an unelected chairman of an agency most people don't give a shit about, and just issuing decrees about what the law means.

• 14:52Speaker 03

  That's just not, that's fully not how it works. And the courts in this country, particularly conservative justices of this country, do not believe that agencies should have any power. So even if you're Brendan Carr, not only does he not have that power, if he tried to use that power, he would run right into the conservative legal movement, which is trying to defang the agencies in a very specific way. So it, it's just, none of this makes sense, except, "Well, if I can wield this weapon over the big platforms, they might do what I say anyway." And that is very much the animus of every attempt to modify 230. No one's actually saying, "We should get rid of this law that allows YouTube to exist." They're saying, "If we threaten this law enough,

• 15:37Speaker 03

  YouTube's Trust and Safety Team will moderate YouTube the way we want."

• 15:41Casey Newton?

  Which I think in practice probably has happened. I think platforms have been responsive to those sorts of threats. You know, if you're the sort of person who likes net neutrality, you like Section 230, Neelai, I feel like people might hear what you're saying and be excited about it. They think, "Okay, cool." So this man is, he's sort of, you know, like, like, you know, banging pots and pans and trying to get everybody all scared, but there's really not a lot of legal basis for what he's threatening. So maybe that you might feel relieved. At the same time... I feel like we're in a world where we can rely less on judicial precedents than we've been able to in the past. So many things that seemed like a slam dunk either turned into a coin flip, or the Supreme Court decides to throw out decades old precedent.

• 16:24Casey Newton?

  So as we move into this new world, this new Trump administration, how are you thinking about that risk and how the internet might change just because we might be sort of living in a legal chaos land?

• 16:36Speaker 03

  I feel very strongly that the First Amendment is under the most direct threat that any of us will ever really experience. The rise of the internet that we know coincided with a period of pretty unfettered expression. The government was told not to regulate the internet. This phrase came up over and over again. Leave these companies alone. We're gonna let a thousand voices bloom. We're gonna get over a lot of weird indecency ideas we have about media in general. More people have more access to speak. That is an unqualified good thing, and we're gonna leave that alone. And that is an interpretation of the First Amendment, or at least a First Amendment environment that I think most people are used to right now. That is our expectation. Those walls are gonna come in closer. What

• 17:20Speaker 03

  you're getting out of the Brendan Cars and the Trump world version of the First Amendment is closing in is my political opponent should be silenced, or the platform should make sure to favor us, and we will wrap it up in what sounds like a defense of free speech. But actually, what it is, is punishment. And you see that over and over again. You see it expressed as punishment. You see Elon Musk, who runs an ISP in this country, saying the hammer of justice is coming for people who publish election hoaxes. Well, lying is legal in America. It's just fully legal. Hate speech is legal in America. We've run this all the way up to the Supreme Court multiple times, and it's just legal to lie.

• 18:05Speaker 03

  It is legal to be racist. We allow the government does not punish these things because we expect the market to punish these things.

• 18:13Casey Newton?

  Yeah, I mean, on that front, Carr recently sent a letter to the CEOs of four big tech tech companies, so Apple, Meta, Microsoft and Google blaming them for what he called an unprecedented surge in censorship, warning them that they might face investigations, not just for their own content moderation, but for work they do with third-party groups like NewsGuard, which do ratings for news sites around bias and accuracy. Do you see that as just kind of more pure intimidation?

• 18:42Speaker 03

  Yes. But, okay, I'm curious, you live in this world, you've covered trust and safety a million times. This idea of groups like NewsGuard, where you have this appeal to a third-party that will tell you how biased your news is, has always been problematic.

• 18:57Casey Newton?

  Yeah.

• 18:58Speaker 03

  But do you think the government should have a role to play in telling you?

• 19:01Casey Newton?

  No, absolutely not. I mean, you know, like, what, what activity could be more protected by the First Amendment than saying, like, I think this website is biased? You know, like, there, there is, there is no even theory of harm there, right? Like, I can, I can see how a car and his allies would come along and say, oh, there's this, like, giant censorship apparatus. But in practice, you know, sites like NewsGuard aren't even particularly widely used, right? And there are all kinds of these rating services that I think most people basically ignore. But to me, that's actually what makes it scary, is this thing that isn't even that influential, you know, is suddenly the target of an FCC commissioner who is now threatening platform owners, saying, do not work with these people. I mean, to

• 19:44Casey Newton?

  me, that seems like the much greater threat to speech than, you know, some website that says Fox News leans conservative.

• 19:50Speaker 03

  Right. And the piece of that that really worries me is there's no legal mechanism to mess with these big companies. We're all basically nation-states unto themselves. Like, you can fire threats at Jeff Bezos all day. Like, what's he gonna get on his yacht and sail away from you as fast as he can with his four support yachts in tow? And he'll just be waving at them in front of the beach. Fine. But there are speakers in America where Brendan will have the power, right? So the actual broadcast networks still use the spectrum.

• 20:18Kevin Roose?

  Right.

• 20:19Speaker 03

  And Kamala Harris shows up on Saturday Night Live and Brendan Carr gets to yell about revoking the broadcast licenses of NBC, which also makes no sense because it's the stations that have the licenses, not NBC proper. And he knows that, but it doesn't matter because you go on Fox News and say, I'm going to revoke NBC's broadcast license for having the temerity for a presidential candidate to be on their program, even though like the next day, Trump was given free airtime during a NASCAR race, right? Which is the, the rule that the government has and NBC is very good at fulfilling this rule because they've been a broadcaster for 5,000 years.

• 20:54Kevin Roose?

  Yes, I have two questions about this letter that Brendan Carr sent to these big tech CEOs. One of them is he, the ones that he included were somewhat mysterious to me. So I get why Meta and Google are on this list. Conservatives have been mad at those two companies in particular for years about perceived censorship. But what are Apple and Microsoft doing on this list? What kind of objectionable content moderation are they doing in Brendan Carr's eyes?

• 21:17Speaker 03

  Apple runs the App Store and in order to have an app on the App Store, you have to pass Apple's rules of acceptable moderation. So I think famously Parler was kicked off the App Store. Gab was kicked off the App Store because they were still allowing all kinds of stuff to go by. Apple doesn't want this to happen. If you're Brendan Carr and you want to make sure that no one gets to control speech in America, except for you, the person who runs the App Store is your greatest enemy because he can keep the platforms off of phones entirely. Microsoft runs a bunch of big platforms. Sure, like you might be worried about being, but they are also a huge developer of AI. And I think Carr is smart enough to know that, you know, the next turn of all of this is what the AI search results are. And if the AI starts to say, hey, this is misinformation, if Grok on X literally says,

• 22:01Speaker 03

  Elon Musk is the greatest source of misinformation on X, which it has said recently, that's a, that's a big problem. And I think putting these companies on notice that, you know, you don't want quote unquote woke AI is a big deal for all of these players.

• 22:12Kevin Roose?

  Yeah.

• 22:13Casey Newton?

  You mentioned the broadcast licenses a minute ago. I wanted to pick that up again because you also established earlier that the FCC does have a bit more legal authority with them. So, you know, I agree with you. It seems like nonsense to say, well, one candidate is allowed to appear on TV, but the other isn't. But at the same time, I also do expect that they will continue making those threats. So what sense can you give us of how easy is it for someone like Brendan Carr to wreak havoc with these broadcast networks? And what do you expect there?

• 22:49Speaker 03

  I think it's tremendously easy for him to wreak havoc with the broadcast networks, not because of the law, but because they are inherently weak counterparties at this moment in American media history. They are dying. This is a historically low moment for broadcast television viewership. And even the things that were keeping it alive, the NFL, are moving to streaming. This is a historically low period for cable television viewership, which is how a bunch of these TV networks are making all their money. We'll see. It just, does anybody there have the fight? Because they could win. I honestly believe if they wanted to win these fights, they could body up against Brendan Carr and say, look, we're not going to do speech police in America, and we're also complying with the rules, right? Fully, we are in compliance with the rules. But I don't think that matters

• 23:33Speaker 03

  in a world where the businesses are dying, the executives just want to cash out and leave, and the audiences don't care because it's not watching anyway. And that is very, very dangerous. When I say that I think the Brendan Carr FCC embedded in the Trump administration represents the biggest threat to free speech that any of us will have ever experienced, that is the mechanism. It's the chilling effect with the power they have, combined with their obvious desire to create new power.

• 23:55Casey Newton?

  Yeah, I think that's right. And I think media executives have not quite fully internalized the degree to which the people who are about to take power in this country are obsessed with destroying them. And I think this is quite different, actually, than the first Trump term, when there were also sort of these grave proclamations about what would happen to the media, but largely, media was, you know, fine, or at least there were pockets of it that had a Trump bump from the first Trump term.

• 24:24Kevin Roose?

  I think this is different because I think for the people who now are going to be running the country, including people like Elon Musk, this is not just something he thinks about. Occasionally, this is one of his driving priorities in life is to delegitimize and undercut and ultimately destroy what he sees as the legacy media. But I'm also just curious, Neelai, as a person who does understand what's coming, does think about this stuff, like, how do you operate in an environment like that, aside from just hiring lawyers to deal with a bunch of bogus defamation claims? What should you do?

• 24:59Speaker 03

  Well, first of all, Kevin, I'm curious if you think the legacy media continues to exist. My view is that it's already dead, right? What this election showed is that actually Trump's mastery of the YouTube podcast format was much more relevant than whatever happened on ABC News, fundamentally. And so I don't want to spend my time worrying about a thing that has already destroyed itself. And so it's like, the real question that I have is like, if our media is all going to be a bunch of independent creators on YouTube, or independent podcasters buffeted by Spotify's ad rates, or whatever, how will those platforms apply this pressure to our speakers in response to the Trump administration? And will anybody

• 25:40Speaker 03

  even be able to follow the causal line of like Brendan Carr yelled at CBS, so the person who runs podcasts at Spotify made sure to promote The Daily Wire more than something else?

• 25:53Kevin Roose?

  I mean, do you think we would ever see something like an equal time mandate for YouTubers, where like if Jake Paul does a video praising Donald Trump, he also has to do one praising whoever's running against Donald Trump?

• 26:05Speaker 03

  I hope not. Elon Musk likes to say he's a free speech absolutist. He is not, but I might actually be one. I have a lot of complicated thoughts about this lately, but I don't think that we should overcome our own First Amendment in that way. There are laws in other countries that are wacky. In India, there was a law proposed that said if you had a YouTube channel over a certain size, you had to register with the government for preemptive regulation.

• 26:32Speaker 03

  Imagine how the heavily armed American population would react to that idea in this country.

• 26:36Kevin Roose?

  Yeah, I only support regulating YouTube channels like Cocohellen, which are a blight on.

• 26:42Speaker 03

  But that's kids, right? If you go and ask politicians of both sides, no matter how credible or consistent or cynical you think they are, you go and say, where can you find a hook that allows you to overcome the First Amendment and pass some speech regulations that everyone will agree on, they will point to children's content universally. And that's why the Kids on the Safety Act exists, right? That's why, hey, we should make sure that at least this group of people that cannot protect themselves, and we don't think they can make choices in the market to benefit themselves, we protect them at the platform level. And that is also why the platforms are fighting against it so hard, right? Because they don't want to accept that responsibility. But that's about it. There's not a world in which we agree that there should be such a thing as the fairness doctrine for podcasts, because the

• 27:26Speaker 03

  solution is to just have more podcasts, right? And that basically is this, like, there's an infinite amount of podcasts, and that should be.

• 27:35Kevin Roose?

  There truly are not, Neelai. You know, you have two. I mean, we can keep creating podcasts in this country.

• 27:43Speaker 03

  I will fix free speech by just starting new podcasts every single day. But that's what I mean. You can either have competition or you can have regulation. And up until recently, our solution has been competition. And I think what we're all kind of realizing or maybe waking up to is actually the recommendation algorithms, you know, the TikTok for you algorithms, they're putting much more of a thumb on a scale than anybody can realize or quantify or see or even research because APIs aren't open. And maybe that's the thing we need. Like, maybe that's where we should point our regulatory effort, is saying, you need more competition there. Because otherwise, you start to get into this really dicey space where you're regulating the content itself, which is what Brendan Carr is trying to do. And I, I just think, no matter if you're super conservative or super liberal, that's

• 28:28Speaker 03

  too dangerous. The government should not have that power.

• 28:33Kevin Roose?

  Well, on that cheery note, Neelai.

• 28:35Speaker 03

  Look, I'm just telling you, the empowering thing, whenever you see a government regulator be like, we should do some speech regulations, just say they're bad. It's great. It's like the most American thing you can do, to look at the speech police and say, no, leave. And it feels good. And there's just, I promise you, I promise all listeners, there's something deeply empowering about that, that you can express at almost every turn of your life.

• 28:56Casey Newton?

  Yeah. All right, we'll give it a shot.

• 28:58Kevin Roose?

  All right, thanks, Neelai.

• 28:59Speaker 03

  Thanks, Seth.

• 29:00Casey Newton?

  Neelai, this was great. Thank you so much.

• 29:05Kevin Roose?

  When we come back, we've got a doctor's appointment. We'll talk to one of the authors of a new study showing how effective ChatGPT can be in diagnosing disease.

• 29:13Casey Newton?

  How much is the copay?

• 29:15Kevin Roose?

  I think it's 20 bucks a month.

• 29:21Speaker 00

  Support for this podcast comes from Dell Technologies. It's time to put AI to work. Is your business ready? The challenge isn't getting started with AI. It's scaling beyond pilots while maintaining control of your data, governance, infrastructure, and costs. Dell AI Factory provides the integrated foundation you need to scale AI securely and efficiently with one trusted partner, Dell, AI on your terms. From idea to impact. Learn more at dell.com/yourwaytoai. That's dell.com/yourwaytoai.

• 29:55Speaker 04

  Companies are pouring money into AI, but few leaders can explain how those AI investments work with the systems you already have. You know, the ones that actually run your business? Without Systems, you don't have to start over with AI. You can choose the projects where AI will have the most impact. And Outsystems gives you the platform to make it agentic on your terms with complete control every step of the way. So you can trust AI with the mission-critical systems that run your business. Own your agentic future. Learn more at outsystems.com/NYT. That's outsystems.com/NYT.

• 30:30Speaker 05

  AI spend is growing fast, but most businesses still can't see exactly where that money is going. Ramp gives businesses visibility and control across AI providers with real-time guardrails by team and model, helping ensure spend goes to the highest ROI areas. Stop guessing where your AI spend is going. Get started with Ramp. And if you're new to Ramp, you can get $250 off at ramp.com/hardfork. Go to ramp.com/hardfork. Ramp.com/hardfork.

• 31:05Kevin Roose?

  Well, Casey, it's time for your annual checkup.

• 31:07Casey Newton?

  Oh, my goodness. That's, you know what? You're joking, but I actually do have my annual checkup later today.

• 31:12Kevin Roose?

  Wait, really?

• 31:13Casey Newton?

  Yeah, I do.

• 31:14Kevin Roose?

  You're going to the doctor?

• 31:14Casey Newton?

  That's right. It's time to find out what's going on with this whole body, Kevin.

• 31:17Kevin Roose?

  Well, just from looking at you, I would say you're not getting enough vitamin D.

• 31:21Casey Newton?

  Well, I was recently diagnosed as handsome.

• 31:26Kevin Roose?

  I think you need to get a second opinion on that. But Casey, I want to talk today about AI and medicine. Because there was a thing that caught my attention recently. My colleague at the New York Times, Gina Colada, wrote a story about a study that came out a few weeks ago over at JAMA, the Journal of the American Medical Association, which showed that, on average, at least in this study, ChatGPT was better at diagnosing illnesses than doctors, even doctors who had access to ChatGPT.

• 31:54Casey Newton?

  And why that's so fascinating to me is for decades, people have been turning to WebMD to do something very similar and mostly, it seems, getting the wrong answer. Certainly, the people posting online said, oh, I typed these three symptoms into WebMD and, you know, it told me I was dying. That is not what appears to be happening with ChatGPT. ChatGPT is actually able to figure out what's going on with these folks.

• 32:15Kevin Roose?

  Yes. So we have so many questions about this study that we invited one of these study's authors, Dr. Adam Rodman, to join us. Dr. Rodman is an internist at Beth Israel Deaconess Medical Center in Massachusetts and the host of a medical history podcast called Bedside Rounds. Let's bring him in.

• 32:33Casey Newton?

  The doctor will see us now?

• 32:35Kevin Roose?

  The doctor will see us now.

• 32:43Kevin Roose?

  Adam Rodman, welcome to Heart of Fork. Thank you, guys, for having me. So let's talk about this study that you helped design. Tell us about the study and sort of what you were aiming to discover.

• 32:54Speaker 06

  Well, we were testing a simple hypothesis in a complicated way. That's what scientists do. We get too much into the details. But you know, one of the presuppositions in my field has been this, this idea that AI plus humans will always be better than AI alone, right? There's something essential about the humans. And a lot of health systems have rolled out these like secure versions of ChatGPT. Sometimes there's other language models with the idea that it'll make doctors better. So we basically tested that hypothesis out. We did a randomized controlled trial where we gave doctors, we gave attending physicians and residents. So those are physicians in training about, it was literally 50/50. And we either randomized them to go through these really complicated cases with ChatGPT or without. And we didn't

• 33:39Speaker 06

  just measure the diagnosis. We did, of course, measure whether they got the diagnosis, but we measure these really nuanced measures of how people think. So were you able to look for evidence that supported what you thought? Were you able to look for evidence that didn't support what you thought? Were you able to do these kind of basic cognitive tasks of a doctor?

• 33:56Kevin Roose?

  Hmm. What kind of information were you presenting to these doctors and these AI models? How detailed was it? Like, the kind of thing that you would get in a medical school exam? Or like, what kinds of problems were they being asked to solve?

• 34:10Casey Newton?

  Yeah, I want to see if we can solve some of them.

• 34:12Speaker 06

  Yeah, I can. Do you want me, do you want me to go through one of them for you?

• 34:15Casey Newton?

  Sure, let's hear one.

• 34:17Speaker 06

  Yeah, you, okay. Yeah. I'm excited to hear you guys attempt to go through a medical case as we go. Let me pull up-

• 34:24Kevin Roose?

  I think it's scurvy.

• 34:25Speaker 06

  Yeah, I don't think any of them were scurvy, unfortunately.

• 34:28Casey Newton?

  If there's one thing we've learned on podcasts is that people love a medical mystery.

• 34:32Speaker 06

  Yeah, this is basically like House MD, right?

• 34:34Casey Newton?

  Okay. Yeah, exactly.

• 34:35Speaker 06

  Here, here you go. A 76-year-old man comes to his doctor complaining of pain in his back and thighs for two weeks. He has no pain sitting or lying, but walking causes severe pain in his lower back, buttocks, and calves. He has a fever, he's tired. He was told by his referring cardiologist that his recent test results, that since his pain started, he has a new anemia, so his blood levels are low, and he has renal failure. And then a few days before the onset of the pain, he had coronary angioplasty. So he had a coronary catheterization of his heart and opened a vessel, and he got heparin during that. And then we go over like the lab values and stuff. This is not an easy case here. This is something that I think every doctor would know. Well, do you want to try to solve it first? Sorry, I should, I should have gone-

• 35:14Casey Newton?

  My first thought was chlamydia.

• 35:17Speaker 06

  Post-cardiologist-acquired chlamydia. Exactly. Kevin, any thoughts?

• 35:22Casey Newton?

  I'm still going with scurvy.

• 35:24Speaker 06

  Okay, great. What was the real answer, Adam? Cholesterol emboli syndrome, of course. No, I'm just kidding. It's actually a very hard diagnosis.

• 35:32Casey Newton?

  That was my second guess.

• 35:33Speaker 06

  Yeah, second guess. Yeah, I mean, and the point is, none of the cases are what are called zebras, right? They're none of the things that are off and on, house empty. They're all things that are tricky to figure out, but you will see and are real. The purpose wasn't really whether or not the humans got the diagnosis, but whether they went through those steps that are essential and generalizable to getting any diagnosis.

• 35:53Kevin Roose?

  So you give these little vignettes, these medical sort of mysteries, to the doctors in the study, and the doctors are given the use of GPT-4 to try to help them diagnose and figure out what's going on with this patient. Then you also had just GPT-4 by itself with no help from human doctors, try to analyze the same cases, and then you compared the analysis or the diagnosis from both groups. Is that right?

• 36:24Speaker 06

  Exactly. And we also let them use any other resources they wanted.

• 36:27Kevin Roose?

  And were these doctors in the study chosen because they had interest in using AI for diagnosis? Were they mostly more tech-savvy doctors? Were they people who had used this stuff before?

• 36:40Speaker 06

  No. So we did the classic trick to get a good subset of doctors, as we paid them. So these doctors were everyone. They had been in practice for a variety amount of time. Some people were experienced ChatGPT users. Those were the minority. Some people had never used it before. Most people fell in between. And what were the findings? Yeah, so the findings were not the most optimistic if you want to make people better, which is that the AI model did not improve human performance. So humans using the AI model did about as well as humans alone. And then, of course, the finding that I think the reason that I'm here and that everyone is angry at me is that the AI model itself drastically outperformed both groups.

• 37:22Kevin Roose?

  Yes, this was the headline, you know, of a lot of the coverage about it, was that the AI had beaten the doctors. Even if you gave the doctors access to AI, the AI by itself appears to do better at diagnosing these things. Now, obviously, we should make some caveats. This is a small study. We obviously would want more studies to sort of confirm this result. But this really stuck out to me because it seems like Sort of reading the study, what happened is that the, you know, the, basically the human doctors did not believe the AI could be as good or better than them at diagnosing. And so they would go in and sort of second guess what the AI had said and end up getting the diagnosis wrong as a result. Is that consistent with the findings?

• 38:01Speaker 06

  Yeah, I say there are two, well, maybe three reasons, two reasons. So one, some people, I mean, despite the basic training, some people didn't quite know how to use a language model to get the most use out of it. So probably some of that is training. Number two, though, when we look at the data, people liked it when the AI model said, oh, this is your, this is your idea. These are the things that agree with it. But when that AI model said, hey, man, you might be wrong. These things don't fit. They disregarded that.

• 38:28Casey Newton?

  Here is why that resonates with me. Have you ever been in an Uber and they have the Google Maps open and Google Maps is like, you might want to take this route? And they say, no, no, no, no, I actually know a better way. And the next thing you know, it takes you an extra 30 minutes to get you wherever you were going. Human, I firmly believe there is no Uber driver who can outsmart Google Maps. And we may be moving into a situation where most doctors cannot outsmart ChatGPT.

• 38:54Speaker 06

  Well, and that brings us to reason number three, the reason that people are angry at me, which, you know, I don't think it's the case now. It might be the case with O1, and it's certainly going to be the case in the next one to two years. Like, maybe AI models are better at making diagnoses than human doctors. Like, I don't think that's the case with GPT-4 Turbo, which was the model that was used here. But it's gonna be true at some point, and we're quickly approaching that.

• 39:17Casey Newton?

  Yeah, and we should say, this study took place last year, right? So like, all of the models that doctors have access to are now almost, they are 12 months better than they were, you know, when you ran the study.

• 39:26Speaker 06

  This is the classic academic publishing lag. And of course, I'm talking about this trial now and doing really other cool stuff. But like, the models have continued to improve, especially in diagnostic domains. Like, they're, they're saturating our benchmarks, right? Everything that we can throw at them and like, this is what humans should accomplish. By the way, humans are like 45, 50%. The new models are like, well, just kidding, I'm 90%. So.

• 39:47Kevin Roose?

  Well, so I have a question about that, which is, you know, ChatGPT released this O1 preview model, which does better reasoning, that's what they tell us. And I have not been able to figure out any prompt that I, as a mere journalist, actually seem to have any need of it. As a doctor, are you already turning to this model for reasoning through difficult medical questions?

• 40:11Speaker 06

  Yes, yes, and I have a preprint that will come out in the next couple of days that shows how dramatic it is.

• 40:16Kevin Roose?

  Yeah, I mean, the reason that this study caught my eye and fascinated me so much is that I think it's possible to imagine that a version of this finding could be found in many different fields. It's not just going to be medicine where the AI is sort of, reaches a point where it is better than either the human practitioners in that field or the human practitioners using AI in that field. And I think when that point happens for many white collar sort of knowledge workers, there's this question of like, how do you, as the practitioner, react? Do you get defensive and say, oh, the AI has to be flawed, it couldn't possibly be better than me, I'm not going to use it?

• 40:58Kevin Roose?

  Do you rebel against the AI and say, we can't, you know, we, these things make things up, they don't always get the thing right? Or do you embrace it and try to get good at the technology and use it in your work? Is that scenario playing out among doctors? Do you see doctors who are really happy about these findings because they say, oh, man, we're going to be able to give patients such better care? Or do you think most of them are sort of reacting from a place of fear and confusion?

• 41:23Speaker 06

  So, yes, yes, and yes, different people are reacting differently. Obviously, the reason that I'm doing this work is I want better care for my patients. And again, it's, if, I like making diagnoses, I'm a huge nerd, I'm like the prototypical internist, I pace around my patient's room like a crazy person trying to figure out what's going on. But if this algorithm helps me take better care from them, I will give that up. Other people are resistant, like, to insult doctors a little bit, we're a profession that really prides ourselves on our cognitive abilities. It gives us a lot of societal power and power over our patients, and this is a A professional challenge to my field.

• 42:01Speaker 06

  I am a pain in the butt, so that's fine. I don't care about that. But there are a lot of people that do. Right now, I'm at the Macy Foundation Conference. It's all the top medical educators to try to figure out what AI means. For how we train the next generation who's going to be practicing medicine for 30 years. These are things that the field is fiercely debating and arguing about right now. I'm just happy we're having the conversation.

• 42:22Casey Newton?

  Well, I have to say, I mean, the results are fascinating. But I do find myself siding in some ways with doctors who might be exasperated with these findings. And the reason is, you know, Kevin, you and I say all the time, hey, don't bet your career on anything that a large language model is telling you. These things do hallucinate. They make up facts all of the time. You and I don't really use them in our work in the context of we look up a quick fact and just drop it into our story. We actually are always going to second guess the LLM. We're always going to try to find a second source before we're like, okay, we actually feel like we can trust this piece of information. And Adam, in your study, basically what you found is that people who did that,

• 43:03Casey Newton?

  which we've been advocating for as the best practice, were worse at diagnosing diseases.

• 43:09Speaker 06

  I know. It's really so I, to be clear, I was shocked at the results. My hypothesis going in was that people using it would be the best. So I am surprised by this. So in the psychological literature on diagnosis, it kind of makes sense. Doctor, it's not doctors, humans are resistant to things that disagree with them. And we have all these heuristics and cognitive shortcuts that we take. So it's not surprising to me that what people did was they anchored on what they thought and what the first things that they thought and they were resistant to something that was giving them a second opinion. Maybe that's something that's actually optimistic because we can align models or try to figure out how to present that information to make humans better. That is what I am trying to do. And I think all the short term uses, like, let's, let's be clear. Like, if the headline is, doctors are over, chat GPT is good. No, absolutely not.

• 43:54Speaker 06

  There's a million things we do. This is just one part of it. And they're not capable of operating without us. I'm not discouraged by this. I'm still working to figure out ways we can use these technologies to make better care of our patients.

• 44:05Casey Newton?

  Yeah, I mean, that's the question. I'm curious.

• 44:07Kevin Roose?

  About is like, what can the medical field do? I mean, I'm imagining a future where patients have access to this stuff. And maybe before you go into the doctor to get your your hip pain checked out, you do sort of an exhaustive prompting exercise with the the model and say, hey, what is this? And then you sort of bring the the readout from the AI into your doctor and say, hey, could you give me this medicine and this medicine? And I need this operation because, you know, and the doctor might say, well, um, you know, let's do some tests and you're saying, I don't need to, the AI already told me.

• 44:36Speaker 06

  That's already happening. That's already happening. I mean, there was a Kaiser Family Foundation's survey on how many patients are putting their information, but it's already happened in my life when people will even put their, to be clear, these things are not HIPAA compliant. Please don't put any of your personal health information. But people are doing it.

• 44:53Kevin Roose?

  Elon Musk told me I should be uploading all my MRIs to Grok. Are you saying he was not correct?

• 44:59Speaker 06

  Well, it depends on if you want someone else to own all your MRI images. So, yeah, keep that in mind.

• 45:04Kevin Roose?

  Yeah, I uploaded it and told me I had the woke mind virus. So that was weird.

• 45:09Speaker 06

  They don't work very well.

• 45:13Kevin Roose?

  Right.

• 45:14Speaker 06

  But, yeah, people are doing it already. I've had patients who do it. This is not a future. Now, does it work that well? Sometimes, but not consistently. And they're very, you know, you have to prompt them, right? But how far are we from somebody selling a commercial tool? That's a dock in a box that works pretty well.

• 45:29Casey Newton?

  I mean, that is the actual implication of your study is that you are better off just asking ChatGPT and not your doctor.

• 45:37Speaker 06

  I, that's not my conclusion from the study. If that's what you want to take from it.

• 45:43Casey Newton?

  Well, you're good. The conclusion is basically one in four doctors were not able to successfully diagnose this. But in 92% of cases, ChatGPT did. Like, if I had to choose one of those two things, I'd probably choose ChatGPT because it also does other things for me, too.

• 45:58Speaker 06

  I would say that the difference is that the people who put the case together, like the information, if you want to think about the prompts, were expert clinicians. We organize it in such a way, like you can imagine, um, you've, you've, I'm assuming, I don't want to talk about your past medical histories, but I've had problems and we don't, humans don't always describe things the right way. We don't know how good ChatGPT is about getting that information out of us. I think it's going to happen, but I, I don't think that ChatGPT can do that now.

• 46:24Kevin Roose?

  I'm curious if these AI tools do become part of the clinical model in hospitals all over the place, it sounds like they are going to. What is it going to mean to be a good doctor in a world where AI is better at diagnosing than you are?

• 46:41Speaker 06

  So I'll give you the, there's the darkest timeline, but we'll go with the optimistic timeline.

• 46:46Kevin Roose?

  Give us both.

• 46:47Speaker 06

  Okay, well, let's go, let's go with the optimistic view first, because this is what I'm hoping. And inspired by, oh, I'm a huge nerd, this should not be a shocker, but inspired by you a little bit, Kevin, it's the Star Trek computer, right? So you have a computer system that's listening in at all times and is saying, hey, hey, Adam, you might be showing some unconscious bias here. Adam, I think you should ask if this person like makes their own snuff because eosinophilic pneumonia is on their differential, like something that's listening in, cueing me to be better, trying to make me a better human, but also listening to the patient and getting more information from the patient. A computer system like that is something that makes the medical encounter more human, which I hope is what we want. You want the darkest timeline next?

• 47:27Kevin Roose?

  Yes, please.

• 47:28Speaker 06

  So I don't know if you guys know this, so AI technologies are already being rapidly spread out in clinical care. They're listening to doctors' encounters with their patients and writing notes. They're writing the first drafts of like, when you talk to your doctor on a portal of those messages. I just wrote a piece in the New England Journal of Medicine where I originally called it Language Models and the Insidification of the Electronic Medical Record. It turns out the New England Journal of Medicine doesn't consider that an academic term, so they changed the degradation. But like, what we're, what we're seeing so far is not the model that I, that I am advocating for and what I'm researching and pushing for, but a system that's obsessed with efficiency isn't really worried about some of the downstream effects on what this means for our relationships and is just going to like, yeah, you'll get these more efficient tools, so you'll see twice the number of patients in a day.

• 48:13Speaker 06

  We'll just put this AI text in the chart so we can bill off of it. And a system that might use these powerful, efficient tools to like squelch out the tiny bit of humanity that remains in medicine. So that, to me, is the darkest timeline and what I want to avoid. I don't think there's, you have to engage with this technology. It's going to change every single like white collar field, we're like the ultimate white collar field. It's going to change our field. And I see a way that we end up like with Dr. Crusher on the Enterprise, but I also see a way that we end up, I don't know, it was a dystopia, like I'd say Blade Runner, but I don't think there are any doctors in Blade Runner, so this analogy is going to fall apart.

• 48:47Casey Newton?

  You know, I mean, to me, like, an optimistic loss on all of this is the upside in making this kind of care much more accessible, right? Like, if all of a sudden, I can just check my basic symptoms with ChatGPT, maybe that does provide me some benefit. Now, obviously, a lot of people have been doing this for decades with WebMD, and there are, you know, sort of a lot of jokes about that. A lot of people are sort of quick to use WebMD to assume that they have the very worst condition, and also, like, constantly seeking medical care can, like, create its own set of problems. But if you're just sort of the median person, I can also just imagine checking in with my virtual doctor a couple of times a month and get some tips about how to live a healthier life.

• 49:26Speaker 06

  Oh, so, yeah, absolutely. We're not there yet, but I think that's the way things are going. And to be clear, the reality is terrible. Like, how long does it take you guys to see your primary care doctor? I'm a doctor, and it takes me forever. So, maybe we'll have a system that can do those basic things, but also recognize when it needs to step you up, like, triage you appropriately. And maybe you'll have a system where, instead of referring you to a specialist, your PCP will be able to work with that system to, you know, answer something that you would have needed a specialist before, or a system that says, hey, you don't need to go through the referral system, go straight to the orthopedic surgeon. So I think there's a lot of hope, and I acknowledge, like, the baseline is terrible. Our medical system isn't really serving our patients. And if we're thoughtful about this, like, it's okay if my power is eroded. Like, we'll get better care for everybody if we're thoughtful about it, which, if you've looked at the history of how medicine

• 50:11Speaker 06

  has happened in this country, that's not always the case.

• 50:14Casey Newton?

  So, Adam, I'm curious. You're a doctor. You trained for many years to become a doctor. You amassed a lot of knowledge that, you know, has made you good at that job. What is your emotional reaction to the findings of your own study?

• 50:28Speaker 06

  Um, yeah, I mean, I'm, I am, it's a lot of emotions, right? I'm both excited and I'm freaked out. I'm, uh, I'm not the typical doctor. I am a historian. I deeply care about how people think. I feel like I'm on the edge of something new, which is exciting. But to me, like, I love talking to people. I love meeting new people. But one of the things that I love is the intellectual part of my work. That's what makes it, I don't love sitting down and writing billing codes and saying, is this a level two? And I hate that part of my job. But the part where I get to, like, talk to somebody and figure out what's going on with them so I can make them better, that's my favorite part. But at the end of the day, I'm here for my patients. So that, like, I'm conflicted, but it's clear to me what the right thing to do is, which is do the right thing for the patient, even if it means giving up something that is dear to me.

• 51:13Kevin Roose?

  Yeah, I mean, that strikes me as a good model for people in all kinds of industries as the AIs do get better at doing our jobs. It seems like the North Star should be like, what is the actual work that I am performing? And if an AI can do that better than I can, then maybe that's better for the world.

• 51:32Casey Newton?

  Well, you know, there is another approach that I wondered if you consider, which is to say that, you know, essentially, these chatbots were trained on a bunch of work that real doctors did. Those doctors are not being compensated. The primary effect of these chatbots being in the world is that the salary of a doctor could go way down. Has there been any talk among doctors of saying, let's actually get together and stop these things from draining all of the money out of our industry?

• 51:56Speaker 06

  Oh, yeah, we're def- I mean, yes, those are- I think that in the grand scheme of things that doctors are worried about threats to their career, this is low right now. These are all theoretical talks, but I suspect we're going to hear more of that. It's weird, like, when you're in a profession, uh, I- I- wait, am I allowed to swear on Hard Fork?

• 52:15Casey Newton?

  Yes.

• 52:16Speaker 06

  Yeah, I actually believe in that old bullshit about, like, the doctor-patient relationship being the most important thing above all else. I believe that. That's why I'm such a pain in the ass. So, like, if the- if this thing can do a better job than me at making my patients' life better, to me, it seems like, regardless of those- of those guild issues, right? It seems to me that's what the right thing to do is.

• 52:39Casey Newton?

  Yeah, it'd just be interesting if we live in a world where the actors have successfully prevented movie studios from replacing them with AI, but the doctors are like, well, I guess that's fine.

• 52:48Speaker 06

  That might happen.

• 52:51Kevin Roose?

  So, after having done this study and continuing to do work in this area of AI and medicine, do you feel more optimistic about the future of medicine? Or do you feel like we're headed into this kind of dark timeline where AI is just making all the decisions and we sort of suck the humanity out of the healthcare system?

• 53:13Speaker 06

  I see the market forces at bear here and my worry and the way that I see things being rolled out now is that we're veering not directly towards the darkest timeline, but that we're heading in that direction. And I think that we need to be really thoughtful. And the we is not just doctors, like patients need to be, have a voice in this also. This is ultimately who this is about, about what type of health system we want and how we want these technologies to be used. But I'm actually worried about us heading, like, the current timeline's pretty dark, guys. You get five minutes, 10 minutes with your doctor and they don't look at you and they type on the computer. Like, that's not good. So with medical errors, like up to 800,000 Americans are killed or seriously injured

• 53:54Speaker 06

  each year because of medical errors. Every one out of $5 that Americans like make go into the healthcare system. So this darkest timeline thing isn't that far away. And I, well, I'm a natural pessimist, but I'm trying. I'm like Don Quixote. I'm trying to like, go for the good timeline, even though it probably won't work.

• 54:14Kevin Roose?

  If any young people are listening to this who may have been interested in becoming a doctor or entering the healthcare profession, what would you advise them? Should they not become doctors because AI is going to take that job?

• 54:25Speaker 06

  Well, the problem is, what are you going to suggest that they do instead? Like, if we're talking about technologies that can do this. At this conference, they played something on Notebook LM with a fantastic podcast host. So I don't know. I don't. So I think that we're talking about tasks of doctors that might be automated and it's going to be working together for a while. And we're not talking about the job as a whole. And fundamentally, it's still a job about human connection and making people better. And if that is what you want, I would do that. Also, surgeons and proceduralists are not going anywhere. So I wouldn't dissuade somebody from medicine, but they should know that's what they're going into it for. And it's not going to be like Dr. House. I actually have never seen House. I always just use this example despite having never seen the

• 55:08Speaker 06

  show, so I'm phony. But it's just, it's not going to be that cognitive part. It's going to be something different. And that's scary because I can't predict. And I like, I would love, I mean, medical students ask me this, and I don't have an answer for them.

• 55:22Kevin Roose?

  Well, it's a fascinating conversation, Doctor, but I will be seeking a second opinion, actually. I think it's just important.

• 55:28Speaker 06

  You should ask ChatGPT.

• 55:32Kevin Roose?

  Thank you so much.

• 55:33Speaker 06

  Thanks, guys. Well, that was fun.

• 55:34Casey Newton?

  Thank you, Adam. I learned a lot.

• 55:40Casey Newton?

  When we come back, crime doesn't pay, but it does play on the Hard Fork Podcast. I see what you did there.

• 55:56Speaker 00

  Support for this podcast comes from Dell Technologies. It's time to put AI to work. Is your business ready? The challenge isn't getting started with AI. It's scaling beyond pilots while maintaining control of your data, governance, infrastructure, and costs. Dell AI Factory provides the integrated foundation you need to scale AI securely and efficiently with one trusted partner. Dell, AI on your terms. From idea to impact. Learn more at dell.com/yourwaytoAI. That's dell.com/yourwaytoAI.

• 56:30Speaker 04

  Companies are pouring money into AI, but few leaders can explain how those AI investments work with the systems you already have. You know, the ones that actually run your business? With OutSystems, you don't have to start over with AI. You can choose the projects where AI will have the most impact. And OutSystems gives you the platform to make it agentic on your terms with complete control every step of the way. So you can trust AI with the mission critical systems that run your business. Own your agentic future. Learn more at outsystems.com/NYT. That's outsystems.com/NYT.

• 57:05Speaker 05

  AI spend is growing fast, but most businesses still can't see exactly where that money is going. Ramp gives businesses visibility and control across AI providers with realtime guardrails by team and model, helping ensure spend goes to the highest ROI areas. Stop guessing where your AI spend is going. Get started with Ramp. And if you're new to Ramp, you can get $250 off at ramp.com/hardfork. Go to ramp.com/hardfork. Ramp.com/hardfork.

• 57:39Casey Newton?

  Well, Kevin, in the criminal justice system, the people are represented by two separate, yet equally important groups: the police, who investigate crime, and the media, who turn those crimes into podcasts.

• 57:50Casey Newton?

  And from time to time here on Hard Fork, we like to survey the landscape of crime and punishment for a segment that we call Hard Fork Crimes Division.

• 58:05Casey Newton?

  All right, now, in this segment, we seek justice. I think it's fair to say we will not be solving any crimes, but we will describe them.

• 58:13Kevin Roose?

  Yes.

• 58:14Casey Newton?

  Or certainly, we will describe what has been alleged.

• 58:16Kevin Roose?

  We, I was just saying, we have not yet solved a crime, but it's not out of the realm of possibility for the future.

• 58:22Casey Newton?

  Not at all. Not at all. We are always gathering evidence. And perhaps we should turn to our first case, Kevin.

• 58:27Kevin Roose?

  Yes.

• 58:28Casey Newton?

  Let me crack open this case file. The FBI searches the home of the founder of the Polymarket betting website. Did you see this one?

• 58:37Kevin Roose?

  I did. This was juicy.

• 58:39Casey Newton?

  So Polymarket founder Shane Copeland had his home searched by the FBI last week as part of a criminal investigation into whether Copeland was running Polymarket as, quote, "an unlicensed commodities exchange", which is apparently illegal. And they seized Copeland's electronic devices, including a phone.

• 58:55Kevin Roose?

  Yeah, that's not a good thing when that happens to you.

• 58:58Casey Newton?

  Now, Kevin, after Shane Copeland's phone was seized, he posted the following on X. "New phone, who dis?" So Kevin, remind us who this Shane Copeland character is.

• 59:10Kevin Roose?

  So this is the young founder of Polymarket, which is the sort of leading crypto prediction betting market platform. It rose to prominence during the election, where people wagered millions of dollars on who was going to win the election. And as my colleague David Yaffey Bellini told us on the show a few weeks ago, it was sort of nominally illegal in the US, but lots of Americans were using it anyway through VPNs and things like that. And it was sort of an open secret that it had this large base of customers in the US despite not technically being allowed here.

• 59:47Casey Newton?

  Yeah, so I think that the FBI has some questions about that. But a Polymarket spokesman said, why not? That the raid was, quote, "obvious political retribution by the outgoing administration."

• 59:59Kevin Roose?

  Yeah, the theory here, at least the one that's being sort of advocated by Polymarket's fans and defenders, is that, you know, the Biden Justice Department and FBI were so mad about the election and the fact that people on Polymarket had predicted that Trump would win, that they, I don't know, went after the company on some bogus charges.

• 1:00:18Casey Newton?

  And here is why I don't think that's true. Could you imagine explaining Polymarket to Joe Biden? Like, Mr. President, it's a prediction market. People bet cryptocurrency on the outcomes of various events. Not in the United States, but they would VPN into it. By the time you've gone to VPN, Joe Biden has truly fallen asleep.

• 1:00:37Kevin Roose?

  I don't think so. I bet Joe Biden has used a VPN.

• 1:00:40Casey Newton?

  You think so, Joe? What, like, watch Netflix movies that are unavailable in the United States?

• 1:00:43Kevin Roose?

  BBC mysteries.

• 1:00:47Kevin Roose?

  So what do we know about why they are being investigated here?

• 1:00:51Casey Newton?

  Well, because if it is true that large numbers of Americans are illegally betting on elections by using VPNs, that could be a violation of the law. You know, D Y B told us that people were openly describing how to get around the ban on US bettors in the poly market Discord.

• 1:01:11Kevin Roose?

  Yes.

• 1:01:11Casey Newton?

  So I think at the very least, the FBI is going to say, you need to tighten this up a little bit and make it a little bit harder for Americans to use this service.

• 1:01:19Kevin Roose?

  Yes, and some reports have said that this investigation predates the election. This was in process long before. It's also not Polly Market's first sort of run in with the law. They previously settled with the CFTC, the Commodity Futures Trading Commission, in 2022 and paid a fine as part of that settlement. But this is something new. This is bigger. And I would say if you are a poly market fan in the US, you probably should stop doing that.

• 1:01:47Casey Newton?

  Can I tell you how I think this one resolves?

• 1:01:49Kevin Roose?

  How?

• 1:01:49Casey Newton?

  Shane Copeland running the Federal Reserve. Stay tuned. It's gonna be a wild 2025.

• 1:01:56Kevin Roose?

  Yes.

• 1:01:56Casey Newton?

  Let's open case number two, Kevin.

• 1:01:59Kevin Roose?

  What do we have?

• 1:02:00Casey Newton?

  Well, Kevin, Razzlecon, crypto's most embarrassing rapper, some say, is going to prison. Remember Razzlecon?

• 1:02:07Kevin Roose?

  I sure do.

• 1:02:08Casey Newton?

  Heather Razzlecon Morgan, who's a former blogger at Forbes and part of the Forbes to Prison Pipeline.

• 1:02:15Casey Newton?

  And creator of cringey crypto tinged rap videos was sentenced to 18 months in federal prison this week after pleading guilty last year to helping her husband, Ilya Dutch Lichtenstein, launder 120,000 Bitcoin he stole by hacking the crypto exchange Bitfinex back in 2016. Do you know how much 120,000 Bitcoin were worth in 2016?

• 1:02:39Kevin Roose?

  Uh, 2016, let's see, probably not as much as they are today.

• 1:02:44Casey Newton?

  They were worth $71 million back then. They are worth $11 billion today.

• 1:02:50Kevin Roose?

  That's quite a haul.

• 1:02:50Casey Newton?

  Old Dutch and Razal Khan really almost got away with it. They would be living large.

• 1:02:55Kevin Roose?

  I was obsessed with this story, uh, when it, when it came out, when they got arrested, because it was sort of like out of like a very, uh, pulpy, like spy novel. Like they had fake passports and they were this like sort of Bitcoin Bonnie and Clyde and they were just like these cringey millennials who were trying to get famous on the internet, but also, uh, stealing a bunch of Bitcoin to make themselves very rich. It was just, I, I, my, my, our friend, uh, Nick Bilton has a documentary coming out about this, uh, this case on Netflix that I'm very excited to watch because I'm truly obsessed.

• 1:03:30Casey Newton?

  Shall we, shall we hear a little bit of Razal Khan's work?

• 1:03:32Kevin Roose?

  Let's do it.

• 1:03:33Casey Newton?

  Yeah. If we could hear a clip, please.

• 1:03:35Speaker 07

  Razal Khan, the Versace better win. Come real far but don't know where I'm heading. Motherfucking crocodile of Wall Street.

• 1:03:42Casey Newton?

  foul language.

• 1:03:43Speaker 07

  Dipper on my fingers and boots on my feet. Always be a goat, not a goddamn sheep. Email me, fuck your message at the beep.

• 1:03:56Casey Newton?

  See, now, I, I, to me, this just goes to show how much the culture has changed because there was a time when people would have looked at what Razal Khan did and simply said, she's being fergalicious. But in sort of the woke moment that we're in now, stealing 120,000 Bitcoin gets you a year and a half in jail.

• 1:04:14Kevin Roose?

  Yeah.

• 1:04:15Casey Newton?

  Really sad.

• 1:04:16Kevin Roose?

  Do you think that like her rap career will be an asset in jail?

• 1:04:20Casey Newton?

  Absolutely.

• 1:04:21Kevin Roose?

  To her reputation?

• 1:04:22Casey Newton?

  I would not be surprised if Razal Khan is the most popular person in the prison that she's in and if it fuels the next phase of her journey. And in fact, she posted on X that she will quote, "Soon be telling my story, sharing my thoughts and telling you more about the creative and other endeavors I've been working on." So, you know, I don't know what that means, but I will say I would love to see a Razal Khan jukebox musical. Tell the story of RazzleCon in her own words through her own music.

• 1:04:46Kevin Roose?

  Yes, and I should say I look forward to RazzleCon's appointment to head the Securities and Exchange Commission.

• 1:04:55Kevin Roose?

  Next crime. What do we have?

• 1:04:58Casey Newton?

  Kevin, Gary Wang, a top FTX executive, has been given no prison time.

• 1:05:03Kevin Roose?

  What did he do?

• 1:05:04Casey Newton?

  Well, Gary Wang, Kevin, was the last of the legal cases against FTX. You might remember some of FTX's more famous co-founders, such as Sam Bankman-Fried, who was sentenced to 25 years in prison for his role in FTX fraud, or Caroline Ellison, who was sentenced to two years in prison. Most recently, Ryan Salami was sentenced to seven and a half years in prison.

• 1:05:27Kevin Roose?

  It's Salem. David Appi Bellini literally has to put a pronunciation guide in his stories for this name, because everyone calls him Ryan Salami, but it's Salem.

• 1:05:37Casey Newton?

  Do you know what they called the case against Ryan Salem, Kevin?

• 1:05:40Kevin Roose?

  I think I know where you're going with this.

• 1:05:41Casey Newton?

  The Salem Rich Trials!

• 1:05:43Kevin Roose?

  Yes. I knew that was gonna happen.

• 1:05:50Kevin Roose?

  More like the Salem Rich Trials, am I right?

• 1:05:53Casey Newton?

  That's a good, better one.

• 1:05:57Kevin Roose?

  Wait, I gotta snort out of you for that?

• 1:05:59Casey Newton?

  That was good. That was good. So, anyway, so that leaves Gary Wang, the fourth member of the crew here. Actually, that's not even true. There's another guy, Nishad Singh, who was sentenced to time served. So Gary Wang was the last of these cases to be, you know, resolved. And it was resolved this week, and he was given no prison time. And the reason is, he snitched so hard on SBF that the government basically gave him a standing ovation. During the sentencing hearing, one prosecutor said that Wang was, quote, "the easiest cooperator they've worked with and provided essential information to them." So he basically got the best snitch award, and it kept him out of jail.

• 1:06:37Kevin Roose?

  Which is a good reminder that cooperating with the government in a fraud investigation can have benefits.

• 1:06:43Casey Newton?

  Now, Kevin, the FTX legal saga has really, you know, taken place from the start of this podcast, you know, and now it's sort of wrapping up. So do you have any sort of feelings of nostalgia or other reminiscences from two years of FTX?

• 1:06:56Kevin Roose?

  You know, I have been just very interested in this whole saga, not just because I think it was a big deal in the world of crypto, but because it has had all of these strange ripple effects, including... I was talking with someone this week about this, but the investment that SBF made in Anthropic, the AI company, has essentially paid back all of the investors who would have lost money on the FTX fraud, because that stake has turned out to be worth a ton of money. And so even though Sam Bankman-Fried was a fraudster and is now serving time in prison, it turns out he was actually a pretty good tech investor.

• 1:07:34Casey Newton?

  If he gets out of prison and you just like run into him and he's like, you know where you should like put your money? Like, would you listen to him?

• 1:07:40Kevin Roose?

  Yes, honestly, I would.

• 1:07:45Casey Newton?

  You know what, I, you know, I might too. I believe in second chances for people.

• 1:07:48Kevin Roose?

  And Sam, if you're listening, I would love your investment advice. I could really use some updates to my portfolio.

• 1:07:53Casey Newton?

  Sam, if you're listening, you're not supposed to have a cellphone in there, so be careful.

• 1:07:56Kevin Roose?

  You don't think you can get podcasts in prison?

• 1:07:58Casey Newton?

  That'd be the worst part about going to jail.

• 1:08:02Casey Newton?

  Well, Kevin, we have one more case to look at. A phone network has employed an AI grandmother to waste scammers' time with meandering conversations.

• 1:08:11Kevin Roose?

  Yes, as you know, there are now these scammers who will call people using an AI voice pretending to be, you know, a long-lost cousin or their grandmother or something and just try to steal money from them by impersonating someone. But this is a story that comes to us from the UK, where the largest mobile phone operator in the UK, O2, has created a new AI system called Daisy to trick scammers into thinking that they are talking to a real person who basically has been given the goal of just rambling and keeping them on the line for as long as possible. So wasting the scammers' time. I'm sure you've seen there are all these YouTube videos now

• 1:08:54Kevin Roose?

  of people whose whole shtick is that they take scam phone calls and then they try to scam the scammers. But that is labor-intensive. And so now, O2 has come along and said, "We can actually build an AI that just wastes the scammers' time for you." And I think that's a great development.

• 1:09:08Casey Newton?

  I agree. I've read that they've sort of designed it to keep the scammers on the phone for as long as possible. But they're also trying to learn what tricks and techniques the scammers are using so that they can, you know, share that with, you know, maybe their customers, maybe the police, and help, you know, prevent people from falling for these things. O2 said that Daisy has managed to keep some people on the phone for up to 40 minutes.

• 1:09:34Kevin Roose?

  I'll just say it. If an AI voice is keeping you on the phone for 40 minutes, you're a bad scammer.

• 1:09:40Casey Newton?

  Terrible scammer.

• 1:09:41Kevin Roose?

  You're bad at your job.

• 1:09:42Casey Newton?

  Yeah.

• 1:09:42Kevin Roose?

  Because you can tell instantly when it's an AI on the other end of the line.

• 1:09:46Casey Newton?

  Yeah.

• 1:09:46Kevin Roose?

  At least I think I can.

• 1:09:48Casey Newton?

  Well, there's usually like some sort of delay.

• 1:09:49Kevin Roose?

  Yes.

• 1:09:50Casey Newton?

  Right? Like, and presumably that's going to disappear, but for now, I guess I feel somewhat confident. Now, I will say that consumers cannot use Daisy, but what O2 did was add it to the list of what they call easy target numbers used by scammers. So sort of sharing it around and saying, hey, you know, this Daisy is a really easy mark. So that's cool. But I will say it does make this feel a little bit more stunt-y to me. Although I guess as I think about it, I'm not exactly sure how consumers would be able to, I don't know, flip a button to get, you know, Daisy to answer their scam.

• 1:10:23Kevin Roose?

  Oh, I think this could be, because you know how Apple or like other mobile devices can now sort of say scam likely?

• 1:10:28Casey Newton?

  Yeah.

• 1:10:29Kevin Roose?

  When, when someone calls you from an unknown number? Yeah, you could just press a button and it would put Daisy on the line and it could just waste their time. I think we should deploy this.

• 1:10:36Casey Newton?

  Wait, that's actually genius. Like, I want to do this.

• 1:10:39Kevin Roose?

  Yes. Do you, do you like these sort of vigilante schemes to take back the power?

• 1:10:44Casey Newton?

  You know, I, I mean, look, there is always pleasure in seeing a justice done.

• 1:10:50Kevin Roose?

  Yes.

• 1:10:50Casey Newton?

  And injustice being righted. You know, I have to say, I have enjoyed YouTube videos of like porch pirates being apprehended.

• 1:10:58Kevin Roose?

  The glitter bombs.

• 1:10:58Casey Newton?

  The glitter bombs. I find that very satisfying.

• 1:11:01Kevin Roose?

  This is when like you disguise like a, something as a package. Someone steals it, they open it up and it sprays glitter everywhere and, you know, sets off an alarm and sets off like horrible smelling stuff.

• 1:11:14Casey Newton?

  Yeah.

• 1:11:14Kevin Roose?

  This is a very popular genre of YouTube video.

• 1:11:16Casey Newton?

  Yeah, most people do not have, very often, an experience of justice.

• 1:11:21Kevin Roose?

  Yes.

• 1:11:21Casey Newton?

  It's like you see injustice everywhere, but like the, the moment that you actually see like a right, a wrong being righted is like transcendent. I remember one time I was on the freeway and everyone was like trying to merge onto a different freeway. And so you're just sitting in bumper to bumper traffic and you're going forward at one inch an hour. And somebody gets impatient and they pull onto the shoulder so they can just get around everybody because I guess they had somewhere to be. And about one second after the person pulled onto the shoulder, I saw siren lights go up and a police officer just went and you know, pulled that person over and, you know, got them in trouble. And that was like my greatest experience of justice.

• 1:11:55Kevin Roose?

  Yeah.

• 1:11:56Casey Newton?

  And that happened 20 years ago when I think about it all the time. I'm like, I'm so glad that happened.

• 1:12:02Kevin Roose?

  Anyway, thanks, Daisy.

• 1:12:04Casey Newton?

  Thanks, Daisy.

• 1:12:05Kevin Roose?

  And the sooner I can have you on my phone to deter the scammers, the happier I'll be.

• 1:12:09Casey Newton?

  And that's the Hard Fork Crimes Division. Case closed.

• 1:12:11Kevin Roose?

  Yeah.

• 1:12:24Speaker 00

  Support for this podcast comes from Dell Technologies. It's time to put AI to work. Is your business ready? The challenge isn't getting started with AI. It's scaling beyond pilots while maintaining control of your data, governance, infrastructure, and costs. Dell AI Factory provides the integrated foundation you need to scale AI securely and efficiently with one trusted partner. Dell, AI on your terms. From idea to impact. Learn more at dell.com/yourwaytoAI. That's dell.com/yourwaytoAI.

• 1:12:58Speaker 04

  Building and scaling AI today can feel pretty chaotic, mostly because teams are juggling too many fragmented tools at once. That's where OutSystems comes in. OutSystems gives you one unified Agentic Systems platform to manage your entire portfolio of agents, apps, and workflows. You get to innovate at the speed of AI, but with the security, governance, and control your enterprise actually needs. And the results are already there. A major manufacturer used OutSystems to build a multi-agent system that analyzes blueprints, and a global bank cut their development time by 60%. Own your Agentic future today at outsystems.com/NYT.

• 1:13:34Speaker 05

  AI spend is growing fast, but most businesses still can't see exactly where that money is going. Ramp gives businesses visibility and control across AI providers with real-time guardrails by team and model, helping ensure spend goes to the highest ROI areas. Stop guessing where your AI spend is going. Get started with Ramp. And if you're new to Ramp, you can get $250 off at ramp.com/hardfork. Go to ramp.com/hardfork. Ramp.com/hardfork.

• 1:14:08Casey Newton?

  Before we go, we have a special request. If you can, we would really appreciate if you filled out a quick survey. You can find the survey at nytimes.com/hardforksurvey. Your answers will not be published in any way. They will just sort of help us make the best show we possibly can and understand more about who listens to the show in the first place. Again, you can find the survey at nytimes.com/hardforksurvey. We'll also drop the link in show notes.

• 1:14:41Casey Newton?

  Hard Fork is produced by Rachel Cohn and Whitney Jones. We're edited by Jen Poyant. This episode was fact-checked by Ina Alvarado. Today's show was engineered by Alyssa Moxley. Original music by Marian Lozano, Diane Wong, Leah Shaw-Damren, and Dan Powell. Our audience editor is Nell Galogly. Video production by Ryan Manning and Chris Schott. You can watch this whole episode on YouTube at youtube.com/hardfork. Special thanks to Paula Schuman, Puyweng Tam, Dahlia Hadad, and Jeffrey Miranda. You can email us at [email protected] with whatever disease ChatGPT just told you that you have.

• 1:15:30Speaker 08

  75% of women who seek menopause care don't receive it. The lying awake at 3 a.m. wondering what's happening to your body ends now. Midi is virtual care built for women in perimenopause and menopause with clinicians who specialize in women's health. Plus, it's covered by most major insurances. Getting started is easy. You can book your first visit with a clinician right away. Midlife needs Midi. Visit joinmidi.com to book your first visit today. That's join m i d i dot com. Insurance coverage varies. Check with your plan for coverage.