I mean, what do you think is going on here?

here?I actually think this is fascinating.

There's a lot going on here.

So s- f- for some context, the FTC has, for decades, been a semi-independent body that was bipartisan.

It was, it's within the executive branch, but it was made up of both Democrats and Republicans traditionally, and so you kind of had to get, uh, both sides on, on board for the commissioners to, to decide on something.

And, uh, even if, even if one side won, the other side got a say.

That has all changed under President Trump.

I can see, you know, the platforms have, have some, uh, plausible deniability in terms of, you know, how, how they approach political content in, in advertising.

I mean, what do you think is going on here?

here?I actually think this is fascinating.

There's a lot going on here.

So s- f- for some context, the FTC has, for decades, been a semi-independent body that was bipartisan.

It was, it's within the executive branch, but it was made up of both Democrats and Republicans traditionally, and so you kind of had to get, uh, both sides on, on board for the commissioners to, to decide on something, and, uh, even if, even if one side won, the other side got a say.

That has all changed under President Trump.

He has asserted that he has the sole authority over the FTC.

I can see, you know, the platforms have, have some, uh, plausible deniability in terms of, you know, how, how they approach political content in, in advertising.

about the White House level, but what about on a state level? Is there any version of this where the state of Virginia decides, actually, we are going to regulate AI to some degree? Or is that even possible, given how the internet's everywhere?

One of the things that that brings up, if I can digress a little bit, is that these risks that we're talking about right now, they fall under the umbrella of what's called AI safety.

It's basically like the risk that the AI systems themselves will go rogue and wreak havoc.

which, again, used to seem crazy.

Now it seems a little less crazy.

But there are a lot of other risks from AI, right? There's mass surveillance.

There's the privacy issues raised by AI.

Is that true? Where do we stand and what is being done internationally about AI safety?

Will, what were you thinking as you listened to that caller?

Yeah, I think that's a very valid point of view.

And again, I think there are really interesting philosophical debates you can have, really interesting semantic debates you can have about words like agency and intelligence and thinking.

And I love philosophical debates.

I was a philosophy major.

Semantics do matter.

But I also know that philosophical debates are really hard to resolve.

But slowing the pace of AI development to the point that it doesn't threaten humanity, but continuing to develop it fast enough that the U.S. stays ahead of China.

Um, Will, what were you thinking as you listened to that caller?

Yeah, I think that's a very valid point of view.

A- and again, I mean, I think there are really interesting philosophical debates you can have, really interesting semantic debates you can have about words like agency and intelligence and thinking.

Uh, and I love philosophical debates.

I was a philosophy major.

Semantics do matter.

But I also know that philosophical debates are really hard to resolve.

Is it clear to you, Will, what kinds of safety practices he's empowering these embedded evaluators to monitor? What the standards or the metrics would be?

W- what though, alongside all of The Atlantic's body of reporting about the threats to democracy, where does the threat to humanity from AI rank?

Well, it's, it feels like it's changing by the week.

Um, it's, uh, you know...

I think the people who are really in deep in the AI safety community believe that it's number one, that it's the clear number one.

Um, I think that there are reasons to think that there are other crises in the world that are at least commensurate.

Uh, but certainly it seems like one that, you know, we should be taking seriously.

And again, I, I think that some of the talk of human extinction or, or robots taking over the world or that sort of thing, uh, it's, it's certainly interesting to think about.

Did anyone respond?

What did you make of the argument, the defence offered by The Wall Street Journal?

It was very interesting because the one thing that everyone in the debate over AI writing had seemed to agree on up to now was that if you're going to allow it, you should at least clearly disclose it.

And the journal's editor said, disclose what? There's nothing to disclose.

And so he leaped past the debate that was being had and started a new one of is it OK to just run AI written material without telling anybody? Do you think they should have disclosed it? I do.

I think we're at a point in grappling with the use of AI to write where knowing when it's been used is valuable for the reader.

I think most of the journal's readers would probably like to know when an op-ed that carries Stanley Druckenmiller's byline has been written by Stanley Druckenmiller and when it's been written by Claude or Chachi Petit.

Are you suggesting some complex manual on what has to be declared and not declared?

It is very confusing, and I think that the line will be different places for different people and publications.

What did you make of the argument, the defence offered by The Wall Street Journal?

It was very interesting because the one thing that everyone in the debate over AI writing had seemed to agree on up to now was that if you're going to allow it, you should at least clearly disclose it.

And the journal's editor said, disclose what? There's nothing to disclose.

And so he leaped past the debate that was being had and started a new one of is it OK to just run AI written material without telling anybody? Do you think they should have disclosed it? I do.

I think we're at a point in grappling with the use of AI to write where knowing when it's been used is valuable for the reader.

I think most of the journal's readers would probably like to know when an op-ed that carries Stanley Druckenmiller's byline has been written by Stanley Druckenmiller and when it's been written by Claude or Chachi Petit.

And if we bring Will back in from the Atlantic, I just wonder, Will, whether you think in a few years time we might listen back to discussions like this and they may feel quite quaint because the degree to which AI has become part of how content is written and produced will be so deep and so intertwined that try to think about disclosing parts of it will be, well, frankly, impossible.

Yeah, I think that's possible.

And if you spend any time at all on LinkedIn or Twitter or Instagram Any sort of corporate setting, I would imagine you're going to just be bombarded with this, right?

Absolutely.

Yeah.

And the tells of AI writing have kind of evolved over time.

I mean, people may forget, but at the beginning, they were really easy.

You would search for, you would like search Amazon for, as an AI language model, I cannot...

And then, you know, whatever it is, but people would forget to take the prompt out or, you know, ChatGPT would always respond by telling you what it can't do and why in the early days.

Like, what are some of the hypotheses as to why AI loves this construction so much?

Like, what are some of the hypotheses as to why AI loves this construction so much?

Right.

So the, the first thing to be clear about is that humans, of course, use this construction too, and in fact, it's a hallmark of some really famous pieces of writing and rhetoric.

Um, at the beginning of my, my Atlantic story on this, I, uh, cite, uh, Shakespeare's Julius Caesar.

Um, you know, "The fault, dear Brutus, is not in our stars but in ourselves," or, "Not that I loved Caesar less but that I loved Rome more." So it was already present in the training data.

Then what happens with these AI systems, so first they just ingest, like, all the te- more or less all the text that's ever been written, and then they, uh, they learn to, uh, kind of regurgitate the patterns in that text.

There is another [laughs] weirder and wonkier theory that I-- To be clear, this is speculative, but I heard it from three different technologists who have worked with AI, so I thought it was interesting and I included it in the story.

It's filling up with its not X, it's Y. They're also sort of using synthetic data, as they call it, you know, just generating AI outputs for the purpose of training, and that can sort of enhance this effect, I guess, right?

So obviously you think it's a pretty good program, but what did Pangram find when it scanned Dagger Mouth? So

the researchers, to be clear, their study was not focused on Dagger Mouth.

They looked at a set of over 14,000 randomly sampled self-published eBooks on Amazon and A lot of authors, if you don't have a traditional publishing deal, you can get your book out to readers through this platform fairly easily these days.

They looked at these books and they ran all of them one chunk of text at a time through PanGram.

They also used another method to try to detect which ones were substantially written with AI.

The study was looking at the fact that books that were written partly or mostly or entirely with AI are now actually making a dent in the sales charts on Amazon.

It's not just pure slop that's designed to trick people.

AI was used in the writing of this book

What do you think of this approach and this kind of wave that's happening?

Yeah, it's interesting.

It's been building for a few years.

I mean, if you think back to a decade ago, Cambridge Analytica, fake news, Russian interference on social media, there was a push at that time to sort of regulate social media overall.

You know, can we you know, can we rein in the power of these giant platforms? Can we pass privacy laws? Can we, you know, force them to be better? And different jurisdictions have taken different approaches to that.

The EU has passed a bunch of regulations.

with varying levels of success.

But, you know, these efforts have met with tremendous backlash from... civil society, civil libertarian groups about kind of the problems of gating the Internet, what that would mean for everyone, for kids, for certain kinds of content, for, you know, government control of Internet content, all of these issues that you mentioned.

What do you think of this approach and this kind of wave that's happening?

Yeah, it's interesting.

It's been building for a few years.

I mean, if you think back to a decade ago, Cambridge Analytica, fake news, Russian interference on social media, there was a push at that time to sort of regulate social media overall.

You know, can we you know, can we rein in the power of these giant platforms? Can we pass privacy laws? Can we, you know, force them to be better? And different jurisdictions have taken different approaches to that.

The EU has passed a bunch of regulations.

With varying levels of success, the US has done pretty much nothing.

But, you know, these efforts have met with tremendous backlash from... civil society, civil libertarian groups about kind of the problems of gating the internet, what that would mean for everyone, for kids, for certain kinds of content, for, you know, government control of internet content, all of these issues that you mentioned.

Yeah.

Yeah, absolutely.

I mean, yeah.

So, so right.

I mean, there's something like on the most surface level, this is supposed to be sort of feel good, unifying, apolitical sentiment, right? Like people from around the world come to the U S they're welcomed, you know, by the locals with open arms, um, sort of debunking the notion that Trump's America is cruel or unwelcoming or hostile to foreigners.

And then not only that, but they just love it.

They love the giant Walmarts.

Maybe it's even true.

And you can get like tinfoil hat like pretty quickly.

Yeah, absolutely.

I mean, yeah.

So, so right.

I mean, there's something like on the most surface level, this is supposed to be sort of feel good, unifying, apolitical sentiment, right? Like people from around the world come to the U S they're welcomed, you know, by the locals with open arms, um, sort of debunking the notion that Trump's America is cruel or unwelcoming or hostile to foreigners.

And then not only that, but they just love it.

They love the giant Walmarts.

It's like everyone learned that term and was like, yeah.

Several state and local officials also tried to dispel rumors that FEMA was blocking or confiscating third-party donations, but there was simply too much noise online.

The reality is that our information environment isn't well-designed for debunkings.

I spoke with Will Areamis, a technology reporter at the Washington Post, as these rumors were spreading in 2024.

We worked with researchers from a nonprofit called the Institute for Strategic Dialogue, who did a report on this.

How many people were reached by 33 of the most viral falsehoods around Hurricane Helene.

And then we just sort of as a back of the envelope analysis looked at, well, what about FEMA's own tweets? How many people are they reaching? And FEMA's most popular tweets since Hurricane Helene have reached 50 times fewer people than the false rumors.

Will, why do you think Meta's doing this?

Well, Meta's playing from behind in the AI chatbot race, right? When people think of, of chatbots, they think of ChatGPT.

I think now increasingly they think of, of Claude.

Tons of people use Google's AI, often without knowing it, right? When you ask a, a Google search, you'll get an AI response.

Um, but, but Gemini is certainly growing in popularity.

Uh, Meta's playing from behind, and I think this is a strategic play by them.

I mean, it's not like all of a sudden they've decided that just, you know, privacy is their, their great North Star as a, as a company.

Why do you think it was AI that finally broke this? [laughs]

What do we know about whether these, these memes and videos are swaying public opinion on the legitimacy of the war in Iran?

That's a great question.

I'm glad you asked it because I, I do think, having covered misinformation and its impacts for a decade, I d- I do think it's nuanced.

Um, I think it's easy to fall into the trap of believing that just because there's false information on social media and people are sharing it, that that means that people are being duped or falling for it or completely changing their minds about something based on false information.

What we know, uh, in general is that much more often, these are people who are already inclined to agree with a certain point of view, right? So these are people who already think that Trump is using the Iran war to distract from the Epstein files, and they see this fake video and they're like, "Ah, here it is," you know, "Here's the proof." Do they...

Are they being deceived? Do they really think it's true? Uh, in some cases, probably yes.

In other cases, they may not care, right? It's like, it's more like sharing a political cartoon or a meme.

Mm-hmm.

Millions of dollars were at stake for bettors.

bettors.Fabian told me that he actually did consider just going along with these people.

Now, he ultimately said no.

He said he worried that if he went along with it in this case, they would come back to him and continue extorting him again and again, and he also just didn't feel it was right.

But he said he was scared, and his concern is that other journalists in the same situation might just take the offer rather than stare down a death threat.

But some states are pushing back, with many noticing that these markets are exempt from their gambling taxes.

Kalshi is legal in the United States for now, but many states have sued the site.

And in fact, earlier this week, the state of Arizona filed criminal charges against Kalshi, alleging that it was effectively operating a sportsbook and allowing betting on elections without a license and in violation of state law.

[chuckles]

Um, so, uh, Google Books, by the way, non-destructively, as I understand it, non-destructively scanned books.

They, they used this much slower process involving robots to go through and scan all the pages without breaking the spines.

It's kind of, like, fascinating.

Um, I went down many rabbit holes in the course of, [chuckles] of reporting this story, and that was one of them.

Um, but in this case, they didn't feel like they had time for that.

There's also a difference between...

And to me, it's just like, if you, if you back it all the way up, if they had just come to the end of the, "Well, we can't do this, this will take a while and cost us a lot of money," and they had decided to take the time and spend the money, I think as a society, we might feel really different about AI.

Um, why is this-- I mean, I, I guess, is this illegal, and why, why do people have concerns about what happened here?

So the potentially illegal part was actually what they were doing before this used book buying.

[music] One of the things that was really interesting to me about this story is that as bad as it sounds to [chuckles] try to go out there and destructively scan all the books in the world, this was actually seen by a lot of people in the industry as a more legal or more ethical way of doing it than what a lot of the other companies were doing at the time.

Authors have also sued Meta, which declined to comment for our story, and by the way, there are pending lawsuits still against OpenAI and Microsoft.

There's a lawsuit against Google.

They're all making broadly similar claims that these companies violated authors' and publishers' copyright.... what some of these companies were apparently doing, and what Anthropic apparently did before Project Panama, was going to these vast repositories online that are known as shadow libraries.

And these are unauthorized copies of millions of works that have been put together into these giant datasets, and they're traded around for free on the internet via this software called Torrent software.

Hmm.

So from the folks you've talked to, are there concerns about the downstream effects of unshackling this industry, of pulling back a lot of the restraints that have been there?

Yeah.

Well, to be honest, there aren't a ton of restraints there now, but uh, what- what restraints are there, um, may be, uh, in some peril after this plan.

Um, what Trump, uh, wants to do is, in his words, roll back a lot of the red tape that is holding back innovation.

And one way that this plan proposes to do this is to try to withhold federal AI spending from states that enact what the administration considers to be burdensome regulations on AI.

Now, they haven't spelled out what they consider burdensome.

They also say they want to leave room for prudent regulation of AI, so there's a lot of subjectivity in there.

Tell us a little bit about that, how the White House views it, and whether that's seen as something that has hindered innovation.

So, who wanted this moratorium on state laws around AI? It seems like big tech really wanted this, yeah?

Yeah, big tech and small tech and medium tech (laughs) and every kinda tech.

Um, they, a- and there are some exceptions.

There are some tech companies that do support certain types of AI policies.

But by and large, the agenda of tech lobbyists was, "Hey, we have an opportunity here.

We hear that Congress really wants us to be able to out-compete China.

Let's seize this moment and get behind this idea to stop states from regulating AI.

I wonder when you talk to people if they are synthesizing those things together or if they're still looking at them in silos.