Trump 正在卖手机 + 试图自动化所有工作的初创公司 + Allison Williams 谈《M3GAN 2.0》

Hello, Kevin.

Hello. I'm here in a beautiful studio in London. Casey, we have been trying to get a studio like this for years, and I think we just figured out that we have to move the show to London and tape it here every week.

You're in this lush red booth with the Hard Fork logo behind you on a TV. It looks like if Hard Fork were a Denny's diner, it would look a little bit like the studio that you're in. I keep waiting for them to bring you a plate of pancakes.

I want the Grand Slam Breakfast now, Casey.

Yeah?

Do you miss me in person? Is it different recording without me?

No, it's great. I have stretched my legs all the way across the studio for the first time. My circulation has never been better. It's markedly cooler in here, both in the temperature and vibe sense of the word. No, if you want to stay over there for a while, you're fine by us.

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

I'm Casey Newton from Platformer.

And this is Hard Fork.

This week, the Trump family is releasing a phone. We'll tell you how the president is using influencer tactics to profit while he's in office. Then, these AI co-founders say they're going to automate away every job. We'll meet the team behind Mechanize. And finally, we're going to the movies. M3GAN 2.0 is almost in theaters, and star Allison Williams is here to talk about it.

Well, Casey, what are we talking about this week?

Well, Kevin, if you hear an ominous ringing in the distance, you may be hearing the Trump phone.

Oh, no. What's the Trump phone?

This week, the Trump family announced 2 initiatives. One, a new cellphone provider called Trump Mobile. The other, a forthcoming smartphone. It's gold-colored, it has Trump branding, and they're calling it the T1, which is the same thing they called the first Terminator movie.

I'm sold. So, Casey, let's talk about this. What is actually going on here? What are they selling, and what is this mobile service they are operating?

Obviously, the Trump family has a lot of initiatives. They love to do a lot of branded merchandise, and for the most part, it all just washes over me, and I don't pay that much attention to it. But once the president of the United States' family says, "We're doing a smartphone, and we're going to have a cellular network," I think, Kevin, we should probably learn a little bit about that.

Yeah, so teach me.

There are 2 aspects of this to talk about. There is the cell network, what's called a mobile virtual network operator, or MVNO. That's Trump Mobile. And then there's the phone. Why don't we start with the MVNO?

Casey, what is an MVNO?

There are real cell networks like the ones owned by AT&T and Verizon, and they spend billions of dollars to build networks all over the country. But they wind up with unused capacity, and that creates the space for the MVNO to come in and say, "Hey, why don't you let us buy that capacity at a wholesale price, and then we'll resell it to other people and perhaps make a tidy profit?" Believe it or not, this has turned out to be a pretty good business for some people. For example, have you heard of Mint Mobile?

I have, actually. This is the one that is run or part-owned, or was part-owned, by the actor Ryan Reynolds.

That's right. So it was founded in 2016 and sold to T-Mobile 8 years later for more than $1 billion, which is not a bad price given that they didn't even have to build any cellphone towers.

Yeah, I remember reading about this and rereading that Ryan Reynolds had somehow made something like $300 million by selling this MVNO thing to T-Mobile. I thought, "That sounds like a great business. Maybe I should learn about it," and then I never did.

I was just learning about it the other day, Kevin. One of our fiercest competitors in the podcasting space, SmartLess, the podcast hosted by Jason Bateman, Will Arnett, and Sean Hayes, has launched its own MVNO. It's SmartLess Mobile.

Really? What you described to me just now is basically a surplus shop for cellphone service? If you're Verizon and, say, only 60% of your tower capacity is used, you could sell that extra space to an MVNO, which could then sell it to customers? How does it work?

Here's the twist. Here's how I would pitch an MVNO: It's cheaper for worse service.

Okay.

If you have a Verizon or an AT&T plan, my guess is you're going to be spending, I don't know, $80-plus a month on your service. But you're going to get to use your cell network during all the busy times. You're going to get priority. If you're on an MVNO, though, your service might be really slow during busy times. But in exchange for that, you might only pay $30 or $40 a month. Trump Mobile says they're going to sell for $47.45 per month, which appears to be a reference to the 45th and 47th presidents, Donald Trump.

So, okay, they're going to sell this service. Is it running on Verizon or AT&T or one of the big mobile carriers? How does it actually work? Who are they subcontracting for?

My understanding is that they're renting capacity from a group of those. It's not just one of them. They're going to bundle up unused capacity from a bunch of them and create the network that way.

I see. Okay. So that is the cellphone service that the Trump family is going to start offering. Now we have to talk about this phone. I know the following things about this phone. One, it's gold.

Yes.

Probably not actual gold.

Gold-colored.

Two, it's an Android phone.

Yeah.

Three, it is billing itself as being made in the USA.

That's right.

So all of those 3 things are correct, but I'm sure you know more about this phone. Tell me what there is to know.

They're calling it the T1 Phone 8002 gold version, which sounds like a Taylor Swift album. It will purportedly be sold for $499. The family has suggested you pre-order it now for $100. But there are many remaining questions about what this thing is. We've seen exactly 1 rendered image of this phone on the Trump Mobile website. There have been questions about whether this is actually a photo of a phone or maybe just a Photoshop render. Some people are even just calling it a concept of a phone, Kevin, if you can believe that.

Yes, I can. In part because I was reading a great story David Pierce had about this at The Verge that had an actual rendering of this T1 phone. It basically made it sound like this thing is either not real, or they have managed to come up with some miracle of supply chain logistics and device manufacturing that no one else who's been thinking about this stuff for decades has managed to come up with.

Typically, in a situation where you're a celebrity and you license your image, likeness, and name to whoever the highest bidder is, that bidder does not tend to be an incredibly innovative operator who is able to work supply chain miracles and create an incredibly premium good at the lowest price you can imagine. That's a pretty rare thing that happens in these cases.

Yeah, but those celebrities did not write The Art of the Deal. So I think we have to give him some credit here for it.

I think that is very fair. But to get to the heart of it, basically no one thinks that you can build a modern smartphone in the United States that runs Android 15 for less than $500.

I was skeptical of the price tag on this thing because I've seen some stories about the people doing the math on what it would cost to, say, make an iPhone in the U.S., and it's many multiples of the cost of manufacturing that overseas. It's not just all of the components; it's all of the fabrication. It takes a lot of specialized equipment to make all this stuff very precise. I cannot think of a scenario in which they could make something like this phone in the U.S., sell it for $500, and still make a profit on it. Am I missing something?

I think you are exactly right. That raises 2 questions. One is, if they are able to deliver it at that price, what shortcut did they take, or how did they get there? Will they be public about that? And if they're simply not telling the truth about the price and it's actually going to be more expensive, that doesn't sound great either.

Yeah, and there actually is, we should say, 1 U.S.-made smartphone that appears to still be shipping. It's called the Liberty Phone, and it is fabricated and assembled in California.

And the starting price of that phone, Casey, would you like to guess?

I'm going to guess $2,000.

$1,999. You lose by “Price Is Right” rules—but you win spiritually. $2,000 is apparently what it costs to have a phone that is assembled in the U.S. So if the Trump family has figured out a way to do that for 25% of that cost, I would be very impressed, but I would also not be surprised if they're just pulling that claim out of thin air.

All right. So why are we talking about this today? Well, one, it is just a funny story during a dark time. It struck both of us, and we thought it would be worth walking through some of those details. But there is also a really dark undercurrent here, Kevin, and it speaks to the utter strangeness of having a president who seems to be openly using public office for private gain.

And while I'm not going to make the case that the Trump phone is the absolute most important story of the week, given all of the tensions abroad and the protests at home, I do think it is worth pointing out to our audience just how many conflicts are baked into an idea as simple as, “Let's have a phone and let's have an MVNO.”

Totally. What I keep thinking about when I hear these stories about the various spinoff businesses that the Trump family is starting, I just feel so bad for Jimmy Carter. They made Jimmy Carter put his peanut farm in a blind trust when he took office because owning a peanut farm was seen as a potentially bad conflict of interest. I'm just glad Jimmy Carter—well, I wish Jimmy Carter were still around, rest in peace, but I'm sort of glad he's not around to see the absolute depths of side hustles that the Trump family has gotten itself into.

Yeah. Well, so let's talk about a few of the areas, Kevin, where this might raise a conflict. One is just the fact that telecommunications is a heavily regulated industry. Trump appoints the head of the Federal Communications Commission, which oversees the telecom industry. So now, if you're Brendan Carr, the head of the FCC, every time you go to make a policy, you're probably going to be asking yourself, “Well, what does this mean for Trump Mobile and the Trump phone?”

I have to imagine that this Trump Mobile MVNO thing is going to start off being a very small operation. But if it were ever to grow into something that was actually competing with the big mobile giants, I think it absolutely would be a very ripe conflict of interest there.

And now let me throw another conflict at you, Kevin. It's extremely common when a new smartphone comes out for a manufacturer like Samsung to go to some of the tech companies out there and say, “Hey, we have a new phone coming out. Would you like to make a deal with us? Pay us a certain amount of money, and we will put your app on our phone.”

Well, now imagine that you're Amazon or you're Meta, and you want to curry favor with the Trump administration because you have a huge amount of business before the government, and you're still working to make inroads with Trump and his family. Wouldn't this be a great time for you to come along and say, “Hey, Donald Trump, you name your price. We would love to get Amazon on the Trump phone. We would love to get Instagram on the Trump phone.”

And all of a sudden, you have opened up a new avenue, essentially, for bribery for these companies to curry favor with the Trump administration.

Yeah. It's fascinating, and it's so troubling for all the reasons you just outlined. But I also think there's a sense in which the Trump family's various business endeavors during this term are really giving us a roadmap to the ways that people have found to monetize influence in the last couple of years.

Mm-hmm.

Just look at the meme-coin business that it has entered, and that is actually making quite a bit of money for the Trump family. That is something that did not really exist in any scaled way a couple of years ago. But now, not just politicians, but lots of celebrities and influencers—people who sort of have their 15 minutes of fame—that is one of the ways that they sometimes try to cash in.

It seems like this MVNO thing is also becoming a way that people like the SmartLess podcast guys, like Ryan Reynolds, like all of the other celebrities who have gotten in on this kind of deal, can turn attention and reputation and influence and fame into money. And so I think it's worth saying that this is something that a lot of politicians are probably going to pay attention to and potentially try to replicate. Because if Donald Trump is able to monetize his influence, I think there will be lots of other people who say, “Well, if he can do it, why not me?”

I'm glad you brought up the meme coins because I think the meme coins offer a really tangible example of these avenues for influence that we have been talking about. The Wall Street Journal reported over the last week on a new financial disclosure from President Trump, and it showed that he had personally made $57 million from his family's crypto firm last year, and it put his current crypto holdings around $1.7 billion at the low end. That's a conservative estimate.

So why does that matter? Well, just as he oversees the FCC, he oversees the crypto industry as well. He appoints the head of the Securities and Exchange Commission, which has a lot of leeway to regulate how crypto is sold or not sold. And so here we can see exactly how much it benefited President Trump to come into office, sweep out all of the anti-crypto regulators, and bring in a bunch of pro-crypto regulators.

And by the way, he was heavily lobbied by the crypto industry to do that. They put a lot of money into his campaign. Well, now he has $57 million. So again, just to say, we try to set up a system in the United States where this could not happen. The Constitution has an emoluments clause that says you cannot accept direct payments or gifts from foreign governments, for example.

And we tried to just create strong norms that said if you're in public office, you cannot use it to just make a bunch of profits for yourself. But that norm, like so many others over the past 6 months, has been shattered, and it just seems like a really troubling precedent, at least to me.

Yeah. Part of what I find so curious about this moment with the Trump family and their business expansion is, typically in politics, you try to reward your supporters. The people who vote for you, typically you have some affection toward them, and you try to give them things that will make their lives better.

In this case, it is a weird inversion of that, where the people who are the most loyal, die-hard Trump supporters are going to be the ones lining up to buy the meme coins, lining up to buy the NFTs, probably lining up to buy the Trump phone and the Trump MVNO service. And they're going to be paying more money for things that are less valuable to them than what some other less Trump-affiliated carrier or seller would provide them.

And so I think it is just a fascinating experiment in how you can pretend like you are rewarding your most loyal followers by really selling them something that's not very valuable. And I wonder how long that will last or whether there will be people who buy this stuff and say, “Hey, wait a minute, my old Verizon service was way better,” or, “My iPhone was way better,” or, “I wish I didn't lose all this money on crypto coins.”

Yeah. It seems like the president has figured out ways to monetize tribalism in ways that have been very beneficial to him, and I suspect pretty enjoyable to the people who are buying at least some of these products, but it just goes against so many precedents in our country's history.

I want to say one more thing about this. The great Russian and American journalist Masha Gessen wrote a piece in The Times recently that really resonated with me. They wrote about the shock of authoritarianism. They've spent a lot of time in Putin's Russia, and there was just one shock after another of this norm being shattered and that norm being shattered.

And the human mind, they write, is always seeking a sense of stability, is always seeking a sense of, “Okay, well, that shock might have happened, but my life is still basically the same.” And they wrote about the danger of that because as shock after shock after shock accumulates, you wake up one day, and you realize you have much less freedom than you used to, and a lot of other bad things have happened.

So in and of itself, a Trump phone, a Trump MVNO might not seem like that big a deal, even though they really are, in their own ways, quite shocking in the context of every other presidency we've ever seen. But my fear, Kevin, is that as more and more of these shocks happen, we do get desensitized to it. We do stop paying attention to it.

We don't bother doing a segment about it on a podcast because we just think, “Oh, well, you know, that's Trump. What are you going to do?” So at least this week we wanted to say, “Hey, this thing that seems crazy, it actually is super crazy.” When we come back, they're coming for your job. Why the team behind Mechanize thinks they can automate all labor.

Well, Kevin, you recently had a very interesting story about a new company called Mechanize.

Yes. This was a fun one to write. This was a startup in San Francisco that got started earlier this year. They raised a bunch of money from people you probably know. Patrick Collison was one of their investors, and Jeff Dean, a big AI honcho at Google, was another one.

They are a very buzzy startup, and what attracted me to writing about them was that they have said their goal is to automate all labor. They want to take away everyone's jobs—yours, mine, everyone we know—and replace them with AI. They think they can do this in the coming decades with a new type of reinforcement learning system.

Well, that is a promise we have heard from a number of Silicon Valley companies, but maybe none as directly as the Mechanize founders are pitching it. What exactly is their secret sauce?

They are building what they call reinforcement learning training environments: basically, simulated environments that these new AI agents can use to learn how to do various white-collar jobs. They believe this approach can scale not just to software engineering, which is the first job they are trying to automate, but to all other kinds of jobs eventually as well.

I thought this was an important conversation to have on the show because we have been talking about AI and jobs for the last couple of episodes. I think this is a conversation that is beginning to grow more and more important and more and more urgent, and we are starting to see evidence of some job displacement from AI.

And then along come these guys at Mechanize who say, “Well, this is actually only the tip of the iceberg. You have not seen anything yet. Our plan is to help these giant AI companies automate a bunch more jobs very quickly.”

Well, that seems to raise a lot of important questions about how society would deal with the fallout of such a thing. So why don't we bring in these founders and see what they have to say about it?

Yep. Let's bring in Matthew Barnett and Ege Erdil from Mechanize. Matthew and Ege, welcome to Hard Fork.

Hi.

Hi.

So I had a lot of fun writing this story about Mechanize, the company that you all founded, along with your third co-founder, Tamer Basaroglu, earlier this year. One of the things that made me interested in what you are doing is that unlike a lot of AI companies I cover, who sort of pretend not to be automating jobs, or say, “You know, we are just making helpful copilots and assistants for workers. We are not going to replace their jobs,” you all were actually coming out and saying, “Yes, we absolutely want to automate jobs, and not just a couple of them. We want to automate all jobs.”

Tell me what inspired you three to leave Epoch AI, the research firm where you were before this, and start this company, and to be so open about the agenda of automating labor.

When we were at Epoch AI, a nonprofit research organization whose mission was informing society about trends in artificial intelligence, we did a bunch of research into the economics of AI. As part of that research, we looked into what the impact would be of AI that could substitute for human workers across the economy on the economic growth rate.

A very robust conclusion of that research was that it would speed up economic growth by enormous amounts—unprecedented amounts, maybe by 10 times or more compared to current rates. That would unlock such a vast abundance of not just material goods, but also services that today can only be provided by humans; technological progress, like medical progress, that currently, no matter how much money you have, you can't really purchase.

We think that if AI automates everyone's job, because we can scale the AI workforce so much more than we can scale the human workforce, that leads to this vast abundance—an enormous increase in a variety of goods, new medicine, new technologies, et cetera—and makes people's lives much better.

So this is a story we have heard from other AI founders. They want to unlock a world of radical abundance. I think a lot of our listeners hear that and think, “Okay, here come the Silicon Valley hype guys, and they are out there raising funds, so they are going to tell me this beautiful story so that they can raise billions of dollars and get rich.” How do you respond to the idea that this is just a bunch of hype that you are selling to benefit your own project?

I would say that the difference between just hype that someone is speculating about and something that is real is that you can test it empirically. You can look at whether it is an implication of robust economic models. You can try to look at the history of automation.

I would say that if you look at the empirical evidence, it is quite clear that automation has been good for most people. People have benefited from the mechanization of agriculture, from refrigeration, and from all these sorts of technologies.

Anyone who thinks that it is just benefiting a small group of people should really study the history of automation. I think almost all the evidence would show that they are wrong.

Well, I have studied the history of automation. I wrote a whole book that touched on the history of automation, and one of the things that I want to really impress upon you all is that I do agree with the statement that automation and technology broadly generally improve people's lives in the long run.

I don't think, for example, that a lot of us would willingly switch places with our great-great-grandparents. They had hard, backbreaking lives of manual labor. But people don't live in the long run. People live in the short run, and in every technological revolution that we have ever had, there have been people who struggle, who fall through the cracks, who are not able to seamlessly make the jump from one era to another.

What do you say to those people who look at you and say, “Well, I get that this future of radical abundance may be possible somewhere down the line, but for me, in the year 2025, what this looks like is my job getting automated away and me not having any way to pay my bills”?

A few things. First of all, I think we expect AI to speed up economic growth, assuming it can substitute for everyone's job and not just the job of a few percent of workers or something like that. The impact of that is so big that I don't actually think the long run is that long. It might be a few decades or something like that, so it might easily be within the lifespan of most people who are currently alive, for example.

But the other thing is, I think the standard that we should only automate jobs, or only embrace new technologies, if there are no losers—if nobody is made worse off by the adoption of a new technology—is extremely strict. I don't think that is a reasonable standard.

If AI can actually substitute for human workers across the entire economy, it is a substitute for human workers. In that case, humans would no longer be getting income from wages, but there are lots of other sources of income.

There are countries in the world today where citizens actually get their income from, say, natural resource endowments. There is just a certain amount of natural resources that a country owns, and the government has maybe a sovereign wealth fund. Maybe they have other ways in which they can distribute the income from that natural resource to the population, and that is something we see in our world today.

It is not actually that far-fetched, and that is the kind of thing I would expect to happen in a world where AIs are vastly more capable and human workers cannot compete.

So you guys published a blog post recently where you wrote about the history of automation in software engineering.

Mm-hmm.

You note that automation has been coming to software engineers over a period of decades, and as that has happened, their jobs largely have not been eliminated. What are you seeing right now that is making you say, “Okay, it really is different this time, and we are going to be able to go that last mile and actually fully automate everything and take away these software engineers' jobs”?

I'm not sure we are actually saying that something is definitely different this time. In fact, 2 of my co-founders think that maybe full automation will take many decades.

And so I don't think we're actually taking the strong view that there's something different in the next 5 or 10 years such that everyone will lose their jobs. I actually think that, especially in the next 5 years, we'll probably see a continuation of past trends, which is that AI automates some tasks within professions. It doesn't completely automate the entire profession in the sense of completely replacing most workers in those professions. So, to take the example of software engineering, we think that these coding assistants will be used to help software engineers.

We don't necessarily think that the coding assistants will be able to do all the jobs that a software engineer can do. For example, a software engineer often needs to coordinate across teams. They often need to plan projects and test the software to make sure that it's up to the design specifications. They need to do a lot of these different types of things that are very hard to automate, which aren't just under the label of coding.

And so we think that if AIs can just do coding but can't do these other things, then in fact it will lead to a productivity increase in these professions and probably raise wages for software engineers, even though it's not taking over their entire job. Of course, in the long run, we just expect that it will be able to take over people's jobs, but this isn't because of some specific thing that we think is different now. We just have a projection of how long we think it will take for AI to replace everything, and we have disagreements about that. It's not that we think that it's a qualitatively different type of automation. We just have a guess at how long that type of thing will take.

Got it. Okay, so let's talk a little bit about what you guys actually do. You've just raised a bunch of money. You're presumably hard at work building something. I've read a little bit about it in Kevin's article, but tell people what it is that you're actually building.

Yeah. So what we're building is a reinforcement learning process where we will design essentially virtual work environments, you could say, for models to acquire the skills that human professionals have that enable them to do their jobs.

Like using a spreadsheet, maybe.

Yeah, like using a spreadsheet, using common software tools that people use in their work, like Slack for messaging or checking their email. If they were doing software engineering, then using tools like GitHub. So what we're doing is creating these work environments with scoring. Basically, we have a bunch of tasks in these environments, and we can score a model to see: Did the model do this task well? How well did the model do this task? And then we give this to an AI company, and they're able to train their model in this environment. They handle the parts where they train it, but we are the ones who supply them with the environments in which they will be doing this training.

The way your co-founder Tammy explained it to me, which I thought was a useful model for my understanding, was that you are basically creating what amounts to very boring video games. Like, a video game in which the goal is to be a software engineer, be a lawyer, or be an accountant. You set up this environment, and then the AI agent goes and plays it a bunch of times, gets the signal for whether it failed or succeeded, and ideally gets better over time until it's good enough to actually do the full job. Is that more or less correct?

I think that's a good description, yeah.

What do you think the next target is after software engineering? What is the next easiest job to automate using this technique?

I guess there are some things that are adjacent to software engineering. Data science, for example, might be a good target. And we have—

Podcasting?

I think that would take a different kind of reward signal, maybe.

Okay, so you're sort of at work on this system. You know, I'm struck by the fact that there's this interesting tension where, on one hand, you're saying that you're going to automate all labor. On the other hand, you're saying, well, it might take 20 or 30 years. You've also raised venture capital, and they all want a return within 7 years. What are those conversations like? What are you promising to give them before the end of the decade?

Even if you just automate a substantial fraction of labor but not all labor, that would still be extraordinary in terms of the amount of progress we would've made over a brief period of time. I think if we could get as far as, very ambitiously, automating 20% of current jobs within the next 5 years, that would be insanely valuable from just a conventional perspective. So I don't think we need to achieve the ambitious long-term goal of automating all jobs for this to be a successful venture within just a brief period of time.

Got it.

I have a question for you guys. When I was at your launch event and stood up during the Q&A and said, “Is any of this ethical?” Matthew, you had a response to me that I thought was interesting, so I want to ask you for an abridged version of that case. Make the case that what you are doing, trying to automate jobs, is ethical.

Well, I think what I said was that it was a question of costs and benefits. I would say that there are costs to automating everything. It's true that people will lose their jobs. However, we need to compare this to the enormous upside potential from automating all jobs, which is this vast prosperity that would be created from automating labor. It'd mean that goods and services would be much easier to produce, which would mean that we'd have a much higher standard of living across virtually all areas of life.

I think one thing in particular that people miss is that the secret to mass consumption is mass production. In order to get people to consume a lot of goods and services, in order to get people to have a higher quality of life, this needs to be backed up by a lot of production, with a lot of goods and services actually being produced. So if you have some mechanism that's able to expand the base of goods and services that are being produced, then, as long as there's some way for this to be shared among the broad population—not necessarily equally, but just as long as there's some way in which these goods and services can be distributed even slightly—I think that just leads to almost everyone becoming better off.

And not just in terms of having a higher material standard of living. I think also their lives would still have meaning. They would find new ways for their lives to have meaning other than work. I think, for example, one thing people perhaps didn't anticipate in 1800 is that, if you were to ask people, “Would all this automation be able to work out if people aren't going to be constantly working on a farm?” they might not have been able to anticipate that, for example, in the next 200 years we would have government-funded education in universities.

People might not have realized that there's this alternative way of enjoying your time, which is going to school, which is going to college. This is a new way for people to spend their time, which I would argue is even more meaningful than these long hours toiling on a farm. I would just think that that's the default that I would predict as a result of this empirical trend that we've already observed.

I know you guys are focused on the capitalism part of this equation, but I'm curious what thoughts you do have on the role of government. We're always struck on this show by the disconnect between, on one hand, entrepreneurs like yourselves saying, “Hey, in the next 18 to 24 months the world is gonna look very different,” and, on the other hand, the politicians mostly are just kind of like, “Okay, cool, go for it, yeah.” Are there things that you would like to see them do, or that you think they could do, to get us ready for a world that was very different within a few years?

I think it's a hard question. It's very hard for me to say what we could do today to make a world in which all jobs have been automated, or most jobs would have been automated, much better. Right now, I only have a very vague sense of what that world's gonna look like. When that world's actually here, we will have a much more detailed understanding of how things are gonna work.

And Ege, I'm gonna cut you off there. I understand that what you're saying is, it's hard to know the future and it's hard to make policy based on something that is moving and changing so quickly. But do you think there are things that governments could or should be doing to cushion the fall for workers who may lose their jobs in the next, say, 5 to 10 years? Should they be rolling out something like a basic income or strengthening the social safety net? Do you have any ideas for how we could cope with mass job loss if what you are predicting comes true?

I don't really think we're predicting mass job loss in the next 5 or 10 years. At least I would assume that initially the impact of AI automation is actually going to drive wages up, and it will lead to some occupations changing in character.

For example, software engineers might still be employed, but their jobs might start looking different because some of the tasks have been automated. I think the world in which there is mass job loss and mass unemployment is further away. I would say that world is definitely more than 10 years away. Maybe Matthew disagrees with me about this a little bit.

Well, I definitely don’t expect mass job loss within the next few years or the next 5 years. So I do think it’s premature to start talking about government programs to cushion against that.

The thing that I would like to emphasize the most here is that I think when people come up with plans for the future, especially if it’s not an immediate plan of what to do in the next year, for example, I think that’s overrated. If you asked people 10 years ago, “What should we do to prepare for people losing their jobs from LLMs?” how many people would’ve been able to come up with a good recommendation of what to do by 2025? I just think that would’ve been overrated.

What’s underrated, I think, is just being honest about our intentions, saying, “This is what we intend to do. This is our roadmap, perhaps. We don’t actually know whether a roadmap will succeed, but this is at least what we’re planning on doing at the moment.” We think that when these things become more apparent and more salient, people will be able to leverage the knowledge that they have at that time, which will be far more detailed than they would have had 5 or 10 years in advance of the event actually occurring. Then they’ll be able to use the tools and knowledge of the time to craft appropriate policy.

Committing ourselves to some sort of plan or policy ahead of time without knowing these details just seems foolish to me.

Well, my counterpoint would be that if you know a pandemic is going to come at some point, you can manufacture and stockpile vaccines, right? Even if you don’t know exactly when it’s going to arrive. My hope would be similar here: if you know that there is going to be massive job loss to come, you could start saying, “In such a world, what kind of solutions might exist?”

Sure.

You have a theory of the case.

I certainly agree that we can look at the character of the solutions that might exist. The character of the solutions that would look reasonable is similar to the character of the solutions that we’ve already seen in the past.

Governments, for example, have already become more generous in terms of redistributing income in the last 100 or 200 years as automation has progressed. We have Social Security, for example, which didn’t used to exist at all. We have Medicare and Medicaid. We have all these different programs for caring for people, for the poor. We have unemployment insurance.

I would say that a continuation of that character—those types of things for taking care of people—seems to make a lot of sense if you care about redistributing income. I would say that’s a reasonable thing to do in the future, but I don’t necessarily have a particular plan for what that would look like. I’m not saying a UBI would be best, but I think that type of thing makes a lot of sense.

As I said in the article, I’m glad that you’re being honest about your intentions. I think it’s useful to have an honest and open conversation about the possibility of job loss through automation with AI.

I do hope that you will find some sort of empathy for people who are really scared about what’s coming. We hear from listeners every single week on this show who say that their jobs are changing in ways that they don’t necessarily like because of AI, and who are anxious that their bosses are trying to automate them out of a job.

I would just say that my free advice to you as you go out making your pitch about automating all jobs is that there are people on the other side of that, and I think those people are concerned and worried, especially when they hear guys from San Francisco talk about how they’re excited to automate their jobs.

Yeah, I would say it’s difficult when you’re doing something that you think has enormous benefits but has some costs. In those cases, you can look at the people who are bearing the costs or who fear that they might suffer the costs, and it’s just a very difficult thing to do.

Of course I have empathy for those people, but I acknowledge that it feels cold for me to say that I feel empathy for you, but I just think that these benefits that I’m listing are much greater than the costs. That’s always, I think, what it sounds like when someone’s performing some sort of utilitarian calculus.

But compared to the vast majority of policies that people actually talk about in the public sphere, this is something that is much more positive-sum than negative-sum than you might have otherwise imagined.

All right.

All right.

Well, I think we’re going to have to leave it there, but hopefully someday you can come back for another round of spirited debate after you’ve pushed the first—

Yeah. Come back when you’ve automated podcasting and give us the news.

All right.

Thanks, guys.

All right.

Take care.

Well, Kevin, when this show started, you and I made a pact: we would see every M3GAN movie released in theaters. This week, we honored that pact.

I don’t remember making that pact, but we did see a new M3GAN movie on Monday night together. We had a little double date with our partners.

Yeah, and this is a movie that has a lot of ideas in it. That may surprise you for a movie that I think is mostly designed to be a lot of fun, but we were laughing as the movie went along at just how many concepts in the film are things we have talked about on the show. We were really excited when we heard from the movie studio that we would actually be able to talk to the star of M3GAN herself, Allison Williams, about this movie.

Yes, and I will say my guess is that this is the only movie of 2025 that will contain a reference to the paperclip maximizer. That is how deep down the rabbit hole the screenwriter and director, Gerard Johnstone, went here. So, M3GAN 2.0 is a movie about AI that managed to get a lot of the inside terminology of the AI world into its script.

It comes out June 27th, but in the meantime, here’s Allison Williams.

Allison Williams, welcome to Hard Fork.

Oh, my gosh, what an honor. Thank you so much for having me.

It’s such a pleasure. Casey and I had a double date on Monday night.

Mm-hmm.

We both went out to see M3GAN 2.0. We had a great time. What struck me about the movie is that it was quite impressively literate about some of the nerdier, more arcane parts of the AI and tech universe. I’m curious: how much research did you do going into this, with concepts like instrumental convergence, which only a couple hundred real AI nerds in San Francisco talk about?

And some actors and actresses now, too. We’re evolving.

A lot of research. On the first movie, I did a lot more research into robotics, engineering, AI, and women in tech and all of those things, because I couldn’t be farther from a woman in STEM. I’m a woman in English and film and television.

On this one, I did much more physical preparation. But yes, to be able to speak cogently about these things and also make it sound like you know what you’re talking about, it’s always smart to keep up with the reading and what there is to know about what’s going on.

But yeah, I give a lot of that credit to Gerard. He doesn't just do the superficial pass where it's gobbledygook that means nothing. We really do want exactly what just happened. We want praise from the 100 people who know what we're talking about. And honestly, we can go home now. This was really all we needed. Thank you guys so much for having me.

Watching it…

Yeah.

I did have that feeling of, “Oh, they put real ideas in this movie.” It really does feel like it has been keeping track of a lot of the big discussions that are happening about the role of tech in society. Your character, Gemma, becomes a bit of a screen-time crusader as this movie begins. She's lobbying against smartphones in schools. I'm curious—

Yeah.

How much of that resonates with you personally?

So much. I think Gemma and Jonathan Haidt would have gone on a tour together, just talking about these issues. I really think that the first movie was sort of posing a hypothetical that then, by the time it came out, was starting to feel very prescient and kind of urgent. This movie is sort of saying, “Okay, hypothetical over. We are here. Now let's have an ethical conversation and a moral conversation about now what?” Sort of about parenthood and stewardship, and the parallels in the first movie between Gemma's motherhood, so to speak, of Cady, her niece, and of M3GAN are still at work in this movie, except it's an even more loaded word, I will say, without spoiling anything.

And it's definitely trying to fully entertain you in the theater. You're on a ride, and then when you get home and get into bed, you're like, “So I guess we should be expressing a little more gratitude to the Roomba? I think we should be.”

Yes. Yes.

And like I did the other day when I rode in one for the first time, say thank you to my Waymo, I think.

You gotta say thank you to the Waymos. They're keeping track.

You gotta. Listen, I feel grateful for anyone who gets me anywhere safely, including an inanimate line of code or many, many lines of code and cameras. So I feel like it is definitely asking us to think critically about our ethical responsibility. It feels like it's asking us to enter into—or realize that we have already entered into—a relational positioning rather than a parasitic one, where we can just use and use and take and take and take and expect and expect, versus being in relationship with…

And it sounds crazy, but it is sort of the main question of the movie: to be in relationship with these types of ways of existing that we have brought to life, so to speak.

Yeah. I will say, this does not feel like a far-future scenario to me. In fact, just recently, OpenAI announced that they're doing a partnership with Mattel, the toy company, and there have been—

Yes, I got that—

—lots of companies—

I got that article 150 times.

I'm sure. I'm sure. For an official merch tie-in, are they going to do a M3GAN doll?

I don't know. It hasn't come up yet. But when I read that headline, I was like, well, you know, sometimes it takes a little longer than this for sci-fi to become reality, but this just feels like it's part of our marketing cycle.

Now, I think we have kids around the same age. Mine's 3.

Yes.

Would you give your kid an OpenAI or just an AI in general toy?

Okay, well, here's the deal. As you know, with a kid at this age, our son is incredibly curious. I love this side of him and how inquisitive he is, and also how dissatisfied he is with any surface level. He can feel when you're phoning something in or when you just don't have expertise, and in his least condescending way possible, he'll always ask for supporting documents and evidence—

Yeah. He—

And he'll be like—

They really do the citation—

“Let's flip to the back.”

—the needed thing a lot.

Exactly.

Yeah.

“Let's flip to the back. Let's go through a bibliography. Let's see where we can go a little deeper here. Let's go past the Wikipedia of it all and get some primary sources.”

So often when I'm trying to explain something to him, like jet propulsion, which he asked me about last week, I turn to, as we all do, ChatGPT. I'll speak into the speaking thing for a hands-free thing if I'm multitasking and say, “In a way that a 3-and-a-half-year-old can understand, can you explain how a rocket launches and how jet propulsion works?”

And then Arlo watches this little orb on the phone just going in and out. It's the least stimulating graphic experience on the planet, and yet his eyes are—I'm watching something so intense happen. His brain is not able to comprehend what this interaction is that's taking place, and he's hearing all this information, and it's distilled perfectly, and it's timed perfectly, and it's exactly what he wants to know. Then he'll ask a follow-up question, and the look on his face is so troubling that I will cut it off and make him stop talking.

Hmm.

The other day, we were talking about a subject; I can't remember what it was. It might have been something related to animal husbandry or something. It's a long story.

Oh.

We were talking about baby deer a lot.

Mm-hmm.

Anyway, he said, “Can you ask Chapatiti?” And I was like, “Chapatiti? Who is Chapatiti?” And then he was like, “The person who talks from your phone and answers my questions.” And I was like, oh my God, we are finished using this technology as parents. This is upsetting.

And honestly, this is free for you, OpenAI, but Chapatiti is a very cute and very benign-sounding nickname for an extremely powerful machine.

I love that story, and it just makes me curious about the tension, because I feel like the tension that you're describing is also in the movie, on the one hand—

Exactly. That's what I was just going to say.

Yeah, you have this thing that is able to mesmerize your son and answer his question. Kids are famously curious. They often exhaust—

Yeah.

—their parents' patience, but now you have something that can just explain everything to them forever. That's kind of the promise of M3GAN, too, right? “Hey, let me take a little bit of this parenting work off your hands. Be an extremely supportive friend. Commit a little light murder—

Yeah.

—for you, if necessary.”

Totally.

So I'm curious, in the real world, when you're away from the killer doll and you just have this mysterious new technology, how you navigate that tension?

Well, it's really interesting. Honestly, I never get tired of explaining stuff to him if it's in my wheelhouse. The times that I reach for ChatGPT are when he has wandered beyond— I can't explain gravity, period. I can't explain it to a 3-and-a-half-year-old for sure. You know who can? ChatGPT.

What I used to do was look it up on ChatGPT, and then I would translate it so the information was coming from me. Then there was a time when I needed my hands, and I just did the voice-activated thing, and I didn't think about it, and now here we are. I think our strategy is to limit the amount of stimulus to an amount that he can tolerate.

Mm.

He and I talk all the time about overstimulation. When he was 2 and a half, he was running around the house at breakneck speed and he was like, “Mama, I think I'm overstimulated.” And I was like, “You have nailed it. That is exactly what we're seeing right now.” In the same way, I think that an amount of knowledge that is delivered in such a confusing vehicle can also be overstimulating. I think there's such a thing as just overdoing it, being overwhelmed by the amount of information that's coming at you.

And as anyone with a toddler will tell you, often the best amount to give them is just enough—

Mm.

—for them to wonder more—

Mm.

—and fill in the gaps for themselves, because sometimes our instinct to overexplain things deprives them of the opportunity to answer some of it on their own, and the things they come up with are so extraordinary—

Mm.

—and the questions they end up having. So I try to remind myself that less is more, and he's just learning so much every single day, and a little bit every day is totally enough.

So that's how you feel about AI as a parent.

Yeah.

I'm curious, as a creative and a person who makes and acts in movies, how you feel about it. Obviously, AI and Hollywood have had tensions going back years.

Yeah.

You've now been in 2 films featuring AI as a major theme.

Yeah.

And I'm curious, when you put your actor hat on, how you feel about this technology.

It's obviously intimidating. My mind goes instantly, of course, to job security, because—

Hmm.

I'm like, “Okay, what about my job can't be performed by an algorithm or an AI or a computer or something like that?” And the answer, honestly, is a little bit embarrassing, but it's the parts of us that are flawed and that make mistakes and that don't do things perfectly, like a hair out of place that looks normal, or a smudge of lipstick, or a lack of continuity here, or a slurred word occasionally in a sentence the way we all speak, or handwriting that's not consistent.

All of those things are so human. I rely on the tiny moments where I’m bad at my job to save my job, frankly, because I think that’s how art ends up feeling human to me: the parts of it that aren’t executed perfectly.

Well, my question in this vein was: a few years from now, it’s time to make M3GAN 5.0. They come to you.

Uh-huh.

They say, “Allison, we have so much footage of you.”

Oh, I have been anticipating this since the first movie, by the way.

Yeah, okay.

Keep going with your question.

Sure. If they come to you and say, “Look, we’re going to write you a really nice check, and you don’t actually have to do anything. You can stay at home, go make another movie, or do whatever you might like to do. We’ll just make M3GAN 5.0 using your digital likeness,” what’s your gut reaction to that kind of offer?

I’ve actually contemplated this from a contract standpoint. I would want my likeness compensated, first of all, to the same amount that my actual personhood would be.

Yeah.

So that’s the first one: I would want to make it as cost-prohibitive as possible—not to say that I’m enormously expensive to hire. I just mean, I would not want it to be the vastly cheaper option, and I actually think, for now at least, it is the more laborious and difficult option. Luckily, I feel safe there unless you’re going for an uncanny kind of thing.

Mm.

But even for the M3GAN movies, we do a combination of human and machine. It’s crucial to our creation of M3GAN and Amelia and for all of the iterations of all of these little beings that we have in this movie. I don’t want to give them all away, but they’re all a collaboration. There will always have to be, I think, a human component for all of it, just because of what it is.

Let’s say something happens to Gemma, she’s no longer alive, and she’s reanimated as AI. Let’s say that’s the situation in M3GAN 5.

Yeah.

I may have just written it with you guys on this podcast. What do I do? Put a flag in it? IP: this is mine.

Yeah, that’s yours.

No, that belongs to the Hard Fork podcast now. Sorry.

Okay.

Kevin.

Shit, we don’t have room for—

Give it to the guys.

Just 4 cooks in the kitchen. That’s fine. It’d be an awesome collab.

Anyway, I think if that were the case, they would still, at this point, need me to be involved. They would need someone’s physical body to be involved for it to work, even from a motion-capture standpoint. But I do want to—I guess I should copyright my likeness. I don’t know.

I already watch cuts of trailers for movies that I’m in that have lines of dialogue that aren’t in the movie but are said by an AI version of my voice.

Mm.

They don’t end up making it to air. They’re just in the draft versions, and then I go in and record them. That’s always really strange because it doesn’t exactly sound like me, but for a second I’m like, “Wait, I don’t remember saying that. Oh, okay, that’s AI.”

Hmm.

It’s just so rudimentary now. But, yeah, as of now, there has to be a human pass on it to make it sound—worse, I guess, in a way that’s more normal, and better, conversely.

I want to ask a follow-up related to how different the 2 films are. M3GAN feels very much like a horror film.

Mm-hmm.

M3GAN 2.0 feels a lot more like an action film. To me, it felt like Mission: Impossible crossed with Terminator 2. I wonder, as you’re reading both scripts—when you read the script for M3GAN 2.0, do you feel as an actor, “The genre has evolved here, and so my performance needs to evolve”?

And if you did feel that way, if maybe that is exactly the sort of thing that’s going to be hard to translate when it comes time to make M3GAN 5 and the genre has shifted yet again, and they don’t actually have the raw materials they need to do the performance that you would have done—

Exactly.

That you would have done.

Exactly. I mean, that’s a great point. I think the thing that’s fun about this movie is that we started having conversations about what it was going to be before the first one even came out, and the clear mandate from every corner of our world was: there’s going to be a second doll.

And so the question then was: okay, let’s extrapolate from there. Obviously, we’re in a world where T2 exists, so let’s not make a direct 1-to-1 reboot of that movie. Let’s figure out how to make it interesting and also live in our world.

And then, as Gerard started to put the pillars of the plot together in a way that felt so logical to him but was so unpredictable to everyone else as you start watching the movie—and made for such a fun, unexpected, and unpredictable ride—he realized that we had to be in an action world for reasons I don’t want to spoil. The stakes get really bigger. The world gets bigger. That’s what a Yale education can give you—and not just a Yale education, an English major. The stakes get a lot higher.

The world expanded in such a way that the story pushed us into action. It wasn’t the tail wagging the dog the other way. Once we realized we were in that genre, it was really cool to realize that we had created a tone in the first movie and characters that could be transported. As long as there’s still that thriller DNA in it, it can be translated into a different genre.

And now, as we are talking about possibly being lucky enough to do another one, we’re asking that question again: “Okay, so where to? Where do we go next?” What should we—where should we take this little team of misfits? Where should we creep other people out in the future? It’s a really fun way to iterate the franchise.

Yeah. I mean, the thing that stuck out to me about M3GAN 2.0 compared with the first one is that it just feels like a much more complicated moral. If the takeaway from the first movie was “AI bad,” there was some complication of that in the second one. I don’t want to spoil any details, but there are points at which AI is part of the solution.

Did you feel like the script that you were reading and then performing was more complicated? Is the moral of M3GAN 2.0, to the extent that the film has a moral, something different from “AI bad”?

Yeah, for sure. I think it had to be, because making a movie with the moral “AI bad” would be a deeply dick move. It would be sort of like—I don’t know. It would just be obnoxious to make a movie that’s like, “Psych, we think this is bad and should go away, but you’re stuck with it, so…” You know what I mean? It would just be counterproductive, whereas I feel like the movie we made is asking a much more nuanced question.

We’re all on a similar moral journey to the one that Gemma’s on. By the end of the movie, without giving anything away, we hear exactly how she feels on the topic, and it is, shocker, very different from the way it is at the beginning of the movie. Not that being cautious is a bad idea, but there needed to be asterisks to her argument because of where we are as a society, and there just are certain realities to that that she needed to contend with.

Yeah.

In the first movie, there’s a wonderful scene that’s kind of a combination dance-murder. In this one, I desperately want to spoil it, but I won’t: there’s an incredible song that brings the house down. How important is it to the creative team for there to be an iconic moment of gay culture in every M3GAN film?

Extremely important.

Okay.

But it’s not engineered like that. It’s just a moment where she behaves in a kind of shocking way—a shocking use of the arts, I guess—especially when it’s coming from tech.

I had this realization the other day that I played a character on Girls who broke into spontaneous song to the detriment of everyone around her quite frequently, never asked permission, and it was never the right time.

Mm-hmm.

I’m almost being karmically punished by watching this other doll—

This is your cross to bear, yes.

It is, sort of. I have now brought another being to life that does this, where she’ll spontaneously dance and/or sing at certain points in a movie when it’s never feeling quite right.

Anyway, I think it’s important to us that we make a movie that feels so tonally specific. Making sure that we hit the same tone as the first movie was priority number 1.

Mm.

Because if we became too self-aware and too in on the joke and too winking at the camera, it would be a totally boring movie to watch. It would feel exhausting.

Mm-hmm.

And so we knew that eventually the camp would be there, and the self-awareness and the in-on-itself would be there. But it would only be there if we executed completely earnestly on the story and the emotional beats as they were there. And I think watching women with really strong personalities committing to an emotional beat—

Mm.

—and then rising to the occasion and their strength is, in and of itself, sort of a celebration somehow of queer culture.

Absolutely.

And so it feels like making sure that we just stick to that in these movies is not at all an effort to, I don't know, placate anybody, but ends up just taking us in that direction every single time.

I'm glad you did. Allison Williams, thanks so much for coming on.

Thanks, Allison.

Thank you for having me. This was so fun. I'm starstruck.

We hope you'll come back for M3GAN 3.

I absolutely will, especially if I get to play myself, a human being.

You would come back.

It would be an honor.

Before we go, for even more Allison Williams, head over to YouTube at youtube.com/hardfork. She participated in one of our favorite segments, Group Chat Chat, about what's been blowing up her phone lately. Hard Fork is produced by Whitney Jones and Rachel Cohn. We're edited by Jen Poyntout. We're fact-checked by Caitlin Love. Today's show was engineered by Katie McMurrin. Original music by Marian Lozano, Diane Wong, Rowan Nemestó, and Dan Powell. Video production by Sawyer Roquet, Pat Gunther, and Chris Schott. You can watch this whole episode on YouTube at youtube.com/hardfork. Special thanks to Paul Schumann, Pui Wing Tam, Dalia Haddad, and Jeffrey Miranda. You can email us at hardfork@nytimes.com with how you would protect yourself from a killer robot.